Tests passing checkpoint

A bit more on the rust backend
Fix update script a bit
2025-06-03 14:57:27 +02:00 · 2025-05-29 17:09:17 +02:00 · 2025-05-23 16:45:37 +00:00 · 2025-05-23 10:55:32 +01:00 · 2025-05-22 17:26:58 +01:00 · 2025-05-22 14:42:49 +01:00
95 changed files with 22747 additions and 4627 deletions
--- a/.gitignore
+++ b/.gitignore
@ -17,3 +17,7 @@ dist/
 Cargo.lock
 src/python/qubed/_version.py
 *.ipynb
+cmake_build/
+tests/data/
+*.secret
+node_modules/
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@ -7,7 +7,8 @@ repos:
    -   id: trailing-whitespace
    -   id: end-of-file-fixer
    # -   id: check-yaml
-    -   id: check-added-large-files
+# -   id: check-added-large-files
+
 - repo: https://github.com/astral-sh/ruff-pre-commit
  rev: v0.9.7
  hooks:
--- a/Cargo.toml
+++ b/Cargo.toml
@ -8,7 +8,9 @@ repository = "https://github.com/ecmwf/qubed"
 # rsfdb = {git = "https://github.com/ecmwf/rsfdb", branch = "develop"}
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
-pyo3 = "0.23"
+pyo3 = "0.25"
+lasso = "0.7.3"
+itertools = "0.14.0"

 [package.metadata.maturin]
 version-from-git = true
--- a/README.md
+++ b/README.md
@ -1,4 +1,4 @@
-# Q<sup>3</sup> Quick Querying of Qubes
+# <p align="center"><img src="https://raw.githubusercontent.com/ecmwf/qubed/refs/heads/main/docs/_static/banner.svg" width="1000"></p> 
 [![Static Badge](https://github.com/ecmwf/codex/raw/refs/heads/main/Project%20Maturity/emerging_badge.svg)](https://github.com/ecmwf/codex/raw/refs/heads/main/Project%20Maturity#emerging)
 [![Docs](https://readthedocs.org/projects/qubed/badge/?version=latest)](https://qubed.readthedocs.io/en/latest/)
 [![PyPi](https://img.shields.io/pypi/v/qubed.svg)](https://pypi.org/project/qubed/)
@ -46,7 +46,7 @@ In addition to this core datastructure, this repostitory contains a collection o
 - 🌟 Implements our proposed [Datacube STAC Extension](./structured_stac.md).
 - 🛠️ Allows efficient traversal of ECMWF's datacubes.
 - Part of the implementation of this is [🌲 Tree Compressor](./tree_compresser), a **compressed tree representation** optimised for storing trees with many duplicated subtress.
- 🔗 **[Live Example](https://climate-catalogue.lumi.apps.dte.destination-earth.eu/api/stac?root=root&activity=story-nudging%2Cscenariomip&class=d1)**.
+- 🔗 **[Live Example](https://qubed.lumi.apps.dte.destination-earth.eu/api/v1/stac/climate-dt/?class=od%2Cd1&dataset=climate-dt)**.

 ---

@ -54,7 +54,7 @@ In addition to this core datastructure, this repostitory contains a collection o
 > **Web Frontend**

 - 👀 Displays data from the **STAC Server** in an intuitive user interface.
- 🌍 **[Try the Live Demo](https://climate-catalogue.lumi.apps.dte.destination-earth.eu/)**.
+- 🌍 **[Try the Live Demo](https://qubed.lumi.apps.dte.destination-earth.eu/)**.

 ---

--- a/chart/Chart.yaml
+++ b/chart/Chart.yaml
@ -1,5 +1,5 @@
 apiVersion: v2
-name: stac-server
+name: qubed
 description: A Helm chart for the STAC Server with frontend, STAC API and caching service.
 type: application
 version: 0.1.0
--- a/chart/templates/ingress.yaml
+++ b/chart/templates/ingress.yaml
@ -10,7 +10,7 @@ spec:
      http:
        paths:
        {{- if .Values.stacServer.enabled }}
-          - path: /api
+          - path: /
            pathType: Prefix
            backend:
              service:
@ -18,15 +18,6 @@ spec:
                port:
                  number: {{ .Values.stacServer.servicePort }}
        {{- end }}
-        {{- if .Values.webQueryBuilder.enabled }}
-          - path: /
-            pathType: Prefix
-            backend:
-              service:
-                name: web-query-builder
-                port:
-                  number: {{ .Values.webQueryBuilder.servicePort }}
-        {{- end }}
  tls:
    - hosts:
        - {{ .Values.ingress.hostname }}
--- a/chart/templates/redis-deployment.yaml
+++ b/chart/templates/redis-deployment.yaml
@ -1,28 +0,0 @@
-# templates/redis-deployment.yaml
-
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: redis
-spec:
-  selector:
-    matchLabels:
-      app: redis
-  template:
-    metadata:
-      labels:
-        app: redis
-    spec:
-      containers:
-        - name: redis
-          image: "redis:alpine"
-          command: ["redis-server", "--appendonly", "yes"]
-          ports:
-            - containerPort: {{ .Values.redis.servicePort }}
-          # volumeMounts:
-          #   - mountPath: /data
-          #     name: redis-data
-      # volumes:
-      #   - name: redis-data
-      #     persistentVolumeClaim:
-      #       claimName: redis-data
--- a/chart/templates/redis-pvc.yaml
+++ b/chart/templates/redis-pvc.yaml
@ -1,14 +0,0 @@
-# templates/redis-pvc.yaml
-
-{{- if .Values.redis.pvc.enabled }}
-apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: redis-data
-spec:
-  accessModes: {{ .Values.redis.pvc.accessModes }}
-  resources:
-    requests:
-      storage: {{ .Values.redis.pvc.size }}
-  storageClassName: {{ .Values.redis.pvc.storageClassName | quote }}
-{{- end }}
--- a/chart/templates/redis-service.yaml
+++ b/chart/templates/redis-service.yaml
@ -1,11 +0,0 @@
-apiVersion: v1
-kind: Service
-metadata:
-  name: redis
-spec:
-  selector:
-    app: redis
-  ports:
-    - protocol: TCP
-      port: {{ .Values.redis.servicePort }}
-      targetPort: {{ .Values.redis.servicePort }}
--- a/chart/templates/stac-server-configmap.yaml
+++ b/chart/templates/stac-server-configmap.yaml
@ -1,11 +0,0 @@
-# apiVersion: v1
-# kind: ConfigMap
-# metadata:
-#   name: stack-server
-# data:
-#   file1.txt: |-
-#     {{ .Files.Get "files/file1.txt" | nindent 2 }}
-#   file2.txt: |-
-#     {{ .Files.Get "files/file2.txt" | nindent 2 }}
-#   file3.txt: |-
-#     {{ .Files.Get "files/file3.txt" | nindent 2 }}
--- a/chart/templates/stac-server-deployment.yaml
+++ b/chart/templates/stac-server-deployment.yaml
@ -5,7 +5,7 @@ kind: Deployment
 metadata:
  name: stac-server
 spec:
-  replicas: 1  # Adjust as needed
+  replicas: {{ .Values.stacServer.replicas }}
  selector:
    matchLabels:
      app: stac-server
@ -14,20 +14,30 @@ spec:
      labels:
        app: stac-server
    spec:
-      initContainers:
-        - name: wait-for-redis
-          image: busybox
-          command:
-            [
-              'sh', '-c',
-              'until nc -z -v -w30 {{ .Values.stacServer.environment.REDIS_HOST }} {{ .Values.redis.service.port }}; do echo "Waiting for Redis..."; sleep 5; done;'
-            ]
      containers:
        - name: stac-server
          image: "{{ .Values.stacServer.image.repository }}:{{ .Values.stacServer.image.tag }}"
          imagePullPolicy: {{ .Values.stacServer.image.pullPolicy }}
          env:
-            - name: REDIS_HOST
-              value: "{{ .Values.stacServer.environment.REDIS_HOST }}"
+            - name: API_KEY
+              valueFrom:
+                secretKeyRef:
+                    name: api-key
+                    key: API_KEY
+            - name: API_URL
+              value: "https://{{ .Values.ingress.hostname }}/api/v1/"
          ports:
            - containerPort: {{ .Values.stacServer.servicePort }}
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: stac-server
+spec:
+  selector:
+    app: stac-server
+  ports:
+    - protocol: TCP
+      port: {{ .Values.stacServer.servicePort }}
+      targetPort: {{ .Values.stacServer.servicePort }}
+  type: ClusterIP
--- a/chart/templates/stac-server-service.yaml
+++ b/chart/templates/stac-server-service.yaml
@ -1,12 +0,0 @@
-apiVersion: v1
-kind: Service
-metadata:
-  name: stac-server
-spec:
-  selector:
-    app: stac-server
-  ports:
-    - protocol: TCP
-      port: {{ .Values.stacServer.servicePort }}
-      targetPort: {{ .Values.stacServer.servicePort }}
-  type: ClusterIP
--- a/chart/templates/web-query-builder-deployment.yaml
+++ b/chart/templates/web-query-builder-deployment.yaml
@ -1,37 +0,0 @@
---
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: web-query-builder
-spec:
-  replicas: {{ .Values.webQueryBuilder.replicas }}
-  selector:
-    matchLabels:
-      app: web-query-builder
-  template:
-    metadata:
-      labels:
-        app: web-query-builder
-    spec:
-      containers:
-        - name: web-query-builder
-          image: "{{ .Values.webQueryBuilder.image.repository }}:{{ .Values.webQueryBuilder.image.tag }}"
-          imagePullPolicy: {{ .Values.webQueryBuilder.image.pullPolicy }}
-          env:
-            - name: API_HOST
-              value: stac-server
-          ports:
-            - containerPort: {{ .Values.webQueryBuilder.servicePort }}
---
-apiVersion: v1
-kind: Service
-metadata:
-  name: web-query-builder
-spec:
-  selector:
-    app: web-query-builder
-  ports:
-    - protocol: TCP
-      port: {{ .Values.webQueryBuilder.servicePort }}
-      targetPort: {{ .Values.webQueryBuilder.servicePort }}
-  type: ClusterIP
--- a/chart/values.yaml
+++ b/chart/values.yaml
@ -1,33 +1,13 @@
-redis:
-  servicePort: 6379
-  pvc:
-    enabled: true
-    storageClassName: ""
-    accessModes:
-      - ReadWriteOnce
-    size: 1Gi
-  service:
-    port: 6379
-
 stacServer:
  enabled: true
+  replicas: 1
  image:
    repository: "eccr.ecmwf.int/qubed/stac_server"
    tag: "latest"
    pullPolicy: Always
  servicePort: 80
-  environment:
-    REDIS_HOST: "redis"
-
-webQueryBuilder:
-  enabled: true
-  image:
-    repository: "eccr.ecmwf.int/qubed/web_query_builder"
-    tag: "latest"
-    pullPolicy: Always
-  servicePort: 80

 ingress:
  enabled: True
  tlsSecretName: "lumi-wildcard-tls"
-  hostname: "climate-catalogue.lumi.apps.dte.destination-earth.eu"
+  hostname: "qubed.lumi.apps.dte.destination-earth.eu"
--- a/compose.yaml
+++ b/compose.yaml
@ -1,16 +1,5 @@

 services:
-  # redis server holds the catalog data blob
-  redis:
-    image: redis:alpine
-    container_name: redis
-    command: ["redis-server", "--appendonly", "yes"]
-    volumes:
-      - ./redis-data:/data
-    ports:
-      - "6379:6379"
-    restart: always
-
  # STAC Server
  stac_server:
    # image: stac-server:latest
@ -20,26 +9,24 @@ services:
      dockerfile: Dockerfile
      target: stac_server
    ports:
-      - "8124:8080"
+      - "8124:80"
    environment:
-      - REDIS_HOST=redis
      - CONFIG_DIR=/config
    volumes:
      - ./stac_server:/code/stac_server
-      - ./TreeTraverser:/code/TreeTraverser
    # restart: always

-  # web_query_builder:
-  #   # image: web_query_builder:latest
-  #   container_name: web_query_builder
-  #   build:
-  #     context: .
-  #     dockerfile: Dockerfile
-  #     target: web_query_builder
-  #   ports:
-  #     - "8125:8080"
-  #   environment:
-  #     - CONFIG_DIR=/config
-  #   volumes:
-  #     - ./web_query_builder:/code/web_query_builder
-    # restart: always
+  web_query_builder:
+    # image: web_query_builder:latest
+    container_name: web_query_builder
+    build:
+      context: .
+      dockerfile: Dockerfile
+      target: web_query_builder
+    ports:
+      - "8125:80"
+    environment:
+      - API_URL=http://127.0.0.1:8124/api/v1/stac/climate-dt
+    volumes:
+      - ./web_query_builder:/code/web_query_builder
+    restart: always
--- a/config/config-climate-dt.yaml
+++ b/config/config-climate-dt.yaml
@ -0,0 +1,6 @@
+---
+type: remote
+host: databridge-prod-catalogue3-ope.ewctest.link
+port: 10000
+engine: remote
+store: remote
--- a/config/config-extremes-dt.yaml
+++ b/config/config-extremes-dt.yaml
@ -0,0 +1,6 @@
+---
+type: remote
+host: databridge-prod-catalogue1-ope.ewctest.link
+port: 10000
+engine: remote
+store: remote
--- a/config/extremes-dt/language.yaml
+++ b/config/extremes-dt/language.yaml
--- a/config/climate-dt/language.yaml
+++ b/config/climate-dt/language.yaml
--- a/config/language/paramids.yaml
+++ b/config/language/paramids.yaml
--- a/config/local/language.yaml
+++ b/config/local/language.yaml
--- a/29
+++ b/29
@ -24,32 +24,13 @@ FROM base AS stac_server
 COPY stac_server/requirements.txt /code/requirements.txt
 RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt

-# Todo: don't embed this here, mount them at runtime
-# ENV CONFIG_DIR=/config/
-# COPY config/destinE/config.yaml /config/config.yaml
-# COPY config/destinE/schema /config/schema
-# COPY config/destinE/language.yaml /config/language.yaml
+COPY ./src /code/qubed/src
+COPY ./pyproject.toml /code/qubed/
+COPY ./Cargo.toml /code/qubed/
+COPY ./README.md /code/qubed/

-COPY ./tree_compresser /code/tree_compresser
-
-# Clone the rsfdb and rsfindlibs repos manually because they're private
-
-# RUN --mount=type=ssh git clone ssh://git@github.com/ecmwf/rsfdb.git
-# RUN --mount=type=ssh git clone ssh://git@github.com/ecmwf/rsfindlibs.git
-COPY stac_server/deps/rsfdb /code/rsfdb
-COPY stac_server/deps/rsfindlibs /code/rsfindlibs
-
-RUN pip install --no-cache-dir -e /code/tree_compresser
+RUN pip install --no-cache-dir -e /code/qubed
 COPY ./stac_server /code/stac_server

 WORKDIR /code/stac_server
 CMD ["fastapi", "dev", "main.py", "--proxy-headers", "--port", "80", "--host", "0.0.0.0"]
-
-FROM base AS web_query_builder
-
-COPY web_query_builder/requirements.txt /code/requirements.txt
-RUN pip install --no-cache-dir --upgrade -r /code/requirements.txt
-
-COPY web_query_builder /code/web_query_builder
-WORKDIR /code/web_query_builder
-CMD ["flask", "run", "--host", "0.0.0.0", "--port", "80"]
--- a/docs/_static/banner.svg
+++ b/docs/_static/banner.svg
@ -0,0 +1,159 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<!-- Created with Inkscape (http://www.inkscape.org/) -->
+
+<svg
+   width="1000"
+   height="200"
+   viewBox="0 0 264.58333 52.916666"
+   version="1.1"
+   id="svg5"
+   xml:space="preserve"
+   inkscape:version="1.2.2 (b0a84865, 2022-12-01)"
+   sodipodi:docname="banner.svg"
+   xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
+   xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
+   xmlns="http://www.w3.org/2000/svg"
+   xmlns:svg="http://www.w3.org/2000/svg"><sodipodi:namedview
+     id="namedview7"
+     pagecolor="#ffffff"
+     bordercolor="#000000"
+     borderopacity="0.25"
+     inkscape:showpageshadow="2"
+     inkscape:pageopacity="0.0"
+     inkscape:pagecheckerboard="0"
+     inkscape:deskcolor="#d1d1d1"
+     inkscape:document-units="mm"
+     showgrid="false"
+     inkscape:zoom="1.4221154"
+     inkscape:cx="509.80392"
+     inkscape:cy="23.908046"
+     inkscape:window-width="2665"
+     inkscape:window-height="1000"
+     inkscape:window-x="96"
+     inkscape:window-y="35"
+     inkscape:window-maximized="0"
+     inkscape:current-layer="g330" /><defs
+     id="defs2"><rect
+       x="641.41612"
+       y="32.816639"
+       width="73.588826"
+       height="29.833308"
+       id="rect2775" /><rect
+       x="500.20513"
+       y="263.52755"
+       width="244.63313"
+       height="143.19988"
+       id="rect2749" /><rect
+       x="467.38849"
+       y="331.14972"
+       width="258.55534"
+       height="132.261"
+       id="rect2743" /><rect
+       x="80.859469"
+       y="61.833711"
+       width="299.65568"
+       height="114.15454"
+       id="rect242" /><rect
+       x="61.833711"
+       y="66.590151"
+       width="313.925"
+       height="114.15454"
+       id="rect236" /></defs><g
+     inkscape:label="Layer 1"
+     inkscape:groupmode="layer"
+     id="layer1"><text
+       xml:space="preserve"
+       transform="scale(0.26458333)"
+       id="text234"
+       style="font-weight:500;font-size:20px;line-height:1.2;font-family:Futura;-inkscape-font-specification:'Futura, Medium';white-space:pre;shape-inside:url(#rect236);display:inline;fill:#000000;stroke-width:0.926667;stroke-miterlimit:4.9;stroke-dasharray:2.78, 0.926667;stroke-opacity:0.40146" /><text
+       xml:space="preserve"
+       transform="scale(0.26458333)"
+       id="text240"
+       style="font-weight:500;font-size:20px;line-height:1.2;font-family:Futura;-inkscape-font-specification:'Futura, Medium';white-space:pre;shape-inside:url(#rect242);display:inline;fill:#000000;stroke-width:0.926667;stroke-miterlimit:4.9;stroke-dasharray:2.78, 0.926667;stroke-opacity:0.40146" /><text
+       xml:space="preserve"
+       transform="scale(0.26458333)"
+       id="text2741"
+       style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:20px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;white-space:pre;shape-inside:url(#rect2743);display:inline;fill:#000000;stroke-width:0.926667;stroke-miterlimit:4.9;stroke-dasharray:2.78, 0.926667;stroke-opacity:0.40146" /><text
+       xml:space="preserve"
+       transform="scale(0.26458333)"
+       id="text2747"
+       style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:20px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;white-space:pre;shape-inside:url(#rect2749);display:inline;fill:#000000;stroke-width:0.926667;stroke-miterlimit:4.9;stroke-dasharray:2.78, 0.926667;stroke-opacity:0.40146" /><text
+       xml:space="preserve"
+       transform="scale(0.26458333)"
+       id="text2773"
+       style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:26.6667px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;white-space:pre;shape-inside:url(#rect2775);display:inline;fill:#000000;stroke-width:0.926667;stroke-miterlimit:4.9;stroke-dasharray:2.78, 0.926667;stroke-opacity:0.40146" /><g
+       id="g349"
+       transform="translate(-5.8208336)"><text
+         xml:space="preserve"
+         style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;fill:#000000;stroke-width:0.245181;stroke-miterlimit:4.9;stroke-dasharray:0.735542, 0.245181;stroke-opacity:0.40146"
+         x="114.48351"
+         y="14.228302"
+         id="text2763"><tspan
+           sodipodi:role="line"
+           id="tspan2761"
+           style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+           x="114.48351"
+           y="14.228302">root</tspan><tspan
+           sodipodi:role="line"
+           style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+           x="114.48351"
+           y="21.701376"
+           id="tspan2765">├── class=cd, stream=lwda/oai, param=1/2/3</tspan><tspan
+           sodipodi:role="line"
+           style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+           x="114.48351"
+           y="29.17445"
+           id="tspan2767">├── class=od, expver=1/2, param=1/2</tspan><tspan
+           sodipodi:role="line"
+           style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+           x="114.48351"
+           y="36.647522"
+           id="tspan2771">├── class=rd, param=1/2/3</tspan><tspan
+           sodipodi:role="line"
+           style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+           x="114.48351"
+           y="44.120598"
+           id="tspan2769">└── ...</tspan></text><g
+         id="g330"><text
+           xml:space="preserve"
+           style="font-weight:500;font-size:14.1111px;line-height:0;font-family:Futura;-inkscape-font-specification:'Futura, Medium';white-space:pre;inline-size:112.115;display:inline;fill:#000000;stroke-width:0.245181;stroke-miterlimit:4.9;stroke-dasharray:0.735542, 0.245181;stroke-opacity:0.40146"
+           x="5.439929"
+           y="17.022402"
+           id="text248"
+           transform="translate(0,-1.5875)"><tspan
+             x="5.439929"
+             y="17.022402"
+             id="tspan532"><tspan
+               style="font-size:12.3472px;line-height:1.2"
+               id="tspan530">Qube</tspan></tspan></text><text
+           xml:space="preserve"
+           style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:5.29167px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;white-space:pre;inline-size:87.6248;display:inline;fill:#000000;stroke-width:0.245181;stroke-miterlimit:4.9;stroke-dasharray:0.735542, 0.245181;stroke-opacity:0.40146"
+           x="4.209815"
+           y="40.519432"
+           id="text2755"
+           transform="translate(1.744648,-4.9844494)"><tspan
+             x="4.209815"
+             y="40.519432"
+             id="tspan534">1. A data structure for efficiently </tspan><tspan
+             x="4.209815"
+             y="46.604852"
+             id="tspan536">representing and querying complex </tspan><tspan
+             x="4.209815"
+             y="52.690271"
+             id="tspan538">tree-like datacubes.</tspan></text><text
+           xml:space="preserve"
+           style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:5.29167px;line-height:1.15;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;fill:#000000;stroke-width:0.245181;stroke-miterlimit:4.9;stroke-dasharray:0.735542, 0.245181;stroke-opacity:0.40146"
+           x="5.4673572"
+           y="26.586193"
+           id="text2759"><tspan
+             sodipodi:role="line"
+             id="tspan2757"
+             style="font-style:normal;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal;stroke-width:0.245181"
+             x="5.4673572"
+             y="26.586193">[kjuːb] <tspan
+   style="font-style:italic;font-variant:normal;font-weight:500;font-stretch:normal;font-size:6.35px;font-family:Futura;-inkscape-font-specification:'Futura, Medium Italic';font-variant-ligatures:normal;font-variant-caps:normal;font-variant-numeric:normal;font-variant-east-asian:normal"
+   id="tspan495">noun</tspan></tspan></text><path
+           style="fill:#000000;stroke:#000000;stroke-width:0.445;stroke-miterlimit:4.9;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1"
+           d="M 6.0516036,18.417924 H 92.221177"
+           id="path2833"
+           sodipodi:nodetypes="cc" /></g></g></g></svg>
--- a/docs/api.md
+++ b/docs/api.md
@ -1,13 +0,0 @@
-# API
-
-## Set Operations
-
-```{code-cell} python3
-from qubed import Qube
-
-A = Qube.from_dict({
-    "a=1": {"b": {1, 2, 3}, "c": {1}},
-    "a=2": {"b": {1, 2, 3}, "c": {1}},
-})
-A
-```
--- a/docs/cmd.md
+++ b/docs/cmd.md
@ -2,21 +2,18 @@

 ```bash
 fdb list class=rd,expver=0001,... | qubed --from=fdblist --to=text
+
+fdb list --minimum-keys=class class=d1,dataset=climate-dt --config prod_remoteFDB.yaml  | qubed convert --from=fdb --to=text
+
 ```

 `--from` options include:
-* `fdblist`
-* `json`
-* `protobuf`
-* `marslist`
-* `constraints`
+* `fdb`

 `--to` options include:
 * `text`
 * `html`
 * `json`
-* `datacubes`
-* `constraints`

 use `--input` and `--output` to specify input and output files respectively.

@ -26,4 +23,22 @@ There's some handy test data in the `tests/data` directory. For example:
 gzip -dc tests/data/fdb_list_compact.gz| qubed convert --from=fdb --to=text --output=qube.txt
 gzip -dc tests/data/fdb_list_porcelain.gz| qubed convert --from=fdb --to=json --output=qube.json
 gzip -dc tests/data/fdb_list_compact.gz | qubed convert --from=fdb --to=html --output=qube.html
+
+// Operational data stream=oper/wave/enfo/waef
+fdb list class=od,expver=0001,date=0,stream=oper --compact >> operational_compact.txt
+operational_compact.txt | qubed convert --from=fdb --to=text --output=operational.txt
 ```
+
+
+
+## Todo
+
+--from for
+* `protobuf`
+* `marslist`
+* `constraints`
+
+--to for
+* `json`
+* `datacubes`
+* `constraints`
--- a/docs/conf.py
+++ b/docs/conf.py
@ -33,3 +33,7 @@ source_suffix = {

 html_theme = "sphinx_rtd_theme"
 html_static_path = ["_static"]
+
+myst_enable_extensions = [
+    "attrs_inline",
+]
--- a/docs/development.md
+++ b/docs/development.md
@ -1,6 +1,19 @@
 # Development

+To install the latest stable release from PyPI (recommended):
+
+```bash
+pip install qubed
+```
+
+To install the latest version from github (requires rust):
+
+```bash
+pip install qubed@git+https://github.com/ecmwf/qubed.git@main
+```
+
 To build the develop branch from source install a rust toolchain and pip install maturin then run:
+
 ```
 git clone -b develop git@github.com:ecmwf/qubed.git
 cd qubed
--- a/docs/fiab.md
+++ b/docs/fiab.md
@ -98,6 +98,40 @@ all_models.select({
 })
 ```

+## Using WildCards
+
+```{code-cell} python3
+daily_surface_means = Qube.from_datacube({
+    "model": "*",
+    "frequency": "*",
+    "levtype": "sfc",
+    "param": "*",
+})
+all_models & daily_surface_means
+```
+
+```{code-cell} python3
+
+daily_level_means = Qube.from_datacube({
+    "model": "*",
+    "frequency": "*",
+    "levtype": "pl",
+    "param": "*",
+    "level": "*"
+})
+all_models & daily_level_means
+```
+
+```{code-cell} python3
+daily_level_mean_products = all_models & daily_surface_means
+for i, identifier in enumerate(daily_level_mean_products.leaves()):
+    print(identifier)
+    if i > 10:
+        print("...")
+        break
+
+```
+
 <!-- ## Choosing the fewest models needed to cover the requested parameter set -->

 <!-- ```{code-cell} python3 -->
--- a/docs/index.md
+++ b/docs/index.md
@ -11,46 +11,43 @@ jupytext:

 ```{toctree}
 :maxdepth: 1
-background.md
 quickstart.md
-api.md
 development.md
+background.md
 algorithms.md
 fiab.md
+cmd.md
 ```

-Qubed provides a datastructure called a Qube which represents sets of data identified by multiple key value pairs as a tree of datacubes. To understand what that means go to [Background](background.md), to just start using the library skip straight to the [Quickstart](quickstart.md).
-
-Here's a real world dataset from the [Climate DT](https://destine.ecmwf.int/climate-change-adaptation-digital-twin-climate-dt/):
-
-```{code-cell} python3
-import requests
-from qubed import Qube
-climate_dt = Qube.from_json(requests.get("https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/climate_dt.json").json())
-climate_dt.html(depth=1)
+Qubed provides a datastructure primitive for working with trees of DataCubes. If a normal tree looks like this:
+```
+root
+├── class=od
+│   ├── expver=0001
+│   │   ├── param=1
+│   │   └── param=2
+│   └── expver=0002
+│       ├── param=1
+│       └── param=2
+└── class=rd
+    ├── expver=0001
+    │   ├── param=1
+    │   ├── param=2
+    │   └── param=3
+    └── expver=0002
+        ├── param=1
+        └── param=2
 ```

-Click the arrows to expand and drill down deeper into the data. Any particular dataset is uniquely identified by a set of key value pairs:
-
-```{code-cell} python3
-import json
-for i, identifier in enumerate(climate_dt.leaves()):
-    print(identifier)
-    break
+A compressed view of the same set would be:
+```
+root
+├── class=od, expver=0001/0002, param=1/2
+└── class=rd
+    ├── expver=0001, param=1/2/3
+    └── expver=0002, param=1/2
 ```

-Here's an idea of the set of values each key can take:
-```{code-cell} python3
-axes = climate_dt.axes()
-for key, values in axes.items():
-    print(f"{key} : {list(sorted(values))[:10]}")
-```
+Qubed provides a datastructure that represents this compressed cube we call a Qube. It defines all the algorithms you would expect such as intersection/union/difference, compression, search, transformation and filtering.

-This dataset isn't dense, you can't choose any combination of the above key values pairs, but it does contain many dense datacubes. Hence it makes sense to store and process the set as a tree of dense datacubes, what we call a Qube. For a sense of scale, this dataset contains about 200 million distinct datasets but only contains a few thousand unique nodes.
-
-```{code-cell} python3
-print(f"""
-Distinct datasets: {climate_dt.n_leaves},
-Number of nodes in the tree: {climate_dt.n_nodes}
-""")
-```
+To get a little more background on the motivation and structure of a Qube go to [Background](background.md), for a more hands on intro, go to [Quickstart](quickstart.md).
--- a/docs/quickstart.md
+++ b/docs/quickstart.md
@ -8,23 +8,54 @@ jupytext:
 ---
 # Quickstart

-## Installation
-To install the latest stable release from PyPI (recommended):
-```bash
-pip install qubed
-```
-Or to build and install the latest version from github (requires cargo):
-```bash
-pip install qubed@git+https://github.com/ecmwf/qubed.git@main
+First install qubed with `pip install qubed`. Now, let's dive in with a real world dataset from the [Climate DT](https://destine.ecmwf.int/climate-change-adaptation-digital-twin-climate-dt/). We'll pull a prebuilt qube from github and render it in it's default HTML representation.
+
+```{code-cell} python3
+import requests
+from qubed import Qube
+climate_dt = Qube.from_json(requests.get("https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/climate_dt.json").json())
+climate_dt.html(depth=1)
 ```

-## Usage
-Make an uncompressed qube:
+Click the arrows to expand and drill down deeper into the data.
+
+```{note}
+There is currently a simple Qube web browser hosted [here](https://qubed.lumi.apps.dte.destination-earth.eu/). Browse that and copy the 'Example Qube Code' to download a Qube representing the selection at that point. You'll get something like `Qube.from_json(requests.get("https://qubed.lumi.apps.dte.destination-earth.eu/api/v1/select/climate-dt/?").json())`{l=python}
+```
+
+Fundamentally a Qube represents a set identifiers which are a set of key value pairs, here's the one leaf in the Climate DT dataset:
+
+```{code-cell} python3
+next(climate_dt.leaves())
+```
+
+We can look at the set of values each key can take:
+```{code-cell} python3
+axes = climate_dt.axes()
+for key, values in axes.items():
+    print(f"{key} : {list(sorted(values))[:10]}")
+```
+
+This dataset isn't dense, you can't choose any combination of the above key values pairs, but it does contain many dense datacubes. Hence it makes sense to store and process the set as a tree of dense datacubes, which is what a Qube. For a sense of scale, this dataset contains about 8 million distinct datasets but only contains a few hundred unique nodes.
+
+```{code-cell} python3
+import objsize
+print(f"""
+Distinct datasets: {climate_dt.n_leaves}
+Number of nodes in the tree: {climate_dt.n_nodes}
+Number of dense datacubes within this qube: {len(list(climate_dt.datacubes()))}
+In memory size according to objsize: {objsize.get_deep_size(climate_dt) / 2**20:.0f} MB
+""")
+```
+
+## Building your own Qubes
+
+You can do it from nested dictionaries with keys in the form "{key=value}":

 ```{code-cell} python3
 from qubed import Qube

-q = Qube.from_dict({
+q1 = Qube.from_dict({
    "class=od" : {
        "expver=0001": {"param=1":{}, "param=2":{}},
        "expver=0002": {"param=1":{}, "param=2":{}},
@ -34,20 +65,63 @@ q = Qube.from_dict({
        "expver=0002": {"param=1":{}, "param=2":{}},
    },
 })
-print(f"{q.n_leaves = }, {q.n_nodes = }")
-q
+print(f"{q1.n_leaves = }, {q1.n_nodes = }")
+q1
 ```

-Compress it:
+If someone sends you a printed qube you can convert that back to a Qube too:

 ```{code-cell} python3
-cq = q.compress()
-assert cq.n_leaves == q.n_leaves
+q2 = Qube.from_tree("""
+    root, frequency=6:00:00
+    ├── levtype=pl, param=t, levelist=850, threshold=-2/-4/-8/2/4/8
+    └── levtype=sfc
+        ├── param=10u/10v, threshold=10/15
+        ├── param=2t, threshold=273.15
+        └── param=tp, threshold=0.1/1/10/100/20/25/5/50
+""")
+q2
+```
+We would not recommend trying to write this representation by hand though.
+
+Finally, quite a flexible approach is to take the union of a series of dense datacubes:
+
+```{code-cell} python3
+q3 = Qube.from_datacube(
+    dict(
+        param="10u/10v/2d/2t/cp/msl/skt/sp/tcw/tp".split("/"),
+        threshold="*",
+        levtype="sfc",
+        frequency="6:00:00",
+    )
+) | Qube.from_datacube(
+    dict(
+        param="q/t/u/v/w/z".split("/"),
+        threshold="*",
+        levtype="pl",
+        level="50/100/150/200/250/300/400/500/600/700/850".split("/"),
+        frequency="6:00:00",
+    )
+)
+q3
+```
+
+## Operations on Qubes
+
+Going back to that first qube:
+```{code-cell} python3
+q1
+```
+
+We can compress it:
+
+```{code-cell} python3
+cq = q1.compress()
+assert cq.n_leaves == q1.n_leaves
 print(f"{cq.n_leaves = }, {cq.n_nodes = }")
 cq
 ```

-
 With the HTML representation you can click on the leaves to expand them. You can copy a path representation of a node to the clipboard by alt/option/⌥ clicking on it. You can then extract that node in code using `[]`:

 ```{code-cell} python3
@ -73,30 +147,7 @@ dq = Qube.from_datacube({
 ```


-
-### Tree Construction
-
-One of the quickest ways to construct non-trivial trees is to use the `Qube.from_datacube` method to construct dense trees and then use the set operations to combine or intersect them:
-
-
-```{code-cell} python3
-q = Qube.from_datacube({
-    "class": "d1",
-    "dataset": ["climate-dt", "another-value"],
-    'generation': ['1', "2", "3"],
-})
-
-r  = Qube.from_datacube({
-    "class": "d1",
-    "dataset": ["weather-dt", "climate-dt"],
-    'generation': ['1', "2", "3", "4"],
-})
-
-q | r
-```
-
-
-### Iteration / Flattening
+## Iteration

 Iterate over the leaves:

@ -108,26 +159,13 @@ for i, identifier in enumerate(cq.leaves()):
        break
 ```

-Iterate over the datacubes:
+Or if you can it's more efficient to iterate over the datacubes:

 ```{code-cell} python3
-cq.datacubes()
-```
-
-### A Real World Example
-
-Load a larger example qube:
-
-```{code-cell} python3
-import requests
-qube_json = requests.get("https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/climate_dt.json").json()
-climate_dt = Qube.from_json(qube_json)
-
-# Using the html or print methods is optional but lets you specify things like the depth of the tree to display.
-print(f"{climate_dt.n_leaves = }, {climate_dt.n_nodes = }")
-climate_dt.html(depth=1) # Limit how much is open initially, click leave to see more.
+list(cq.datacubes())
 ```

+## Selection
 Select a subset of the tree:

 ```{code-cell} python3
@ -151,7 +189,7 @@ for key, values in axes.items():
 ```


-### Set Operations
+## Set Operations

 The union/intersection/difference of two dense datacubes is not itself dense.

@ -186,7 +224,7 @@ Symmetric Difference:
 (A ^ B).print();
 ```

-### Transformations
+## Transformations

 `q.transform` takes a python function from one node to one or more nodes and uses this to build a new tree. This can be used for simple operations on the key or values but also to split or remove nodes. Note that you can't use it to merge nodes beause it's only allowed to see one node at a time.

@ -194,3 +232,42 @@ Symmetric Difference:
 def capitalize(node): return node.replace(key = node.key.capitalize())
 climate_dt.transform(capitalize).html(depth=1)
 ```
+
+## Save to disk
+
+There is currently a very simple JSON serialisation format. More compact binary serialisations are planned.
+```{code-cell} python3
+json = climate_dt.to_json()
+Qube.from_json(json) == climate_dt
+```
+
+## Advanced Selection
+
+There is currently partial support for different datatypes in addition to strings. Here we can convert datatypes by key to ints and timedeltas and then use functions as filters in select.
+
+```{code-cell} python3
+from datetime import timedelta, datetime
+def to_timedelta(t):
+    dt = datetime.strptime(t, "%H:%M:%S")
+    return timedelta(hours=dt.hour, minutes=dt.minute, seconds=dt.second)
+
+q = Qube.from_tree("""
+root, frequency=6:00:00
+├── levtype=pl, levelist=850, threshold=-2/-4/-8/2/4/8
+└── levtype=sfc
+    ├── param=10u/10v, threshold=10/15
+    ├── param=2t, threshold=273.15
+    └── param=tp, threshold=0.1/1/10/100/20/25/5/50
+""").convert_dtypes({
+    "threshold": float,
+    "levelist": int,
+    "frequency": to_timedelta,
+})
+
+r = q.select({
+        "threshold": lambda t: t > 5,
+        "frequency": lambda dt: dt > timedelta(hours = 2),
+})
+
+r
+```
--- a/docs/requirements.txt
+++ b/docs/requirements.txt
@ -1,2 +1,3 @@
 numpy
 scipy
+objsize
--- a/fiab/extract.py
+++ b/fiab/extract.py
@ -1,4 +1,3 @@
-
 import json
 from collections import defaultdict

--- a/notebooks/DataCubeTree.py
+++ b/notebooks/DataCubeTree.py
@ -1,472 +0,0 @@
-import dataclasses
-from abc import ABC, abstractmethod
-from dataclasses import dataclass
-from datetime import date, datetime, timedelta
-from typing import Any, Callable, Iterable, Literal
-
-
-@dataclass(frozen=True)
-class HTML:
-    html: str
-
-    def _repr_html_(self):
-        return self.html
-
-
-@dataclass(frozen=True)
-class Values(ABC):
-    @abstractmethod
-    def summary(self) -> str:
-        pass
-
-    @abstractmethod
-    def __len__(self) -> int:
-        pass
-
-    @abstractmethod
-    def __contains__(self, value: Any) -> bool:
-        pass
-
-    @abstractmethod
-    def from_strings(self, values: list[str]) -> list["Values"]:
-        pass
-
-
-@dataclass(frozen=True)
-class Enum(Values):
-    """
-    The simplest kind of key value is just a list of strings.
-    summary -> string1/string2/string....
-    """
-
-    values: list[Any]
-
-    def __len__(self) -> int:
-        return len(self.values)
-
-    def summary(self) -> str:
-        return "/".join(sorted(self.values))
-
-    def __contains__(self, value: Any) -> bool:
-        return value in self.values
-
-    def from_strings(self, values: list[str]) -> list["Values"]:
-        return [Enum(values)]
-
-
-@dataclass(frozen=True)
-class Range(Values, ABC):
-    dtype: str = dataclasses.field(kw_only=True)
-
-
-@dataclass(frozen=True)
-class DateRange(Range):
-    start: date
-    end: date
-    step: timedelta
-    dtype: Literal["date"] = dataclasses.field(kw_only=True, default="date")
-
-    @classmethod
-    def from_strings(self, values: list[str]) -> list["DateRange"]:
-        dates = sorted([datetime.strptime(v, "%Y%m%d") for v in values])
-        if len(dates) < 2:
-            return [DateRange(start=dates[0], end=dates[0], step=timedelta(days=0))]
-
-        ranges = []
-        current_range, dates = (
-            [
-                dates[0],
-            ],
-            dates[1:],
-        )
-        while len(dates) > 1:
-            if dates[0] - current_range[-1] == timedelta(days=1):
-                current_range.append(dates.pop(0))
-
-            elif len(current_range) == 1:
-                ranges.append(
-                    DateRange(
-                        start=current_range[0],
-                        end=current_range[0],
-                        step=timedelta(days=0),
-                    )
-                )
-                current_range = [
-                    dates.pop(0),
-                ]
-
-            else:
-                ranges.append(
-                    DateRange(
-                        start=current_range[0],
-                        end=current_range[-1],
-                        step=timedelta(days=1),
-                    )
-                )
-                current_range = [
-                    dates.pop(0),
-                ]
-        return ranges
-
-    def __contains__(self, value: Any) -> bool:
-        v = datetime.strptime(value, "%Y%m%d").date()
-        return self.start <= v <= self.end and (v - self.start) % self.step == 0
-
-    def __len__(self) -> int:
-        return (self.end - self.start) // self.step
-
-    def summary(self) -> str:
-        def fmt(d):
-            return d.strftime("%Y%m%d")
-
-        if self.step == timedelta(days=0):
-            return f"{fmt(self.start)}"
-        if self.step == timedelta(days=1):
-            return f"{fmt(self.start)}/to/{fmt(self.end)}"
-
-        return (
-            f"{fmt(self.start)}/to/{fmt(self.end)}/by/{self.step // timedelta(days=1)}"
-        )
-
-
-@dataclass(frozen=True)
-class TimeRange(Range):
-    start: int
-    end: int
-    step: int
-    dtype: Literal["time"] = dataclasses.field(kw_only=True, default="time")
-
-    @classmethod
-    def from_strings(self, values: list[str]) -> list["TimeRange"]:
-        if len(values) == 0:
-            return []
-
-        times = sorted([int(v) for v in values])
-        if len(times) < 2:
-            return [TimeRange(start=times[0], end=times[0], step=100)]
-
-        ranges = []
-        current_range, times = (
-            [
-                times[0],
-            ],
-            times[1:],
-        )
-        while len(times) > 1:
-            if times[0] - current_range[-1] == 1:
-                current_range.append(times.pop(0))
-
-            elif len(current_range) == 1:
-                ranges.append(
-                    TimeRange(start=current_range[0], end=current_range[0], step=0)
-                )
-                current_range = [
-                    times.pop(0),
-                ]
-
-            else:
-                ranges.append(
-                    TimeRange(start=current_range[0], end=current_range[-1], step=1)
-                )
-                current_range = [
-                    times.pop(0),
-                ]
-        return ranges
-
-    def __len__(self) -> int:
-        return (self.end - self.start) // self.step
-
-    def summary(self) -> str:
-        def fmt(d):
-            return f"{d:04d}"
-
-        if self.step == 0:
-            return f"{fmt(self.start)}"
-        return f"{fmt(self.start)}/to/{fmt(self.end)}/by/{self.step}"
-
-    def __contains__(self, value: Any) -> bool:
-        v = int(value)
-        return self.start <= v <= self.end and (v - self.start) % self.step == 0
-
-
-@dataclass(frozen=True)
-class IntRange(Range):
-    dtype: Literal["int"]
-    start: int
-    end: int
-    step: int
-    dtype: Literal["int"] = dataclasses.field(kw_only=True, default="int")
-
-    def __len__(self) -> int:
-        return (self.end - self.start) // self.step
-
-    def summary(self) -> str:
-        def fmt(d):
-            return d.strftime("%Y%m%d")
-
-        return f"{fmt(self.start)}/to/{fmt(self.end)}/by/{self.step}"
-
-    def __contains__(self, value: Any) -> bool:
-        v = int(value)
-        return self.start <= v <= self.end and (v - self.start) % self.step == 0
-
-
-def values_from_json(obj) -> Values:
-    if isinstance(obj, list):
-        return Enum(obj)
-
-    match obj["dtype"]:
-        case "date":
-            return DateRange(**obj)
-        case "time":
-            return TimeRange(**obj)
-        case "int":
-            return IntRange(**obj)
-        case _:
-            raise ValueError(f"Unknown dtype {obj['dtype']}")
-
-
-@dataclass(frozen=True)
-class Node:
-    key: str
-    values: Values  # Must support len()
-    metadata: dict[str, str]  # Applies to all children
-    payload: list[Any]  # List of size product(len(n.values) for n in  ancestors(self))
-    children: list["Node"]
-
-
-def summarize_node(node: Node) -> tuple[str, Node]:
-    """
-    Extracts a summarized representation of the node while collapsing single-child paths.
-    Returns the summary string and the last node in the chain that has multiple children.
-    """
-    summary = []
-
-    while True:
-        values_summary = node.values.summary()
-        if len(values_summary) > 50:
-            values_summary = values_summary[:50] + "..."
-        summary.append(f"{node.key}={values_summary}")
-
-        # Move down if there's exactly one child, otherwise stop
-        if len(node.children) != 1:
-            break
-        node = node.children[0]
-
-    return ", ".join(summary), node
-
-
-def node_tree_to_string(node: Node, prefix: str = "", depth=None) -> Iterable[str]:
-    summary, node = summarize_node(node)
-
-    if depth is not None and depth <= 0:
-        yield summary + " - ...\n"
-        return
-    # Special case for nodes with only a single child, this makes the printed representation more compact
-    elif len(node.children) == 1:
-        yield summary + ", "
-        yield from node_tree_to_string(node.children[0], prefix, depth=depth)
-        return
-    else:
-        yield summary + "\n"
-
-    for index, child in enumerate(node.children):
-        connector = "└── " if index == len(node.children) - 1 else "├── "
-        yield prefix + connector
-        extension = "    " if index == len(node.children) - 1 else "│   "
-        yield from node_tree_to_string(
-            child, prefix + extension, depth=depth - 1 if depth is not None else None
-        )
-
-
-def node_tree_to_html(
-    node: Node, prefix: str = "", depth=1, connector=""
-) -> Iterable[str]:
-    summary, node = summarize_node(node)
-
-    if len(node.children) == 0:
-        yield f'<span class="leaf">{connector}{summary}</span>'
-        return
-    else:
-        open = "open" if depth > 0 else ""
-        yield f"<details {open}><summary>{connector}{summary}</summary>"
-
-    for index, child in enumerate(node.children):
-        connector = "└── " if index == len(node.children) - 1 else "├── "
-        extension = "    " if index == len(node.children) - 1 else "│   "
-        yield from node_tree_to_html(
-            child, prefix + extension, depth=depth - 1, connector=prefix + connector
-        )
-    yield "</details>"
-
-
-@dataclass(frozen=True)
-class CompressedTree:
-    root: Node
-
-    @classmethod
-    def from_json(cls, json: dict) -> "CompressedTree":
-        def from_json(json: dict) -> Node:
-            return Node(
-                key=json["key"],
-                values=values_from_json(json["values"]),
-                metadata=json["metadata"] if "metadata" in json else {},
-                payload=json["payload"] if "payload" in json else [],
-                children=[from_json(c) for c in json["children"]],
-            )
-
-        return CompressedTree(root=from_json(json))
-
-    def __str__(self):
-        return "".join(node_tree_to_string(node=self.root))
-
-    def html(self, depth=2) -> HTML:
-        return HTML(self._repr_html_(depth=depth))
-
-    def _repr_html_(self, depth=2):
-        css = """
-        <style>
-        .qubed-tree-view {
-            font-family: monospace;
-            white-space: pre;
-        }
-        .qubed-tree-view details {
-            # display: inline;
-            margin-left: 0;
-        }
-        .qubed-tree-view summary {
-            list-style: none;
-            cursor: pointer;
-            text-overflow: ellipsis;
-            overflow: hidden;
-            text-wrap: nowrap;
-            display: block;
-        }
-
-        .qubed-tree-view .leaf {
-            text-overflow: ellipsis;
-            overflow: hidden;
-            text-wrap: nowrap;
-            display: block;
-        }
-
-        .qubed-tree-view summary:hover,span.leaf:hover {
-            background-color: #f0f0f0;
-        }
-        .qubed-tree-view details > summary::after {
-            content: ' ';
-        }
-        .qubed-tree-view details:not([open]) > summary::after {
-            content: " ▼";
-        }
-        </style>
-
-        """
-        nodes = "".join(
-            cc
-            for c in self.root.children
-            for cc in node_tree_to_html(node=c, depth=depth)
-        )
-        return f"{css}<pre class='qubed-tree-view'>{nodes}</pre>"
-
-    def print(self, depth=None):
-        print(
-            "".join(
-                cc
-                for c in self.root.children
-                for cc in node_tree_to_string(node=c, depth=depth)
-            )
-        )
-
-    def transform(self, func: Callable[[Node], Node]) -> "CompressedTree":
-        "Call a function on every node of the tree, any changes to the children of a node will be ignored."
-
-        def transform(node: Node) -> Node:
-            new_node = func(node)
-            return dataclasses.replace(
-                new_node, children=[transform(c) for c in node.children]
-            )
-
-        return CompressedTree(root=transform(self.root))
-
-    def guess_datatypes(self) -> "CompressedTree":
-        def guess_datatypes(node: Node) -> list[Node]:
-            # Try to convert enum values into more structured types
-            children = [cc for c in node.children for cc in guess_datatypes(c)]
-
-            if isinstance(node.values, Enum):
-                match node.key:
-                    case "time":
-                        range_class = TimeRange
-                    case "date":
-                        range_class = DateRange
-                    case _:
-                        range_class = None
-
-                if range_class is not None:
-                    return [
-                        dataclasses.replace(node, values=range, children=children)
-                        for range in range_class.from_strings(node.values.values)
-                    ]
-            return [dataclasses.replace(node, children=children)]
-
-        children = [cc for c in self.root.children for cc in guess_datatypes(c)]
-        return CompressedTree(root=dataclasses.replace(self.root, children=children))
-
-    def select(
-        self,
-        selection: dict[str, str | list[str]],
-        mode: Literal["strict", "relaxed"] = "relaxed",
-    ) -> "CompressedTree":
-        # make all values lists
-        selection = {k: v if isinstance(v, list) else [v] for k, v in selection.items()}
-
-        def not_none(xs):
-            return [x for x in xs if x is not None]
-
-        def select(node: Node) -> Node | None:
-            # Check if the key is specified in the selection
-            if node.key not in selection:
-                if mode == "strict":
-                    return None
-                return dataclasses.replace(
-                    node, children=not_none(select(c) for c in node.children)
-                )
-
-            # If the key is specified, check if any of the values match
-            values = Enum([c for c in selection[node.key] if c in node.values])
-
-            if not values:
-                return None
-
-            return dataclasses.replace(
-                node, values=values, children=not_none(select(c) for c in node.children)
-            )
-
-        return CompressedTree(
-            root=dataclasses.replace(
-                self.root, children=not_none(select(c) for c in self.root.children)
-            )
-        )
-
-    def to_list_of_cubes(self):
-        def to_list_of_cubes(node: Node) -> list[list[Node]]:
-            return [
-                [node] + sub_cube
-                for c in node.children
-                for sub_cube in to_list_of_cubes(c)
-            ]
-
-        return to_list_of_cubes(self.root)
-
-    def info(self):
-        cubes = self.to_list_of_cubes()
-        print(f"Number of distinct paths: {len(cubes)}")
-
-
-# What should the interace look like?
-
-# tree = CompressedTree.from_json(...)
-# tree = CompressedTree.from_protobuf(...)
-
-# tree.print(depth = 5) # Prints a nice tree representation
--- a/pyproject.toml
+++ b/pyproject.toml
@ -19,6 +19,7 @@ dynamic = ["version"]
 dependencies = [
    "frozendict",
    "numpy",
+    "protobuf",

    # CLI
    "rich",
--- a/scripts/build_images.sh
+++ b/scripts/build_images.sh
@ -7,9 +7,3 @@ sudo docker build \
    --target=stac_server \
    .
 sudo docker push eccr.ecmwf.int/qubed/stac_server:latest
-
-sudo docker build \
-    --tag=eccr.ecmwf.int/qubed/web_query_builder:latest \
-    --target=web_query_builder \
-    .
-sudo docker push eccr.ecmwf.int/qubed/web_query_builder:latest
--- a/scripts/deploy.sh
+++ b/scripts/deploy.sh
@ -1 +1,2 @@
-helm upgrade stac-server chart -n stac-server
+# helm install qubed chart -n qubed
+helm upgrade qubed chart -n qubed
--- a/scripts/load_redis.py
+++ b/scripts/load_redis.py
@ -1,19 +0,0 @@
-#! .venv/bin/python
-
-import redis
-import yaml
-import json
-
-print("Opening redis connection")
-r = redis.Redis(host="redis", port=6379, db=0)
-
-print("Loading data from local files")
-with open("config/climate-dt/compressed_tree.json") as f:
-    compressed_catalog = json.load(f)
-
-with open("config/climate-dt/language.yaml") as f:
-    mars_language = yaml.safe_load(f)["_field"]
-
-print("Storing data in redis")
-r.set("compressed_catalog", json.dumps(compressed_catalog))
-r.set("mars_language", json.dumps(mars_language))
--- a/scripts/logs.sh
+++ b/scripts/logs.sh
@ -0,0 +1 @@
+kubectl -n qubed logs deployment/stac-server
--- a/scripts/restart.sh
+++ b/scripts/restart.sh
@ -1,3 +1,2 @@
 # kubectl rollout restart deployment/redis
-kubectl rollout restart deployment/web-query-builder
-kubectl rollout restart deployment/stac-server
+kubectl -n qubed rollout restart deployment/stac-server
--- a/scripts/setup.sh
+++ b/scripts/setup.sh
@ -1,3 +0,0 @@
-python3 -m venv .venv
-source .venv/bin/activate
-pip install pyyaml redis
--- a/src/python/qubed/Qube.py
+++ b/src/python/qubed/Qube.py
@ -1,90 +1,180 @@
-import dataclasses
-from collections import defaultdict
-from dataclasses import dataclass
-from functools import cached_property
-from typing import Any, Callable, Iterable, Literal, Sequence
+# This causes python types to be evaluated later,
+# allowing you to reference types like Qube inside the definion of the Qube class
+# without having to do "Qube"
+from __future__ import annotations

+import dataclasses
+import functools
+import json
+from collections import defaultdict
+from collections.abc import Callable
+from dataclasses import dataclass, field
+from functools import cached_property
+from pathlib import Path
+from typing import Any, Iterable, Iterator, Literal, Mapping, Self, Sequence
+
+import numpy as np
 from frozendict import frozendict

 from . import set_operations
-from .node_types import NodeData, RootNodeData
+from .metadata import from_nodes
+from .protobuf.adapters import proto_to_qube, qube_to_proto
 from .tree_formatters import (
    HTML,
+    _display,
    node_tree_to_html,
    node_tree_to_string,
 )
-from .value_types import QEnum, ValueGroup, values_from_json
+from .value_types import (
+    QEnum,
+    ValueGroup,
+    WildcardGroup,
+    values_from_json,
+)
+
+
+@dataclass
+class AxisInfo:
+    key: str
+    type: Any
+    depths: set[int]
+    values: set
+
+    def combine(self, other: Self):
+        self.key = other.key
+        self.type = other.type
+        self.depths.update(other.depths)
+        self.values.update(other.values)
+        # print(f"combining {self} and {other} getting {result}")
+
+    def to_json(self):
+        return {
+            "key": self.key,
+            "type": self.type.__name__,
+            "values": list(self.values),
+            "depths": list(self.depths),
+        }
+
+
+@dataclass(frozen=True, eq=True, order=True, unsafe_hash=True)
+class QubeNamedRoot:
+    "Helper class to print a custom root name"
+
+    key: str
+    children: tuple[Qube, ...] = ()
+
+    def summary(self) -> str:
+        return self.key


@dataclass(frozen=False, eq=True, order=True, unsafe_hash=True)
 class Qube:
-    data: NodeData
-    children: tuple["Qube", ...]
+    key: str
+    values: ValueGroup
+    metadata: frozendict[str, np.ndarray] = field(
+        default_factory=lambda: frozendict({}), compare=False
+    )
+    children: tuple[Qube, ...] = ()
+    is_root: bool = False
+    is_leaf: bool = False
+    depth: int = field(default=0, compare=False)
+    shape: tuple[int, ...] = field(default=(), compare=False)

-    @property
-    def key(self) -> str:
-        return self.data.key
+    @classmethod
+    def make_node(
+        cls,
+        key: str,
+        values: Iterable | QEnum | WildcardGroup,
+        children: Iterable[Qube],
+        metadata: Mapping[str, np.ndarray] = {},
+        is_root: bool = False,
+        is_leaf: bool | None = None,
+    ) -> Qube:
+        if isinstance(values, ValueGroup):
+            values = values
+        else:
+            values = QEnum(values)

-    @property
-    def values(self) -> ValueGroup:
-        return self.data.values
+        if not isinstance(values, WildcardGroup) and not is_root:
+            assert len(values) > 0, "Nodes must have at least one value"

-    @property
-    def metadata(self) -> frozendict[str, Any]:
-        return self.data.metadata
+        children = tuple(sorted(children, key=lambda n: ((n.key, n.values.min()))))

-    def replace(self, **kwargs) -> "Qube":
-        data_keys = {
-            k: v for k, v in kwargs.items() if k in ["key", "values", "metadata"]
-        }
-        node_keys = {k: v for k, v in kwargs.items() if k == "children"}
-        if not data_keys and not node_keys:
-            return self
-        if not data_keys:
-            return dataclasses.replace(self, **node_keys)
-
-        return dataclasses.replace(
-            self, data=dataclasses.replace(self.data, **data_keys), **node_keys
+        return cls(
+            key,
+            values=values,
+            children=children,
+            metadata=frozendict(metadata),
+            is_root=is_root,
+            is_leaf=(not len(children)) if is_leaf is None else is_leaf,
        )

+    @classmethod
+    def make_root(cls, children: Iterable[Qube], metadata={}) -> Qube:
+        def update_depth_shape(children, depth, shape):
+            for child in children:
+                child.depth = depth + 1
+                child.shape = shape + (len(child.values),)
+                update_depth_shape(child.children, child.depth, child.shape)
+
+        update_depth_shape(children, depth=0, shape=(1,))
+
+        return cls.make_node(
+            "root",
+            values=QEnum(("root",)),
+            children=children,
+            metadata=metadata,
+            is_root=True,
+        )
+
+    def replace(self, **kwargs) -> Qube:
+        return dataclasses.replace(self, **kwargs)
+
    def summary(self) -> str:
-        return self.data.summary()
+        if self.is_root:
+            return self.key
+        return f"{self.key}={self.values.summary()}" if self.key != "root" else "root"

    @classmethod
-    def make(cls, key: str, values: ValueGroup, children, **kwargs) -> "Qube":
-        return cls(
-            data=NodeData(key, values, metadata=kwargs.get("metadata", frozendict())),
-            children=tuple(sorted(children, key=lambda n: ((n.key, n.values.min())))),
-        )
+    def load(cls, path: str | Path) -> Qube:
+        with open(path, "r") as f:
+            return Qube.from_json(json.load(f))

    @classmethod
-    def root_node(cls, children: Iterable["Qube"]) -> "Qube":
-        return cls.make("root", QEnum(("root",)), children)
-
-    @classmethod
-    def from_datacube(cls, datacube: dict[str, str | Sequence[str]]) -> "Qube":
+    def from_datacube(cls, datacube: Mapping[str, str | Sequence[str]]) -> Qube:
        key_vals = list(datacube.items())[::-1]

-        children: list["Qube"] = []
+        children: list[Qube] = []
        for key, values in key_vals:
-            if not isinstance(values, list):
-                values = [values]
-            children = [cls.make(key, QEnum(values), children)]
+            values_group: ValueGroup
+            if values == "*":
+                values_group = WildcardGroup()
+            elif isinstance(values, list):
+                values_group = QEnum(values)
+            else:
+                values_group = QEnum([values])

-        return cls.root_node(children)
+            children = [cls.make_node(key, values_group, children)]
+
+        return cls.make_root(children)

    @classmethod
-    def from_json(cls, json: dict) -> "Qube":
-        def from_json(json: dict) -> Qube:
-            return Qube.make(
+    def from_json(cls, json: dict) -> Qube:
+        def from_json(json: dict, depth=0) -> Qube:
+            return Qube.make_node(
                key=json["key"],
                values=values_from_json(json["values"]),
                metadata=frozendict(json["metadata"]) if "metadata" in json else {},
-                children=(from_json(c) for c in json["children"]),
+                children=(from_json(c, depth + 1) for c in json["children"]),
+                is_root=(depth == 0),
            )

        return from_json(json)

+    @classmethod
+    def from_nodes(cls, nodes: dict[str, dict], add_root: bool = True):
+        return from_nodes(cls, nodes, add_root)
+
    def to_json(self) -> dict:
        def to_json(node: Qube) -> dict:
            return {
@ -97,73 +187,161 @@ class Qube:
        return to_json(self)

    @classmethod
-    def from_dict(cls, d: dict) -> "Qube":
-        def from_dict(d: dict) -> list[Qube]:
-            return [
-                Qube.make(
-                    key=k.split("=")[0],
-                    values=QEnum((k.split("=")[1].split("/"))),
+    def from_dict(cls, d: dict) -> Qube:
+        def from_dict(d: dict) -> Iterator[Qube]:
+            for k, children in d.items():
+                key, values = k.split("=")
+                values = values.split("/")
+                # children == {"..." : {}}
+                # is a special case to represent trees with leaves we don't know about
+                if frozendict(children) == frozendict({"...": {}}):
+                    yield Qube.make_node(
+                        key=key,
+                        values=values,
+                        children={},
+                        is_leaf=False,
+                    )
+
+                # Special case for Wildcard values
+                if values == ["*"]:
+                    values = WildcardGroup()
+                else:
+                    values = QEnum(values)
+
+                yield Qube.make_node(
+                    key=key,
+                    values=values,
                    children=from_dict(children),
                )
-                for k, children in d.items()
-            ]

-        return Qube.root_node(from_dict(d))
+        return Qube.make_root(list(from_dict(d)))
+
+    def to_dict(self) -> dict:
+        def to_dict(q: Qube) -> tuple[str, dict]:
+            key = f"{q.key}={','.join(str(v) for v in q.values)}"
+            return key, dict(to_dict(c) for c in q.children)
+
+        return to_dict(self)[1]

    @classmethod
-    def empty(cls) -> "Qube":
-        return Qube.root_node([])
+    def from_protobuf(cls, msg: bytes) -> Qube:
+        return proto_to_qube(cls, msg)

-    def __str__(self, depth=None, name=None) -> str:
-        node = (
-            dataclasses.replace(
-                self,
-                data=RootNodeData(key=name, values=self.values, metadata=self.metadata),
-            )
-            if name is not None
-            else self
-        )
-        return "".join(node_tree_to_string(node=node, depth=depth))
+    def to_protobuf(self) -> bytes:
+        return qube_to_proto(self)
+
+    @classmethod
+    def from_tree(cls, tree_str):
+        lines = tree_str.splitlines()
+        stack = []
+        root = {}
+
+        initial_indent = None
+        for line in lines:
+            if not line.strip():
+                continue
+            # Remove tree characters and measure indent level
+            stripped = line.lstrip(" │├└─")
+            indent = (len(line) - len(stripped)) // 4
+            if initial_indent is None:
+                initial_indent = indent
+            indent = indent - initial_indent
+
+            # Split multiple key=value parts into nested structure
+            keys = [item.strip() for item in stripped.split(",")]
+            current = bottom = {}
+            for key in reversed(keys):
+                current = {key: current}
+
+            # Adjust the stack to current indent level
+            # print(len(stack), stack)
+            while len(stack) > indent:
+                stack.pop()
+
+            if stack:
+                # Add to the dictionary at current stack level
+                parent = stack[-1]
+                key = list(current.keys())[0]
+                if key in parent:
+                    raise ValueError(
+                        f"This function doesn't yet support reading in uncompressed trees, repeated key is {key}"
+                    )
+                parent[key] = current[key]
+            else:
+                # Top level
+                key = list(current.keys())[0]
+                if root:
+                    raise ValueError(
+                        f"This function doesn't yet support reading in uncompressed trees, repeated key is {key}"
+                    )
+                root = current[key]
+
+            # Push to the stack
+            stack.append(bottom)
+
+        return cls.from_dict(root)
+
+    @classmethod
+    def empty(cls) -> Qube:
+        return Qube.make_root([])
+
+    def __str_helper__(self, depth=None, name=None) -> str:
+        node = self
+        if name is not None:
+            node = node.replace(key=name)
+        out = "".join(node_tree_to_string(node=node, depth=depth))
+        if out[-1] == "\n":
+            out = out[:-1]
+        return out
+
+    def __str__(self):
+        return self.__str_helper__()
+
+    def __repr__(self):
+        return f"Qube({self.__str_helper__()})"

    def print(self, depth=None, name: str | None = None):
-        print(self.__str__(depth=depth, name=name))
+        print(self.__str_helper__(depth=depth, name=name))

-    def html(self, depth=2, collapse=True, name: str | None = None) -> HTML:
-        node = (
-            dataclasses.replace(
-                self,
-                data=RootNodeData(key=name, values=self.values, metadata=self.metadata),
-            )
-            if name is not None
-            else self
+    def html(
+        self,
+        depth=2,
+        collapse=True,
+        name: str | None = None,
+        info: Callable[[Qube], str] | None = None,
+    ) -> HTML:
+        node = self
+        if name is not None:
+            node = node.replace(key=name)
+        return HTML(
+            node_tree_to_html(node=node, depth=depth, collapse=collapse, info=info)
        )
-        return HTML(node_tree_to_html(node=node, depth=depth, collapse=collapse))

    def _repr_html_(self) -> str:
        return node_tree_to_html(self, depth=2, collapse=True)

    # Allow "key=value/value" / qube to prepend keys
-    def __rtruediv__(self, other: str) -> "Qube":
+    def __rtruediv__(self, other: str) -> Qube:
        key, values = other.split("=")
-        values = QEnum((values.split("/")))
-        return Qube.root_node([Qube.make(key, values, self.children)])
+        values_enum = QEnum((values.split("/")))
+        return Qube.make_root([Qube.make_node(key, values_enum, self.children)])

-    def __or__(self, other: "Qube") -> "Qube":
+    def __or__(self, other: Qube) -> Qube:
        return set_operations.operation(
            self, other, set_operations.SetOperation.UNION, type(self)
        )

-    def __and__(self, other: "Qube") -> "Qube":
+    def __and__(self, other: Qube) -> Qube:
        return set_operations.operation(
            self, other, set_operations.SetOperation.INTERSECTION, type(self)
        )

-    def __sub__(self, other: "Qube") -> "Qube":
+    def __sub__(self, other: Qube) -> Qube:
        return set_operations.operation(
            self, other, set_operations.SetOperation.DIFFERENCE, type(self)
        )

-    def __xor__(self, other: "Qube") -> "Qube":
+    def __xor__(self, other: Qube) -> Qube:
        return set_operations.operation(
            self, other, set_operations.SetOperation.SYMMETRIC_DIFFERENCE, type(self)
        )
@ -179,43 +357,85 @@ class Qube:
                    else:
                        yield leaf

-    def datacubes(self) -> "Qube":
-        def to_list_of_cubes(node: Qube) -> Iterable[Qube]:
-            if not node.children:
-                yield node
-            # print(node.key)
-            for c in node.children:
-                # print(c)
-                for sub_cube in to_list_of_cubes(c):
-                    yield dataclasses.replace(node, children=[sub_cube])
+    def leaf_nodes(self) -> "Iterable[tuple[dict[str, str], Qube]]":
+        for value in self.values:
+            if not self.children:
+                yield ({self.key: value}, self)
+            for child in self.children:
+                for leaf in child.leaf_nodes():
+                    if self.key != "root":
+                        yield ({self.key: value, **leaf[0]}, leaf[1])
+                    else:
+                        yield leaf

-        return Qube.root_node((q for c in self.children for q in to_list_of_cubes(c)))
+    def leaves_with_metadata(
+        self, indices=()
+    ) -> Iterator[tuple[dict[str, str], dict[str, str | np.ndarray]]]:
+        if self.key == "root":
+            for c in self.children:
+                yield from c.leaves_with_metadata(indices=())
+            return

-    def __getitem__(self, args) -> "Qube":
+        for index, value in enumerate(self.values):
+            indexed_metadata = {
+                k: vs[indices + (index,)] for k, vs in self.metadata.items()
+            }
+            indexed_metadata = {
+                k: v.item() if v.shape == () else v for k, v in indexed_metadata.items()
+            }
+            if not self.children:
+                yield {self.key: value}, indexed_metadata
+
+            for child in self.children:
+                for leaf, metadata in child.leaves_with_metadata(
+                    indices=indices + (index,)
+                ):
+                    if self.key != "root":
+                        yield {self.key: value, **leaf}, metadata | indexed_metadata
+                    else:
+                        yield leaf, metadata
+
+    def datacubes(self) -> Iterable[dict[str, Any | list[Any]]]:
+        def to_list_of_cubes(node: Qube) -> Iterable[dict[str, Any | list[Any]]]:
+            if node.key == "root":
+                for c in node.children:
+                    yield from to_list_of_cubes(c)
+
+            else:
+                if not node.children:
+                    yield {node.key: list(node.values)}
+
+                for c in node.children:
+                    for sub_cube in to_list_of_cubes(c):
+                        yield {node.key: list(node.values)} | sub_cube
+
+        return to_list_of_cubes(self)
+
+    def __getitem__(self, args) -> Qube:
        if isinstance(args, str):
            specifiers = args.split(",")
            current = self
            for specifier in specifiers:
-                key, values = specifier.split("=")
-                values = values.split("/")
+                key, values_str = specifier.split("=")
+                values = values_str.split("/")
                for c in current.children:
                    if c.key == key and set(values) == set(c.values):
                        current = c
                        break
                else:
                    raise KeyError(
-                        f"Key '{key}' not found in children of '{current.key}'"
+                        f"Key '{key}' not found in children of '{current.key}', available keys are {[c.key for c in current.children]}"
                    )
-            return Qube.root_node(current.children)
+            return Qube.make_root(current.children)

        elif isinstance(args, tuple) and len(args) == 2:
            key, value = args
            for c in self.children:
                if c.key == key and value in c.values:
-                    return Qube.root_node(c.children)
-            raise KeyError(f"Key {key} not found in children of {self.key}")
+                    return Qube.make_root(c.children)
+            raise KeyError(f"Key '{key}' not found in children of {self.key}")
        else:
-            raise ValueError("Unknown key type")
+            raise ValueError(f"Unknown key type {args}")

    @cached_property
    def n_leaves(self) -> int:
@ -232,7 +452,7 @@ class Qube:
            return 0
        return 1 + sum(c.n_nodes for c in self.children)

-    def transform(self, func: "Callable[[Qube], Qube | Iterable[Qube]]") -> "Qube":
+    def transform(self, func: "Callable[[Qube], Qube | Iterable[Qube]]") -> Qube:
        """
        Call a function on every node of the Qube, return one or more nodes.
        If multiple nodes are returned they each get a copy of the (transformed) children of the original node.
@ -240,7 +460,7 @@ class Qube:
        """

        def transform(node: Qube) -> list[Qube]:
-            children = [cc for c in node.children for cc in transform(c)]
+            children = tuple(sorted(cc for c in node.children for cc in transform(c)))
            new_nodes = func(node)
            if isinstance(new_nodes, Qube):
                new_nodes = [new_nodes]
@ -248,48 +468,128 @@ class Qube:
            return [new_node.replace(children=children) for new_node in new_nodes]

        children = tuple(cc for c in self.children for cc in transform(c))
-        return dataclasses.replace(self, children=children)
+        return self.replace(children=children)
+
+    def remove_by_key(self, keys: str | list[str]):
+        _keys: list[str] = keys if isinstance(keys, list) else [keys]
+
+        def remove_key(node: Qube) -> Qube:
+            children: list[Qube] = []
+            for c in node.children:
+                if c.key in _keys:
+                    grandchildren = tuple(sorted(remove_key(cc) for cc in c.children))
+                    grandchildren = remove_key(Qube.make_root(grandchildren)).children
+                    children.extend(grandchildren)
+                else:
+                    children.append(remove_key(c))
+
+            return node.replace(children=tuple(sorted(children)))
+
+        return remove_key(self).compress()
+
+    def convert_dtypes(self, converters: dict[str, Callable[[Any], Any]]):
+        def convert(node: Qube) -> Qube:
+            if node.key in converters:
+                converter = converters[node.key]
+                values = [converter(v) for v in node.values]
+                new_node = node.replace(values=QEnum(values))
+                return new_node
+            return node
+
+        return self.transform(convert)

    def select(
        self,
-        selection: dict[str, str | list[str]],
+        selection: dict[str, str | list[str] | Callable[[Any], bool]],
        mode: Literal["strict", "relaxed"] = "relaxed",
-        prune=True,
-    ) -> "Qube":
-        # make all values lists
-        selection = {k: v if isinstance(v, list) else [v] for k, v in selection.items()}
+        consume=False,
+    ) -> Qube:
+        # Find any bare str values and replace them with [str]
+        _selection: dict[str, list[str] | Callable[[Any], bool]] = {}
+        for k, v in selection.items():
+            if isinstance(v, list):
+                _selection[k] = v
+            elif callable(v):
+                _selection[k] = v
+            else:
+                _selection[k] = [v]

        def not_none(xs):
            return tuple(x for x in xs if x is not None)

-        def select(node: Qube) -> Qube | None:
-            # Check if the key is specified in the selection
+        def select(
+            node: Qube,
+            selection: dict[str, list[str] | Callable[[Any], bool]],
+            matched: bool,
+        ) -> Qube | None:
+            # If this node has no children but there are still parts of the request
+            # that have not been consumed, then prune this whole branch
+            if consume and not node.children and selection:
+                return None
+
+            # If the key isn't in the selection then what we do depends on the mode:
+            # In strict mode we just stop here
+            # In next_level mode we include the next level down so you can tell what keys to add next
+            # In relaxed mode we skip the key if it't not in the request and carry on
            if node.key not in selection:
                if mode == "strict":
                    return None

-                new_children = not_none(select(c) for c in node.children)
+                elif mode == "next_level":
+                    return node.replace(
+                        children=(),
+                        metadata=self.metadata
+                        | {"is_leaf": np.array([not bool(node.children)])},
+                    )

-                # prune==true then remove any non-leaf nodes
-                # which have had all their children removed
-                if prune and node.children and not new_children:
+                elif mode == "relaxed":
+                    pass
+                else:
+                    raise ValueError(f"Unknown mode argument {mode}")
+
+            # If the key IS in the selection then check if the values match
+            if node.key in _selection:
+                # If the key is specified, check if any of the values match
+                selection_criteria = _selection[node.key]
+                if callable(selection_criteria):
+                    values = QEnum((c for c in node.values if selection_criteria(c)))
+                elif isinstance(selection_criteria, list):
+                    values = QEnum((c for c in selection_criteria if c in node.values))
+                else:
+                    raise ValueError(f"Unknown selection type {selection_criteria}")
+
+                # Here modes don't matter because we've explicitly filtered on this key and found nothing
+                if not values:
                    return None

-                return dataclasses.replace(node, children=new_children)
+                matched = True
+                node = node.replace(values=values)

-            # If the key is specified, check if any of the values match
-            values = QEnum((c for c in selection[node.key] if c in node.values))
+            if consume:
+                selection = {k: v for k, v in selection.items() if k != node.key}

-            if not values:
-                return None
-
-            data = dataclasses.replace(node.data, values=values)
-            return dataclasses.replace(
-                node, data=data, children=not_none(select(c) for c in node.children)
+            # Prune nodes that had had all their children pruned
+            new_children = not_none(
+                select(c, selection, matched) for c in node.children
            )

-        return dataclasses.replace(
-            self, children=not_none(select(c) for c in self.children)
+            if node.children and not new_children:
+                return None
+
+            metadata = dict(node.metadata)
+
+            if mode == "next_level":
+                metadata["is_leaf"] = np.array([not bool(node.children)])
+
+            return node.replace(
+                children=new_children,
+                metadata=metadata,
+            )
+
+        return self.replace(
+            children=not_none(
+                select(c, _selection, matched=False) for c in self.children
+            )
        )

    def span(self, key: str) -> list[str]:
@ -311,6 +611,26 @@ class Qube:
            axes[self.key].update(self.values)
        return dict(axes)

+    def axes_info(self, depth=0) -> dict[str, AxisInfo]:
+        axes = defaultdict(
+            lambda: AxisInfo(key="", type=str, depths=set(), values=set())
+        )
+        for c in self.children:
+            for k, info in c.axes_info(depth=depth + 1).items():
+                axes[k].combine(info)
+
+        if self.key != "root":
+            axes[self.key].combine(
+                AxisInfo(
+                    key=self.key,
+                    type=type(next(iter(self.values))),
+                    depths={depth},
+                    values=set(self.values),
+                )
+            )
+
+        return dict(axes)
+
    @cached_property
    def structural_hash(self) -> int:
        """
@ -325,12 +645,48 @@ class Qube:

        return hash_node(self)

-    def compress(self) -> "Qube":
-        # First compress the children (this recursively compresses all the way to the leaves)
-        new_children = [child.compress() for child in self.children]
+    def compress(self) -> Qube:
+        """
+        This method is quite computationally heavy because of trees like this:
+        root, class=d1, generation=1
+        ├── time=0600, many identical keys, param=8,78,79
+        ├── time=0600, many identical keys, param=8,78,79
+        └── time=0600, many identical keys, param=8,78,79
+        This tree compresses dow n

-        # Now compress the set of children at this level
-        new_children = set_operations.compress_children(new_children)
+        """

-        # Return the now compressed node
-        return Qube.make(self.key, self.values, new_children)
+        def union(a: Qube, b: Qube) -> Qube:
+            b = type(self).make_root(children=(b,))
+            out = set_operations.operation(
+                a, b, set_operations.SetOperation.UNION, type(self)
+            )
+            return out
+
+        new_children = [c.compress() for c in self.children]
+        if len(new_children) > 1:
+            new_children = list(
+                functools.reduce(union, new_children, Qube.empty()).children
+            )
+
+        return self.replace(children=tuple(sorted(new_children)))
+
+    def add_metadata(self, **kwargs: dict[str, Any]):
+        metadata = {
+            k: np.array(
+                [
+                    v,
+                ]
+            )
+            for k, v in kwargs.items()
+        }
+        return self.replace(metadata=metadata)
+
+    def strip_metadata(self) -> Qube:
+        def strip(node):
+            return node.replace(metadata=frozendict({}))
+
+        return self.transform(strip)
+
+    def display(self):
+        _display(self)
--- a/src/python/qubed/init.py
+++ b/src/python/qubed/init.py
@ -1,3 +1,4 @@
+from . import protobuf
 from .Qube import Qube

-__all__ = ["Qube"]
+__all__ = ["Qube", "protobuf"]
--- a/src/python/qubed/main.py
+++ b/src/python/qubed/main.py
@ -1,3 +1,4 @@
+import json
 import time

 import click
@ -107,7 +108,15 @@ def convert(input, output, from_format, to_format):
                n0 = q.n_leaves
                t = time.time()

-    output_content = str(q) if to_format == "text" else q.html().html
+    if to_format == "text":
+        output_content = str(q)
+    elif to_format == "json":
+        output_content = json.dumps(q.to_json())
+    elif to_format == "html":
+        output_content = q.html().html
+    else:
+        output_content = str(q)
+
    output.write(output_content)


--- a/src/python/qubed/convert.py
+++ b/src/python/qubed/convert.py
@ -8,8 +8,10 @@ def parse_key_value_pairs(text: str):
    for segment in text.split(","):
        if "=" not in segment:
            print(segment)
-        key, values = segment.split("=", 1)  # Ensure split only happens at first "="
-        values = values.split("/")
+        key, values_str = segment.split(
+            "=", 1
+        )  # Ensure split only happens at first "="
+        values = values_str.split("/")
        result[key] = values

    return result
--- a/src/python/qubed/metadata.py
+++ b/src/python/qubed/metadata.py
@ -0,0 +1,43 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Iterator
+
+import numpy as np
+
+from .value_types import QEnum
+
+if TYPE_CHECKING:
+    from .Qube import Qube
+
+
+def make_node(
+    cls,
+    key: str,
+    values: Iterator,
+    shape: list[int],
+    children: tuple[Qube, ...],
+    metadata: dict[str, np.ndarray] | None = None,
+):
+    return cls.make_node(
+        key=key,
+        values=QEnum(values),
+        metadata={k: np.array(v).reshape(shape) for k, v in metadata.items()}
+        if metadata is not None
+        else {},
+        children=children,
+    )
+
+
+def from_nodes(cls, nodes, add_root=True):
+    shape = [len(n["values"]) for n in nodes.values()]
+    nodes = nodes.items()
+    *nodes, (key, info) = nodes
+    root = make_node(cls, shape=shape, children=(), key=key, **info)
+
+    for key, info in reversed(nodes):
+        shape.pop()
+        root = make_node(cls, shape=shape, children=(root,), key=key, **info)
+
+    if add_root:
+        return cls.make_root(children=(root,))
+    return root
--- a/src/python/qubed/node_types.py
+++ b/src/python/qubed/node_types.py
@ -1,26 +0,0 @@
-from dataclasses import dataclass, field
-from typing import Hashable
-
-from frozendict import frozendict
-
-from .value_types import ValueGroup
-
-
-@dataclass(frozen=False, eq=True, order=True, unsafe_hash=True)
-class NodeData:
-    key: str
-    values: ValueGroup
-    metadata: dict[str, tuple[Hashable, ...]] = field(
-        default_factory=frozendict, compare=False
-    )
-
-    def summary(self) -> str:
-        return f"{self.key}={self.values.summary()}" if self.key != "root" else "root"
-
-
-@dataclass(frozen=False, eq=True, order=True)
-class RootNodeData(NodeData):
-    "Helper class to print a custom root name"
-
-    def summary(self) -> str:
-        return self.key
--- a/src/python/qubed/protobuf/init.py
+++ b/src/python/qubed/protobuf/init.py
--- a/src/python/qubed/protobuf/adapters.py
+++ b/src/python/qubed/protobuf/adapters.py
@ -0,0 +1,109 @@
+from __future__ import annotations
+
+import warnings
+from typing import TYPE_CHECKING
+
+import numpy as np
+from frozendict import frozendict
+
+from ..value_types import QEnum
+
+with warnings.catch_warnings():
+    warnings.filterwarnings(
+        "ignore",
+        "Protobuf gencode version",
+        UserWarning,
+        "google.protobuf.runtime_version",
+    )
+    from . import qube_pb2
+
+
+if TYPE_CHECKING:
+    from ..Qube import Qube
+
+
+def _ndarray_to_proto(arr: np.ndarray) -> qube_pb2.NdArray:
+    """np.ndarray → NdArray message"""
+    return qube_pb2.NdArray(
+        shape=list(arr.shape),
+        dtype=str(arr.dtype),
+        raw=arr.tobytes(order="C"),
+    )
+
+
+def _ndarray_from_proto(msg: qube_pb2.NdArray) -> np.ndarray:
+    """NdArray message → np.ndarray (immutable view)"""
+    return np.frombuffer(msg.raw, dtype=msg.dtype).reshape(tuple(msg.shape))
+
+
+def _py_to_valuegroup(value: list[str] | np.ndarray) -> qube_pb2.ValueGroup:
+    """Accept str-sequence *or* ndarray and return ValueGroup."""
+    vg = qube_pb2.ValueGroup()
+    if isinstance(value, np.ndarray):
+        vg.tensor.CopyFrom(_ndarray_to_proto(value))
+    else:
+        vg.s.items.extend(value)
+    return vg
+
+
+def _valuegroup_to_py(vg: qube_pb2.ValueGroup) -> list[str] | np.ndarray:
+    """ValueGroup → list[str]  *or* ndarray"""
+    arm = vg.WhichOneof("payload")
+    if arm == "tensor":
+        return _ndarray_from_proto(vg.tensor)
+
+    return QEnum(vg.s.items)
+
+
+def _py_to_metadatagroup(value: np.ndarray) -> qube_pb2.MetadataGroup:
+    """Accept str-sequence *or* ndarray and return ValueGroup."""
+    vg = qube_pb2.MetadataGroup()
+    if not isinstance(value, np.ndarray):
+        value = np.array([value])
+
+    vg.tensor.CopyFrom(_ndarray_to_proto(value))
+    return vg
+
+
+def _metadatagroup_to_py(vg: qube_pb2.MetadataGroup) -> np.ndarray:
+    """ValueGroup → list[str]  *or* ndarray"""
+    arm = vg.WhichOneof("payload")
+    if arm == "tensor":
+        return _ndarray_from_proto(vg.tensor)
+
+    raise ValueError(f"Unknown arm {arm}")
+
+
+def _qube_to_proto(q: Qube) -> qube_pb2.Qube:
+    """Frozen Qube dataclass → protobuf Qube message (new object)."""
+    return qube_pb2.Qube(
+        key=q.key,
+        values=_py_to_valuegroup(q.values),
+        metadata={k: _py_to_metadatagroup(v) for k, v in q.metadata.items()},
+        children=[_qube_to_proto(c) for c in q.children],
+        is_root=q.is_root,
+    )
+
+
+def qube_to_proto(q: Qube) -> bytes:
+    return _qube_to_proto(q).SerializeToString()
+
+
+def _proto_to_qube(cls: type, msg: qube_pb2.Qube) -> Qube:
+    """protobuf Qube message → frozen Qube dataclass (new object)."""
+
+    return cls.make_node(
+        key=msg.key,
+        values=_valuegroup_to_py(msg.values),
+        metadata=frozendict(
+            {k: _metadatagroup_to_py(v) for k, v in msg.metadata.items()}
+        ),
+        children=tuple(_proto_to_qube(cls, c) for c in msg.children),
+        is_root=msg.is_root,
+    )
+
+
+def proto_to_qube(cls: type, wire: bytes) -> Qube:
+    msg = qube_pb2.Qube()
+    msg.ParseFromString(wire)
+    return _proto_to_qube(cls, msg)
--- a/src/python/qubed/protobuf/qube_pb2.py
+++ b/src/python/qubed/protobuf/qube_pb2.py
@ -0,0 +1,45 @@
+# -*- coding: utf-8 -*-
+# Generated by the protocol buffer compiler.  DO NOT EDIT!
+# NO CHECKED-IN PROTOBUF GENCODE
+# source: qube.proto
+# Protobuf Python Version: 5.29.0
+"""Generated protocol buffer code."""
+
+from google.protobuf import descriptor as _descriptor
+from google.protobuf import descriptor_pool as _descriptor_pool
+from google.protobuf import runtime_version as _runtime_version
+from google.protobuf import symbol_database as _symbol_database
+from google.protobuf.internal import builder as _builder
+
+_runtime_version.ValidateProtobufRuntimeVersion(
+    _runtime_version.Domain.PUBLIC, 5, 29, 0, "", "qube.proto"
+)
+# @@protoc_insertion_point(imports)
+
+_sym_db = _symbol_database.Default()
+
+
+DESCRIPTOR = _descriptor_pool.Default().AddSerializedFile(
+    b'\n\nqube.proto"4\n\x07NdArray\x12\r\n\x05shape\x18\x01 \x03(\x03\x12\r\n\x05\x64type\x18\x02 \x01(\t\x12\x0b\n\x03raw\x18\x03 \x01(\x0c"\x1c\n\x0bStringGroup\x12\r\n\x05items\x18\x01 \x03(\t"N\n\nValueGroup\x12\x19\n\x01s\x18\x01 \x01(\x0b\x32\x0c.StringGroupH\x00\x12\x1a\n\x06tensor\x18\x02 \x01(\x0b\x32\x08.NdArrayH\x00\x42\t\n\x07payload"6\n\rMetadataGroup\x12\x1a\n\x06tensor\x18\x01 \x01(\x0b\x32\x08.NdArrayH\x00\x42\t\n\x07payload"\xd1\x01\n\x04Qube\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\x1b\n\x06values\x18\x02 \x01(\x0b\x32\x0b.ValueGroup\x12%\n\x08metadata\x18\x03 \x03(\x0b\x32\x13.Qube.MetadataEntry\x12\r\n\x05\x64type\x18\x04 \x01(\t\x12\x17\n\x08\x63hildren\x18\x05 \x03(\x0b\x32\x05.Qube\x12\x0f\n\x07is_root\x18\x06 \x01(\x08\x1a?\n\rMetadataEntry\x12\x0b\n\x03key\x18\x01 \x01(\t\x12\x1d\n\x05value\x18\x02 \x01(\x0b\x32\x0e.MetadataGroup:\x02\x38\x01\x62\x06proto3'
+)
+
+_globals = globals()
+_builder.BuildMessageAndEnumDescriptors(DESCRIPTOR, _globals)
+_builder.BuildTopDescriptorsAndMessages(DESCRIPTOR, "qube_pb2", _globals)
+if not _descriptor._USE_C_DESCRIPTORS:
+    DESCRIPTOR._loaded_options = None
+    _globals["_QUBE_METADATAENTRY"]._loaded_options = None
+    _globals["_QUBE_METADATAENTRY"]._serialized_options = b"8\001"
+    _globals["_NDARRAY"]._serialized_start = 14
+    _globals["_NDARRAY"]._serialized_end = 66
+    _globals["_STRINGGROUP"]._serialized_start = 68
+    _globals["_STRINGGROUP"]._serialized_end = 96
+    _globals["_VALUEGROUP"]._serialized_start = 98
+    _globals["_VALUEGROUP"]._serialized_end = 176
+    _globals["_METADATAGROUP"]._serialized_start = 178
+    _globals["_METADATAGROUP"]._serialized_end = 232
+    _globals["_QUBE"]._serialized_start = 235
+    _globals["_QUBE"]._serialized_end = 444
+    _globals["_QUBE_METADATAENTRY"]._serialized_start = 381
+    _globals["_QUBE_METADATAENTRY"]._serialized_end = 444
+# @@protoc_insertion_point(module_scope)
--- a/src/python/qubed/py.typed
+++ b/src/python/qubed/py.typed
--- a/src/python/qubed/set_operations.py
+++ b/src/python/qubed/set_operations.py
@ -1,186 +1,464 @@
+"""
+# Set Operations
+
+The core of this is the observation that for two sets A and B, if we compute (A - B), (A ∩ B) amd (B - A)
+then we can get the other operations by taking unions of the above three objects.
+Union: All of them
+Intersection: Just take A ∩ B
+Difference: Take either A - B or B - A
+Symmetric Difference (XOR): Take A - B and B - A
+
+We start with a shallow implementation of this algorithm that only deals with a pair of nodes, not the whole tree:
+
+shallow_set_operation(A: Qube, B: Qube) -> SetOpsResult
+
+This takes two qubes and (morally) returns (A - B), (A ∩ B) amd (B - A) but only for the values and metadata at the top level.
+
+For technical reasons that will become clear we actually return a struct with two copies of (A ∩ B). One has the metadata from A and the children of A call it A', and the other has them from B call it B'. This is relevant when we extend the shallow algorithm to work with a whole tree because we will recurse and compute the set operation for each pair of the children of A' and B'.
+
+NB: Currently there are two kinds of values, QEnums, that store a list of values and Wildcards that 'match with everything'. shallow_set_operation checks the type of values and dispatches to different methods depending on the combination of types it finds.
+
+"""
+
+from __future__ import annotations
+
 from collections import defaultdict
-from dataclasses import replace
+from dataclasses import dataclass
 from enum import Enum

 # Prevent circular imports while allowing the type checker to know what Qube is
-from typing import TYPE_CHECKING, Iterable
+from typing import TYPE_CHECKING, Any, Iterable

+import numpy as np
 from frozendict import frozendict

-from .node_types import NodeData
-from .value_types import QEnum, ValueGroup
+from .value_types import QEnum, ValueGroup, WildcardGroup

 if TYPE_CHECKING:
-    from .qube import Qube
+    from .Qube import Qube


 class SetOperation(Enum):
+    "Map from set operations to which combination of (A - B), (A ∩ B), (B - A) we need."
+
    UNION = (1, 1, 1)
    INTERSECTION = (0, 1, 0)
    DIFFERENCE = (1, 0, 0)
    SYMMETRIC_DIFFERENCE = (1, 0, 1)


-def fused_set_operations(
-    A: "ValueGroup", B: "ValueGroup"
-) -> tuple[list[ValueGroup], list[ValueGroup], list[ValueGroup]]:
-    if isinstance(A, QEnum) and isinstance(B, QEnum):
-        set_A, set_B = set(A), set(B)
-        intersection = set_A & set_B
-        just_A = set_A - intersection
-        just_B = set_B - intersection
-        return (
-            [
-                QEnum(just_A),
-            ],
-            [
-                QEnum(intersection),
-            ],
-            [
-                QEnum(just_B),
-            ],
+@dataclass(eq=True, frozen=True)
+class ValuesIndices:
+    "Helper class to hold the values and indices from a node."
+
+    values: ValueGroup
+    indices: tuple[int, ...]
+
+    @classmethod
+    def from_values(cls, values: ValueGroup):
+        return cls(values=values, indices=tuple(range(len(values))))
+
+    @classmethod
+    def empty(cls):
+        return cls(values=QEnum([]), indices=())
+
+    def enumerate(self) -> Iterable[tuple[Any, int]]:
+        return zip(self.indices, self.values)
+
+
+def get_indices(
+    metadata: frozendict[str, np.ndarray], indices: tuple[int, ...]
+) -> frozendict[str, np.ndarray]:
+    "Given a metadata dict and some indices, return a new metadata dict with only the values indexed by the indices"
+    return frozendict(
+        {k: v[..., indices] for k, v in metadata.items() if isinstance(v, np.ndarray)}
+    )
+
+
+@dataclass(eq=True, frozen=True)
+class SetOpResult:
+    """
+    Given two sets A and B, all possible set operations can be constructed from A - B, A ∩ B, B - A
+    That is, what's only in A, the intersection and what's only in B
+    However because we need to recurse on children we actually return two intersection node:
+    only_A is a qube with:
+        The values in A but not in B
+        The metadata corresponding to this values
+        All the children A had
+
+    intersection_A is a qube with:
+      The values that intersected with B
+      The metadata from that intersection
+      All the children A had
+
+    And vice versa for only_B and intersection B
+    """
+
+    only_A: ValuesIndices
+    intersection_A: ValuesIndices
+    intersection_B: ValuesIndices
+    only_B: ValuesIndices
+
+
+def shallow_qenum_set_operation(A: ValuesIndices, B: ValuesIndices) -> SetOpResult:
+    """
+    For two sets of values, partition the overlap into four groups:
+    only_A: values and indices of values that are in A but not B
+    intersection_A: values and indices of values that are in both A and B
+    And vice versa for only_B and intersection_B.
+
+    Note that intersection_A and intersection_B contain the same values but the indices are different.
+    """
+
+    # create four groups that map value -> index
+    only_A: dict[Any, int] = {val: i for i, val in A.enumerate()}
+    only_B: dict[Any, int] = {val: i for i, val in B.enumerate()}
+    intersection_A: dict[Any, int] = {}
+    intersection_B: dict[Any, int] = {}
+
+    # Go through all the values and move any that are in the intersection
+    # to the corresponding group, keeping the indices
+    for val in A.values:
+        if val in B.values:
+            intersection_A[val] = only_A.pop(val)
+            intersection_B[val] = only_B.pop(val)
+
+    def package(values_indices: dict[Any, int]) -> ValuesIndices:
+        return ValuesIndices(
+            values=QEnum(list(values_indices.keys())),
+            indices=tuple(values_indices.values()),
        )

-    raise NotImplementedError(
-        "Fused set operations on values types other than QEnum are not yet implemented"
+    return SetOpResult(
+        only_A=package(only_A),
+        only_B=package(only_B),
+        intersection_A=package(intersection_A),
+        intersection_B=package(intersection_B),
    )


-def node_intersection(
-    A: "ValueGroup", B: "ValueGroup"
-) -> tuple[ValueGroup, ValueGroup, ValueGroup]:
-    if isinstance(A, QEnum) and isinstance(B, QEnum):
-        set_A, set_B = set(A), set(B)
-        intersection = set_A & set_B
-        just_A = set_A - intersection
-        just_B = set_B - intersection
-        return QEnum(just_A), QEnum(intersection), QEnum(just_B)
+def shallow_wildcard_set_operation(A: ValuesIndices, B: ValuesIndices) -> SetOpResult:
+    """
+    WildcardGroups behave as if they contain all the values of whatever they match against.
+    For two wildcards we just return both.
+    For A == wildcard and B == enum we have to be more careful:
+        1. All of B is in the intersection so only_B is None too.
+        2. The wildcard may need to match against other things so only_A is A
+        3. We return B in the intersection_B and intersection_A slot.
+
+    This last bit happens because the wildcard basically adopts the values of whatever it sees.
+    """
+    # Two wildcard groups have full overlap.
+    if isinstance(A.values, WildcardGroup) and isinstance(B.values, WildcardGroup):
+        return SetOpResult(ValuesIndices.empty(), A, B, ValuesIndices.empty())
+
+    # If A is a wildcard matcher and B is not
+    # then the intersection is everything from B
+    if isinstance(A.values, WildcardGroup):
+        return SetOpResult(A, B, B, ValuesIndices.empty())
+
+    # If B is a wildcard matcher and A is not
+    # then the intersection is everything from A
+    if isinstance(B.values, WildcardGroup):
+        return SetOpResult(ValuesIndices.empty(), A, A, B)

    raise NotImplementedError(
-        "Fused set operations on values types other than QEnum are not yet implemented"
+        f"One of {type(A.values)} and {type(B.values)} should be WildCardGroup"
    )


-def operation(A: "Qube", B: "Qube", operation_type: SetOperation, node_type) -> "Qube":
+def shallow_set_operation(
+    A: ValuesIndices,
+    B: ValuesIndices,
+) -> SetOpResult:
+    if isinstance(A.values, QEnum) and isinstance(B.values, QEnum):
+        return shallow_qenum_set_operation(A, B)
+
+    # WildcardGroups behave as if they contain all possible values.
+    if isinstance(A.values, WildcardGroup) or isinstance(B.values, WildcardGroup):
+        return shallow_wildcard_set_operation(A, B)
+
+    raise NotImplementedError(
+        f"Set operations on values types {type(A.values)} and {type(B.values)} not yet implemented"
+    )
+
+
+def operation(
+    A: Qube, B: Qube, operation_type: SetOperation, node_type, depth=0
+) -> Qube | None:
+    # print(f"operation({A}, {B})")
    assert A.key == B.key, (
        "The two Qube root nodes must have the same key to perform set operations,"
        f"would usually be two root nodes. They have {A.key} and {B.key} respectively"
    )
+    node_key = A.key
+
+    assert A.is_root == B.is_root
+    is_root = A.is_root

    assert A.values == B.values, (
        f"The two Qube root nodes must have the same values to perform set operations {A.values = }, {B.values = }"
    )
+    node_values = A.values

    # Group the children of the two nodes by key
-    nodes_by_key = defaultdict(lambda: ([], []))
+    nodes_by_key: defaultdict[str, tuple[list[Qube], list[Qube]]] = defaultdict(
+        lambda: ([], [])
+    )
+    new_children: list[Qube] = []
+
+    # Sort out metadata into what can stay at this level and what must move down
+    stayput_metadata: dict[str, np.ndarray] = {}
+    pushdown_metadata_A: dict[str, np.ndarray] = {}
+    pushdown_metadata_B: dict[str, np.ndarray] = {}
+    for key in set(A.metadata.keys()) | set(B.metadata.keys()):
+        if key not in A.metadata:
+            pushdown_metadata_B[key] = B.metadata[key]
+            continue
+
+        if key not in B.metadata:
+            pushdown_metadata_A[key] = A.metadata[key]
+            continue
+
+        A_val = A.metadata[key]
+        B_val = B.metadata[key]
+        if np.allclose(A_val, B_val):
+            # print(f"{'  ' * depth}Keeping metadata key '{key}' at this level")
+            stayput_metadata[key] = A.metadata[key]
+        else:
+            # print(f"{'  ' * depth}Pushing down metadata key '{key}' {A_val} {B_val}")
+            pushdown_metadata_A[key] = A_val
+            pushdown_metadata_B[key] = B_val
+
+    # Add all the metadata that needs to be pushed down to the child nodes
+    # When pushing down the metadata we need to account for the fact it now affects more values
+    # So expand the metadata entries from shape (a, b, ..., c) to (a, b, ..., c, d)
+    # where d is the length of the node values
    for node in A.children:
+        N = len(node.values)
+        meta = {
+            k: np.broadcast_to(v[..., np.newaxis], v.shape + (N,))
+            for k, v in pushdown_metadata_A.items()
+        }
+        node = node.replace(metadata=node.metadata | meta)
        nodes_by_key[node.key][0].append(node)
+
    for node in B.children:
+        N = len(node.values)
+        meta = {
+            k: np.broadcast_to(v[..., np.newaxis], v.shape + (N,))
+            for k, v in pushdown_metadata_B.items()
+        }
+        node = node.replace(metadata=node.metadata | meta)
        nodes_by_key[node.key][1].append(node)

-    new_children = []
+    # print(f"{nodes_by_key = }")

    # For every node group, perform the set operation
    for key, (A_nodes, B_nodes) in nodes_by_key.items():
-        new_children.extend(
-            _operation(key, A_nodes, B_nodes, operation_type, node_type)
+        output = list(
+            _operation(A_nodes, B_nodes, operation_type, node_type, depth + 1)
        )
+        # print(f"{'  '*depth}_operation {operation_type.name} {A_nodes} {B_nodes} out = [{output}]")
+        new_children.extend(output)
+
+    # print(f"{'  '*depth}operation {operation_type.name} [{A}] [{B}] new_children = [{new_children}]")
+
+    # If there are now no children as a result of the operation, return nothing.
+    if (A.children or B.children) and not new_children:
+        if A.key == "root":
+            return node_type.make_root(children=())
+        else:
+            return None

    # Whenever we modify children we should recompress them
    # But since `operation` is already recursive, we only need to compress this level not all levels
    # Hence we use the non-recursive _compress method
-    new_children = compress_children(new_children)
+    new_children = list(compress_children(new_children))

    # The values and key are the same so we just replace the children
-    return replace(A, children=new_children)
+    if A.key == "root":
+        return node_type.make_root(
+            children=new_children,
+            metadata=stayput_metadata,
+        )
+    return node_type.make_node(
+        key=node_key,
+        values=node_values,
+        children=new_children,
+        metadata=stayput_metadata,
+        is_root=is_root,
+    )


-# The root node is special so we need a helper method that we can recurse on
 def _operation(
-    key: str, A: list["Qube"], B: list["Qube"], operation_type: SetOperation, node_type
-) -> Iterable["Qube"]:
-    # We need to deal with the case where only one of the trees has this key.
-    # To do so we can insert a dummy node with no children and no values into both A and B
-    keep_just_A, keep_intersection, keep_just_B = operation_type.value
+    A: list[Qube],
+    B: list[Qube],
+    operation_type: SetOperation,
+    node_type,
+    depth: int,
+) -> Iterable[Qube]:
+    """
+    This operation assumes that we've found two nodes that match and now want to do a set operation on their children. Hence we take in two lists of child nodes all of which have the same key but different values.
+    We then loop over all pairs of children from each list and compute the intersection.
+    """
+    # print(f"_operation({A}, {B})")
+    keep_only_A, keep_intersection, keep_only_B = operation_type.value

-    # Iterate over all pairs (node_A, node_B)
-    values = {}
-    for node in A + B:
-        values[node] = node.values
+    # We're going to progressively remove values from the starting nodes as we do intersections
+    # So we make a node -> ValuesIndices mapping here for both a and b
+    only_a: dict[Qube, ValuesIndices] = {
+        n: ValuesIndices.from_values(n.values) for n in A
+    }
+    only_b: dict[Qube, ValuesIndices] = {
+        n: ValuesIndices.from_values(n.values) for n in B
+    }

+    def make_new_node(source: Qube, values_indices: ValuesIndices):
+        return source.replace(
+            values=values_indices.values,
+            metadata=get_indices(source.metadata, values_indices.indices),
+        )
+
+    # Iterate over all pairs (node_A, node_B) and perform the shallow set operation
+    # Update our copy of the original node to remove anything that appears in an intersection
    for node_a in A:
        for node_b in B:
-            # Compute A - B, A & B, B - A
-            # Update the values for the two source nodes to remove the intersection
-            just_a, intersection, just_b = node_intersection(
-                values[node_a],
-                values[node_b],
-            )
+            set_ops_result = shallow_set_operation(only_a[node_a], only_b[node_b])

-            # Remove the intersection from the source nodes
-            values[node_a] = just_a
-            values[node_b] = just_b
+            # Save reduced values back to nodes
+            only_a[node_a] = set_ops_result.only_A
+            only_b[node_b] = set_ops_result.only_B

-            if keep_intersection:
-                if intersection:
-                    new_node_a = replace(
-                        node_a, data=replace(node_a.data, values=intersection)
-                    )
-                    new_node_b = replace(
-                        node_b, data=replace(node_b.data, values=intersection)
-                    )
-                    yield operation(new_node_a, new_node_b, operation_type, node_type)
+            if (
+                set_ops_result.intersection_A.values
+                and set_ops_result.intersection_B.values
+            ):
+                result = operation(
+                    make_new_node(node_a, set_ops_result.intersection_A),
+                    make_new_node(node_b, set_ops_result.intersection_B),
+                    operation_type,
+                    node_type,
+                    depth=depth + 1,
+                )
+                if result is not None:
+                    # If we're doing a difference or xor we might want to throw away the intersection
+                    # However we can only do this once we get to the leaf nodes, otherwise we'll
+                    # throw away nodes too early!
+                    # Consider Qube(root, a=1, b=1/2) - Qube(root, a=1, b=1)
+                    # We can easily throw away the whole a node by accident here!
+                    if keep_intersection or result.children:
+                        yield result
+            elif (
+                not set_ops_result.intersection_A.values
+                and not set_ops_result.intersection_B.values
+            ):
+                continue
+            else:
+                raise ValueError(
+                    f"Only one of set_ops_result.intersection_A and set_ops_result.intersection_B is None, I didn't think that could happen! {set_ops_result = }"
+                )

-    # Now we've removed all the intersections we can yield the just_A and just_B parts if needed
-    if keep_just_A:
-        for node in A:
-            if values[node]:
-                yield node_type.make(key, values[node], node.children)
-    if keep_just_B:
-        for node in B:
-            if values[node]:
-                yield node_type.make(key, values[node], node.children)
+    if keep_only_A:
+        for node, vi in only_a.items():
+            if vi.values:
+                yield make_new_node(node, vi)
+
+    if keep_only_B:
+        for node, vi in only_b.items():
+            if vi.values:
+                yield make_new_node(node, vi)


-def compress_children(children: Iterable["Qube"]) -> tuple["Qube"]:
+def compress_children(children: Iterable[Qube], depth=0) -> tuple[Qube, ...]:
    """
    Helper method tht only compresses a set of nodes, and doesn't do it recursively.
    Used in Qubed.compress but also to maintain compression in the set operations above.
    """
-    # Now take the set of new children and see if any have identical key, metadata and children
+    # Take the set of new children and see if any have identical key, metadata and children
    # the values may different and will be collapsed into a single node
-    identical_children = defaultdict(set)
+
+    identical_children = defaultdict(list)
    for child in children:
        # only care about the key and children of each node, ignore values
-        key = hash((child.key, tuple((cc.structural_hash for cc in child.children))))
-        identical_children[key].add(child)
+        h = hash((child.key, tuple((cc.structural_hash for cc in child.children))))
+        identical_children[h].append(child)

    # Now go through and create new compressed nodes for any groups that need collapsing
    new_children = []
-    for child_set in identical_children.values():
-        if len(child_set) > 1:
-            child_set = list(child_set)
-            node_type = type(child_set[0])
-            key = child_set[0].key
+    for child_list in identical_children.values():
+        # If the group is size one just keep it
+        if len(child_list) == 1:
+            new_child = child_list.pop()

-            # Compress the children into a single node
-            assert all(isinstance(child.data.values, QEnum) for child in child_set), (
-                "All children must have QEnum values"
-            )
-
-            node_data = NodeData(
-                key=key,
-                metadata=frozendict(),  # Todo: Implement metadata compression
-                values=QEnum(
-                    (v for child in child_set for v in child.data.values.values)
-                ),
-            )
-            new_child = node_type(data=node_data, children=child_set[0].children)
        else:
-            # If the group is size one just keep it
-            new_child = child_set.pop()
+            example = child_list[0]
+            node_type = type(example)
+            value_type = type(example.values)
+
+            assert all(isinstance(child.values, value_type) for child in child_list), (
+                f"All nodes to be grouped must have the same value type, expected {value_type}"
+            )
+
+            # We know the children of this group of nodes all have the same structure
+            # but we still need to merge the metadata across them
+            # children = example.children
+            children = merge_metadata(child_list, example.depth)
+
+            # Do we need to recusively compress here?
+            # children = compress_children(children, depth=depth+1)
+
+            if value_type is QEnum:
+                values = QEnum(set(v for child in child_list for v in child.values))
+            elif value_type is WildcardGroup:
+                values = example.values
+            else:
+                raise ValueError(f"Unknown value type: {value_type}")
+
+            new_child = node_type.make_node(
+                key=example.key,
+                metadata=example.metadata,
+                values=values,
+                children=children,
+            )

        new_children.append(new_child)
-    return tuple(
-        sorted(new_children, key=lambda n: ((n.key, tuple(sorted(n.values.values)))))
-    )
+
+    return tuple(sorted(new_children, key=lambda n: ((n.key, n.values.min()))))
+
+
+def merge_metadata(qubes: list[Qube], axis) -> Iterable[Qube]:
+    """
+    Given a list of qubes with identical structure,
+    match up the children of each node and merge the metadata
+    """
+    # Group the children of each qube and merge them
+    # Exploit the fact that they have the same shape and ordering
+    example = qubes[0]
+    node_type = type(example)
+
+    for i in range(len(example.children)):
+        group = [q.children[i] for q in qubes]
+        group_example = group[0]
+        assert len(set((c.structural_hash for c in group))) == 1
+
+        # Collect metadata by key
+        metadata_groups = {
+            k: [q.metadata[k] for q in group] for k in group_example.metadata.keys()
+        }
+
+        # Concatenate the metadata together
+        metadata: frozendict[str, np.ndarray] = frozendict(
+            {
+                k: np.concatenate(metadata_group, axis=axis)
+                for k, metadata_group in metadata_groups.items()
+            }
+        )
+
+        group_children = merge_metadata(group, axis)
+        yield node_type.make_node(
+            key=group_example.key,
+            metadata=metadata,
+            values=group_example.values,
+            children=group_children,
+        )
--- a/src/python/qubed/tree_formatters.py
+++ b/src/python/qubed/tree_formatters.py
@ -1,16 +1,16 @@
+from __future__ import annotations
+
 import random
 from dataclasses import dataclass
-from typing import Iterable, Protocol, Sequence, runtime_checkable
+from typing import TYPE_CHECKING, Callable, Iterable

+try:
+    from IPython.display import display
+except ImportError:
+    display = None

-@runtime_checkable
-class TreeLike(Protocol):
-    @property
-    def children(
-        self,
-    ) -> Sequence["TreeLike"]: ...  # Supports indexing like node.children[i]
-
-    def summary(self) -> str: ...
+if TYPE_CHECKING:
+    from .Qube import Qube


@dataclass(frozen=True)
@ -22,8 +22,8 @@ class HTML:


 def summarize_node(
-    node: TreeLike, collapse=False, **kwargs
-) -> tuple[str, str, TreeLike]:
+    node: Qube, collapse=False, max_summary_length=50, **kwargs
+) -> tuple[str, str, Qube]:
    """
    Extracts a summarized representation of the node while collapsing single-child paths.
    Returns the summary string and the last node in the chain that has multiple children.
@ -33,9 +33,10 @@ def summarize_node(

    while True:
        summary = node.summary(**kwargs)
+
        paths.append(summary)
-        if len(summary) > 50:
-            summary = summary[:50] + "..."
+        if len(summary) > max_summary_length:
+            summary = summary[:max_summary_length] + "..."
        summaries.append(summary)
        if not collapse:
            break
@ -45,10 +46,14 @@ def summarize_node(
            break
        node = node.children[0]

+    # Add a "..." to represent nodes that we don't know about
+    if (not node.children) and (not node.is_leaf):
+        summaries.append("...")
+
    return ", ".join(summaries), ",".join(paths), node


-def node_tree_to_string(node: TreeLike, prefix: str = "", depth=None) -> Iterable[str]:
+def node_tree_to_string(node: Qube, prefix: str = "", depth=None) -> Iterable[str]:
    summary, path, node = summarize_node(node)

    if depth is not None and depth <= 0:
@ -71,17 +76,73 @@ def node_tree_to_string(node: TreeLike, prefix: str = "", depth=None) -> Iterabl
        )


+def summarize_node_html(
+    node: Qube,
+    collapse=False,
+    max_summary_length=50,
+    info: Callable[[Qube], str] | None = None,
+    **kwargs,
+) -> tuple[str, Qube]:
+    """
+    Extracts a summarized representation of the node while collapsing single-child paths.
+    Returns the summary string and the last node in the chain that has multiple children.
+    """
+    if info is None:
+
+        def info_func(node: Qube, /):
+            return (
+                # f"dtype: {node.dtype}\n"
+                f"metadata: {dict(node.metadata)}\n"
+            )
+    else:
+        info_func = info
+
+    summaries = []
+
+    while True:
+        path = node.summary(**kwargs)
+        summary = path
+
+        if len(summary) > max_summary_length:
+            summary = summary[:max_summary_length] + "..."
+
+        info_string = info_func(node)
+
+        summary = f'<span class="qubed-node" data-path="{path}" title="{info_string}">{summary}</span>'
+        summaries.append(summary)
+        if not collapse:
+            break
+
+        # Move down if there's exactly one child, otherwise stop
+        if len(node.children) != 1:
+            break
+        node = node.children[0]
+
+    if (not node.children) and (not node.is_leaf):
+        summary = (
+            '<span class="qubed-node" data-path="" title="Truncated Nodes">...</span>'
+        )
+        summaries.append(summary)
+
+    return ", ".join(summaries), node
+
+
 def _node_tree_to_html(
-    node: TreeLike, prefix: str = "", depth=1, connector="", **kwargs
+    node: Qube,
+    prefix: str = "",
+    depth=1,
+    connector="",
+    info: Callable[[Qube], str] | None = None,
+    **kwargs,
 ) -> Iterable[str]:
-    summary, path, node = summarize_node(node, **kwargs)
+    summary, node = summarize_node_html(node, info=info, **kwargs)

    if len(node.children) == 0:
-        yield f'<span class="qubed-node leaf" data-path="{path}">{connector}{summary}</span>'
+        yield f'<span class="qubed-level">{connector}{summary}</span>'
        return
    else:
        open = "open" if depth > 0 else ""
-        yield f'<details {open} data-path="{path}"><summary class="qubed-node">{connector}{summary}</summary>'
+        yield f'<details {open}><summary class="qubed-level">{connector}{summary}</summary>'

    for index, child in enumerate(node.children):
        connector = "└── " if index == len(node.children) - 1 else "├── "
@ -91,13 +152,23 @@ def _node_tree_to_html(
            prefix + extension,
            depth=depth - 1,
            connector=prefix + connector,
+            info=info,
            **kwargs,
        )
    yield "</details>"


-def node_tree_to_html(node: TreeLike, depth=1, **kwargs) -> str:
-    css_id = f"qubed-tree-{random.randint(0, 1000000)}"
+def node_tree_to_html(
+    node: Qube,
+    depth=1,
+    include_css=True,
+    include_js=True,
+    css_id=None,
+    info: Callable[[Qube], str] | None = None,
+    **kwargs,
+) -> str:
+    if css_id is None:
+        css_id = f"qubed-tree-{random.randint(0, 1000000)}"

    # It's ugle to use an f string here because css uses {} so much so instead
    # we use CSS_ID as a placeholder and replace it later
@ -114,7 +185,7 @@ def node_tree_to_html(node: TreeLike, depth=1, **kwargs) -> str:
                margin-left: 0;
            }

-            .qubed-node a {
+            .qubed-level a {
                margin-left: 10px;
                text-decoration: none;
            }
@ -128,7 +199,7 @@ def node_tree_to_html(node: TreeLike, depth=1, **kwargs) -> str:
                display: block;
            }

-            summary:hover,span.leaf:hover {
+            span.qubed-node:hover {
                background-color: #f0f0f0;
            }

@ -140,7 +211,7 @@ def node_tree_to_html(node: TreeLike, depth=1, **kwargs) -> str:
                content: " ▼";
            }

-            .leaf {
+            .qubed-level {
                text-overflow: ellipsis;
                overflow: hidden;
                text-wrap: nowrap;
@ -175,9 +246,26 @@ def node_tree_to_html(node: TreeLike, depth=1, **kwargs) -> str:
            await navigator.clipboard.writeText(path);
        }

-        const nodes = document.querySelectorAll("#CSS_ID .qubed-node");
+        const nodes = document.querySelectorAll("#CSS_ID.qubed-node");
        nodes.forEach(n => n.addEventListener("click", nodeOnClick));
        </script>
        """.replace("CSS_ID", css_id)
-    nodes = "".join(_node_tree_to_html(node=node, depth=depth, **kwargs))
-    return f"{js}{css}<pre class='qubed-tree' id='{css_id}'>{nodes}</pre>"
+    nodes = "".join(_node_tree_to_html(node=node, depth=depth, info=info, **kwargs))
+    return f"{js if include_js else ''}{css if include_css else ''}<pre class='qubed-tree' id='{css_id}'>{nodes}</pre>"
+
+
+def _display(qube: Qube, **kwargs):
+    if display is None:
+        print(qube)
+    else:
+
+        def info(node: Qube):
+            return f"""\
+structural_hash = {node.structural_hash}
+metadata = {dict(node.metadata)}
+is_root = {node.is_root}
+is_leaf = {node.is_leaf}
+"""
+
+        kwargs = {"info": info} | kwargs
+        display(qube.html(**kwargs))
--- a/src/python/qubed/value_types.py
+++ b/src/python/qubed/value_types.py
@ -1,8 +1,19 @@
+from __future__ import annotations
+
 import dataclasses
 from abc import ABC, abstractmethod
-from dataclasses import dataclass, replace
+from dataclasses import dataclass
 from datetime import date, datetime, timedelta
-from typing import TYPE_CHECKING, Any, FrozenSet, Iterable, Literal, TypeVar
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    FrozenSet,
+    Iterable,
+    Iterator,
+    Literal,
+    Sequence,
+    TypeVar,
+)

 if TYPE_CHECKING:
    from .Qube import Qube
@ -11,13 +22,13 @@ if TYPE_CHECKING:
@dataclass(frozen=True)
 class ValueGroup(ABC):
    @abstractmethod
-    def summary(self) -> str:
-        "Provide a string summary of the value group."
+    def dtype(self) -> str:
+        "Provide a string rep of the datatype of these values"
        pass

    @abstractmethod
-    def __len__(self) -> int:
-        "Return how many values this group contains."
+    def summary(self) -> str:
+        "Provide a string summary of the value group."
        pass

    @abstractmethod
@ -26,14 +37,8 @@ class ValueGroup(ABC):
        pass

    @abstractmethod
-    def __iter__(self) -> Iterable[Any]:
-        "Iterate over the values in the group."
-        pass
-
-    @classmethod
-    @abstractmethod
-    def from_strings(cls, values: Iterable[str]) -> list["ValueGroup"]:
-        "Given a list of strings, return a one or more ValueGroups of this type."
+    def to_json(self) -> dict:
+        "Return a JSON serializable representation of the value group."
        pass

    @abstractmethod
@ -41,15 +46,39 @@ class ValueGroup(ABC):
        "Return the minimum value in the group."
        pass

+    @classmethod
    @abstractmethod
-    def to_json(self) -> dict:
-        "Return a JSON serializable representation of the value group."
+    def from_strings(cls, values: Iterable[str]) -> Sequence[ValueGroup]:
+        "Given a list of strings, return a one or more ValueGroups of this type."
+        pass
+
+    @abstractmethod
+    def __iter__(self) -> Iterator:
+        "Iterate over the values in the group."
+        pass
+
+    @abstractmethod
+    def __len__(self) -> int:
        pass


 T = TypeVar("T")
 EnumValuesType = FrozenSet[T]

+_dtype_map: dict[str, type] = {
+    "str": str,
+    "int64": int,
+    "float64": float,
+    "date": datetime,
+}
+_dtype_map_inv: dict[type, str] = {v: k for k, v in _dtype_map.items()}
+_dtype_formatters = {
+    "str": str,
+    "int64": int,
+    "float64": float,
+    "date": datetime.fromisoformat,
+}
+

@dataclass(frozen=True, order=True)
 class QEnum(ValueGroup):
@ -59,9 +88,15 @@ class QEnum(ValueGroup):
    """

    values: EnumValuesType
+    _dtype: str = "str"

-    def __init__(self, obj):
-        object.__setattr__(self, "values", frozenset(obj))
+    def __init__(self, obj, dtype="str"):
+        object.__setattr__(self, "values", tuple(sorted(obj)))
+        object.__setattr__(
+            self,
+            "_dtype",
+            dtype,
+        )

    def __post_init__(self):
        assert isinstance(self.values, tuple)
@ -78,14 +113,60 @@ class QEnum(ValueGroup):
    def __contains__(self, value: Any) -> bool:
        return value in self.values

-    def from_strings(self, values: Iterable[str]) -> list["ValueGroup"]:
-        return [type(self)(tuple(values))]
+    def dtype(self):
+        return self._dtype
+
+    @classmethod
+    def from_strings(cls, values: Iterable[str]) -> Sequence[ValueGroup]:
+        return [cls(tuple(values))]

    def min(self):
        return min(self.values)

    def to_json(self):
-        return list(self.values)
+        return {"type": "enum", "dtype": self.dtype(), "values": self.values}
+
+    # @classmethod
+    # def from_json(cls, type: Literal["enum"], dtype: str, values: list):
+    #     dtype_formatter = _dtype_formatters[dtype]
+
+    @classmethod
+    def from_list(cls, obj):
+        example = obj[0]
+        dtype = type(example)
+        assert [type(v) is dtype for v in obj]
+        return cls(obj, dtype=_dtype_map_inv[dtype])
+
+
+@dataclass(frozen=True, order=True)
+class WildcardGroup(ValueGroup):
+    def summary(self) -> str:
+        return "*"
+
+    def __contains__(self, value: Any) -> bool:
+        return True
+
+    def to_json(self):
+        return "*"
+
+    def min(self):
+        return "*"
+
+    def __len__(self):
+        return 1
+
+    def __iter__(self):
+        return ["*"]
+
+    def __bool__(self):
+        return True
+
+    def dtype(self):
+        return "*"
+
+    @classmethod
+    def from_strings(cls, values: Iterable[str]) -> Sequence[ValueGroup]:
+        return [WildcardGroup()]


 class DateEnum(QEnum):
@ -95,37 +176,6 @@ class DateEnum(QEnum):

        return "/".join(map(fmt, sorted(self.values)))

-@dataclass(frozen=True)
-class WildCard(ValueGroup):
-    def summary(self) -> str:
-        "Provide a string summary of the value group."
-        return "Includes all options"
-
-    def __len__(self) -> int:
-        "Return how many values this group contains."
-        return -1
-
-    def __contains__(self, value: Any) -> bool:
-        "Given a value, coerce to the value type and determine if it is in the value group."
-        return True
-
-    @abstractmethod
-    def __iter__(self) -> Iterable[Any]:
-        "Iterate over the values in the group."
-        pass
-
-    @classmethod
-    def from_strings(cls, values: Iterable[str]) -> list["ValueGroup"]:
-        "Given a list of strings, return a one or more ValueGroups of this type."
-        return [cls()]
-
-    def min(self):
-        "Return the minimum value in the group."
-        return -1
-
-    def to_json(self) -> dict:
-        "Return a JSON serializable representation of the value group."
-        return dataclasses.asdict(self)

@dataclass(frozen=True)
 class Range(ValueGroup, ABC):
@ -138,7 +188,7 @@ class Range(ValueGroup, ABC):
    def min(self):
        return self.start

-    def __iter__(self) -> Iterable[Any]:
+    def __iter__(self) -> Iterator[Any]:
        i = self.start
        while i <= self.end:
            yield i
@ -158,19 +208,19 @@ class DateRange(Range):
    def __len__(self) -> int:
        return (self.end - self.start) // self.step

-    def __iter__(self) -> Iterable[date]:
+    def __iter__(self) -> Iterator[date]:
        current = self.start
        while current <= self.end if self.step.days > 0 else current >= self.end:
            yield current
            current += self.step

    @classmethod
-    def from_strings(cls, values: Iterable[str]) -> "list[DateRange | QEnum]":
+    def from_strings(cls, values: Iterable[str]) -> Sequence[DateRange | DateEnum]:
        dates = sorted([datetime.strptime(v, "%Y%m%d") for v in values])
        if len(dates) < 2:
            return [DateEnum(dates)]

-        ranges = []
+        ranges: list[DateEnum | DateRange] = []
        current_group, dates = (
            [
                dates[0],
@ -256,7 +306,7 @@ class TimeRange(Range):
    def min(self):
        return self.start

-    def __iter__(self) -> Iterable[Any]:
+    def __iter__(self) -> Iterator[Any]:
        return super().__iter__()

    @classmethod
@ -366,23 +416,19 @@ class IntRange(Range):
        return ranges


-def values_from_json(obj) -> ValueGroup:
+def values_from_json(obj: dict | list) -> ValueGroup:
    if isinstance(obj, list):
-        return QEnum(tuple(obj))
+        return QEnum.from_list(obj)

-    match obj["dtype"]:
-        case "date":
-            return DateRange(**obj)
-        case "time":
-            return TimeRange(**obj)
-        case "int":
-            return IntRange(**obj)
+    match obj["type"]:
+        case "enum":
+            QEnum.from_json(**obj)
        case _:
            raise ValueError(f"Unknown dtype {obj['dtype']}")


 def convert_datatypes(q: "Qube", conversions: dict[str, ValueGroup]) -> "Qube":
-    def _convert(q: "Qube") -> Iterable["Qube"]:
+    def _convert(q: "Qube") -> Iterator["Qube"]:
        if q.key in conversions:
            data_type = conversions[q.key]
            assert isinstance(q.values, QEnum), (
@ -390,7 +436,7 @@ def convert_datatypes(q: "Qube", conversions: dict[str, ValueGroup]) -> "Qube":
            )
            for values_group in data_type.from_strings(q.values):
                # print(values_group)
-                yield replace(q, data=replace(q.data, values=values_group))
+                yield q.replace(values=values_group)
        else:
            yield q

--- a/src/qube.proto
+++ b/src/qube.proto
@ -0,0 +1,32 @@
+syntax = "proto3";
+
+message NdArray {
+  repeated int64 shape = 1;
+  string  dtype  = 2;
+  bytes   raw    = 3;
+}
+
+message StringGroup {repeated string items = 1; }
+
+// Stores values i.e class=1/2/3 the 1/2/3 part
+message ValueGroup {
+    oneof payload {
+    StringGroup s = 1;
+    NdArray tensor = 2;
+  }
+}
+
+message MetadataGroup {
+    oneof payload {
+    NdArray tensor = 1;
+  }
+}
+
+message Qube {
+  string key          = 1;
+  ValueGroup values   = 2;
+  map<string, MetadataGroup> metadata = 3;
+  string dtype        = 4;
+  repeated Qube children = 5;
+  bool is_root        = 6;
+}
--- a/src/rust/compressed_tree.rs
+++ b/src/rust/compressed_tree.rs
@ -1,334 +0,0 @@
-#![allow(dead_code)]
-
-use std::rc::Rc;
-use smallstr::SmallString;
-
-use slotmap::{new_key_type, SlotMap};
-
-new_key_type! {
-    struct NodeId;
-}
-
-type CompactString = SmallString<[u8; 16]>;
-
-#[derive(Clone)]
-enum NodeValueTypes {
-    String(CompactString),
-    Int(i32),
-}
-
-impl From<&str> for NodeValueTypes {
-    fn from(s: &str) -> Self {
-        NodeValueTypes::String(CompactString::from(s))
-    }
-}
-
-impl From<i32> for NodeValueTypes {
-    fn from(i: i32) -> Self {
-        NodeValueTypes::Int(i)
-    }
-}
-
-enum NodeValue {
-    Single(NodeValueTypes),
-    Multiple(Vec<NodeValueTypes>),
-}
-
-struct Node<Payload> {
-    key: Rc<String>,
-    value: NodeValue,
-    parent: Option<NodeId>,
-    prev_sibling: Option<NodeId>,
-    next_sibling: Option<NodeId>,
-    // vector may be faster for traversal, but linkedlist should be faster for insertion
-    children: Option<(NodeId, NodeId)>, // (first_child, last_child)
-    data: Option<Payload>,
-}
-
-struct QueryTree<Payload> {
-    nodes: SlotMap<NodeId, Node<Payload>>,
-}
-
-impl<Payload> QueryTree<Payload> {
-    fn new() -> Self {
-        QueryTree {
-            nodes: SlotMap::with_key(),
-        }
-    }
-
-    // Adds a node with a key and single value
-    fn add_node<S>(&mut self, key: &Rc<String>, value: S, parent: Option<NodeId>) -> NodeId
-    where
-        S: Into<NodeValueTypes>,
-    {
-        let node_id = self.nodes.insert_with_key(|_| Node {
-            key: Rc::clone(key),
-            value: NodeValue::Single(value.into()),
-            parent,
-            prev_sibling: None,
-            next_sibling: None,
-            children: None,
-            data: None,
-        });
-
-        if let Some(parent_id) = parent {
-            // Determine if parent has existing children
-            if let Some((first_child_id, last_child_id)) = self.nodes[parent_id].children {
-                // Update the last child's `next_sibling`
-                {
-                    let last_child = &mut self.nodes[last_child_id];
-                    last_child.next_sibling = Some(node_id);
-                }
-
-                // Update the new node's `prev_sibling`
-                {
-                    let new_node = &mut self.nodes[node_id];
-                    new_node.prev_sibling = Some(last_child_id);
-                }
-
-                // Update parent's last child
-                let parent_node = &mut self.nodes[parent_id];
-                parent_node.children = Some((first_child_id, node_id));
-            } else {
-                // No existing children
-                let parent_node = &mut self.nodes[parent_id];
-                parent_node.children = Some((node_id, node_id));
-            }
-        }
-
-        node_id
-    }
-
-    // Add a single value to a node
-    fn add_value<S>(&mut self, node_id: NodeId, value: S)
-    where
-        S: Into<NodeValueTypes>,
-    {
-        if let Some(node) = self.nodes.get_mut(node_id) {
-            match &mut node.value {
-                NodeValue::Single(v) => {
-                    let values = vec![v.clone(), value.into()];
-                    node.value = NodeValue::Multiple(values);
-                }
-                NodeValue::Multiple(values) => {
-                    values.push(value.into());
-                }
-            }
-        }
-    }
-
-    // Add multiple values to a node
-    fn add_values<S>(&mut self, node_id: NodeId, values: Vec<S>)
-    where
-        S: Into<NodeValueTypes>,
-    {
-        if let Some(node) = self.nodes.get_mut(node_id) {
-            match &mut node.value {
-                NodeValue::Single(v) => {
-                    let mut new_values = vec![v.clone()];
-                    new_values.extend(values.into_iter().map(|v| v.into()));
-                    node.value = NodeValue::Multiple(new_values);
-                }
-                NodeValue::Multiple(existing_values) => {
-                    existing_values.extend(values.into_iter().map(|v| v.into()));
-                }
-            }
-        }
-    }
-
-    fn get_node(&self, node_id: NodeId) -> Option<&Node<Payload>> {
-        self.nodes.get(node_id)
-    }
-
-    // TODO: better if this returns an iterator?
-    fn get_children(&self, node_id: NodeId) -> Vec<NodeId> {
-        let mut children = Vec::new();
-
-        if let Some(node) = self.get_node(node_id) {
-            if let Some((first_child_id, _)) = node.children {
-                let mut current_id = Some(first_child_id);
-                while let Some(cid) = current_id {
-                    children.push(cid);
-                    current_id = self.nodes[cid].next_sibling;
-                }
-            }
-        }
-
-        children
-    }
-
-    fn remove_node(&mut self, node_id: NodeId) {
-        // Remove the node and update parent and siblings
-        if let Some(node) = self.nodes.remove(node_id) {
-            // Update parent's children
-            if let Some(parent_id) = node.parent {
-                let parent_node = self.nodes.get_mut(parent_id).unwrap();
-                if let Some((first_child_id, last_child_id)) = parent_node.children {
-                    if first_child_id == node_id && last_child_id == node_id {
-                        // Node was the only child
-                        parent_node.children = None;
-                    } else if first_child_id == node_id {
-                        // Node was the first child
-                        parent_node.children = Some((node.next_sibling.unwrap(), last_child_id));
-                    } else if last_child_id == node_id {
-                        // Node was the last child
-                        parent_node.children = Some((first_child_id, node.prev_sibling.unwrap()));
-                    }
-                }
-            }
-
-            // Update siblings
-            if let Some(prev_id) = node.prev_sibling {
-                self.nodes[prev_id].next_sibling = node.next_sibling;
-            }
-            if let Some(next_id) = node.next_sibling {
-                self.nodes[next_id].prev_sibling = node.prev_sibling;
-            }
-
-            // Recursively remove children
-            let children_ids = self.get_children(node_id);
-            for child_id in children_ids {
-                self.remove_node(child_id);
-            }
-        }
-    }
-
-    fn is_root(&self, node_id: NodeId) -> bool {
-        self.nodes[node_id].parent.is_none()
-    }
-
-    fn is_leaf(&self, node_id: NodeId) -> bool {
-        self.nodes[node_id].children.is_none()
-    }
-
-    fn add_payload(&mut self, node_id: NodeId, payload: Payload) {
-        if let Some(node) = self.nodes.get_mut(node_id) {
-            node.data = Some(payload);
-        }
-    }
-
-    fn print_tree(&self) {
-        // Find all root nodes (nodes without a parent)
-        let roots: Vec<NodeId> = self
-            .nodes
-            .iter()
-            .filter_map(|(id, node)| {
-                if node.parent.is_none() {
-                    Some(id)
-                } else {
-                    None
-                }
-            })
-            .collect();
-
-        // Iterate through each root node and print its subtree
-        for (i, root_id) in roots.iter().enumerate() {
-            let is_last = i == roots.len() - 1;
-            self.print_node(*root_id, String::new(), is_last);
-        }
-    }
-
-    /// Recursively prints a node and its children.
-    ///
-    /// - `node_id`: The current node's ID.
-    /// - `prefix`: The string prefix for indentation and branch lines.
-    /// - `is_last`: Boolean indicating if the node is the last child of its parent.
-    fn print_node(&self, node_id: NodeId, prefix: String, is_last: bool) {
-        // Retrieve the current node
-        let node = match self.nodes.get(node_id) {
-            Some(n) => n,
-            None => return, // Node not found; skip
-        };
-
-        // Determine the branch character
-        let branch = if prefix.is_empty() {
-            "" // Root node doesn't have a branch
-        } else if is_last {
-            "└── " // Last child
-        } else {
-            "├── " // Middle child
-        };
-
-        // Print the current node's key and values
-        print!("{}{}{}", prefix, branch, node.key);
-        match &node.value {
-            NodeValue::Single(v) => match v {
-                NodeValueTypes::String(s) => println!(": ({})", s),
-                NodeValueTypes::Int(i) => println!(": ({})", i),
-            },
-            NodeValue::Multiple(vs) => {
-                let values: Vec<String> = vs
-                    .iter()
-                    .map(|v| match v {
-                        NodeValueTypes::String(s) => s.to_string(),
-                        NodeValueTypes::Int(i) => i.to_string(),
-                    })
-                    .collect();
-                println!(": ({})", values.join(", "));
-            }
-        }
-
-        // Prepare the prefix for child nodes
-        let new_prefix = if prefix.is_empty() {
-            if is_last {
-                "    ".to_string()
-            } else {
-                "│   ".to_string()
-            }
-        } else {
-            if is_last {
-                format!("{}    ", prefix)
-            } else {
-                format!("{}│   ", prefix)
-            }
-        };
-
-        // Retrieve and iterate through child nodes
-        if let Some((_first_child_id, _last_child_id)) = node.children {
-            let children = self.get_children(node_id);
-            let total = children.len();
-            for (i, child_id) in children.iter().enumerate() {
-                let child_is_last = i == total - 1;
-                self.print_node(*child_id, new_prefix.clone(), child_is_last);
-            }
-        }
-    }
-}
-
-fn main() {
-    let mut tree: QueryTree<i16> = QueryTree::new();
-
-    let value = "hello";
-    let axis = Rc::new("foo".to_string());
-
-    let root_id = tree.add_node(&axis, value, None);
-
-    use std::time::Instant;
-    let now = Instant::now();
-
-    for _ in 0..100 {
-        // let child_value = format!("child_val{}", i);
-        let child_id = tree.add_node(&axis, value, Some(root_id));
-        // tree.add_value(child_id, value);
-
-        for _ in 0..100 {
-            // let gchild_value = format!("gchild_val{}", j);
-            let gchild_id = tree.add_node(&axis, value, Some(child_id));
-            // tree.add_values(gchild_id, vec![1, 2]);
-
-            for _ in 0..1000 {
-                // let ggchild_value = format!("ggchild_val{}", k);
-                let _ggchild_id = tree.add_node(&axis, value, Some(gchild_id));
-                // tree.add_value(_ggchild_id, value);
-                // tree.add_values(_ggchild_id, vec![1, 2, 3, 4]);
-            }
-        }
-    }
-
-    assert_eq!(tree.nodes.len(), 10_010_101);
-
-    let elapsed = now.elapsed();
-    println!("Elapsed: {:.2?}", elapsed);
-
-    // tree.print_tree();
-}
--- a/src/rust/connectors/fdb.rs
+++ b/src/rust/connectors/fdb.rs
@ -0,0 +1,76 @@
+use rsfdb::listiterator::KeyValueLevel;
+use rsfdb::request::Request;
+use rsfdb::FDB;
+
+use serde_json::{json, Value};
+use std::time::Instant;
+
+
+use std::collections::HashMap;
+
+pub mod tree;
+use std::sync::Arc;
+use std::sync::Mutex;
+use tree::TreeNode;
+
+#[pyclass(unsendable)]
+pub struct PyFDB {
+    pub fdb: FDB,
+}
+
+#[pymethods]
+impl PyFDB {
+    #[new]
+    #[pyo3(signature = (fdb_config=None))]
+    pub fn new(fdb_config: Option<&str>) -> PyResult<Self> {
+        let fdb = FDB::new(fdb_config)
+            .map_err(|e| PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(e.to_string()))?;
+        Ok(PyFDB { fdb })
+    }
+
+    /// Traverse the FDB with the given request.
+    pub fn traverse_fdb(
+        &self,
+        py: Python<'_>,
+        request: HashMap<String, Vec<String>>,
+    ) -> PyResult<PyObject> {
+        let start_time = Instant::now();
+
+        let list_request = Request::from_json(json!(request))
+            .map_err(|e| PyErr::new::<pyo3::exceptions::PyValueError, _>(e.to_string()))?;
+
+        // Use `fdb_guard` instead of `self.fdb`
+        let list = self
+            .fdb
+            .list(&list_request, true, true)
+            .map_err(|e| PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(e.to_string()))?;
+
+        let mut root = TreeNode::new(KeyValueLevel {
+            key: "root".to_string(),
+            value: "root".to_string(),
+            level: 0,
+        });
+
+        for item in list {
+            py.check_signals()?;
+
+            if let Some(request) = &item.request {
+                root.insert(&request);
+            }
+        }
+
+        let duration = start_time.elapsed();
+        println!("Total runtime: {:?}", duration);
+
+        let py_dict = root.to_py_dict(py)?;
+        Ok(py_dict)
+    }
+}
+
+use pyo3::prelude::*;
+
+#[pymodule]
+fn rust(m: &Bound<'_, PyModule>) -> PyResult<()> {
+    m.add_class::<PyFDB>()?;
+    Ok(())
+}
--- a/src/rust/formatters/mod.rs
+++ b/src/rust/formatters/mod.rs
@ -0,0 +1,147 @@
+use crate::{Node, NodeId, Qube};
+use itertools::Itertools;
+use itertools::Position;
+
+impl Node {
+    /// Generate a human readable summary of the node
+    /// Examples include: key=value1/value2/.../valueN, key=value1/to/value1, key=*, root etc
+    pub fn summary(&self, qube: &Qube) -> String {
+        if self.is_root() {
+            return "root".to_string();
+        }
+        let key = &qube[self.key];
+        let values: String =
+            Itertools::intersperse(self.values.iter().map(|id| &qube[*id]), "/").collect();
+
+        format!("{}={}", key, values)
+    }
+
+    pub fn html_summary(&self, qube: &Qube) -> String {
+        if self.is_root() {
+            return r#"<span class="qubed-node">root</span>"#.to_string();
+        }
+        let key = &qube[self.key];
+        let values: String =
+            Itertools::intersperse(self.values.iter().map(|id| &qube[*id]), "/").collect();
+
+        let summary = format!("{}={}", key, values);
+        let path = summary.clone();
+        let info = format!("is_root: {}", self.is_root());
+        format!(r#"<span class="qubed-node" data-path="{path}" title="{info}">{summary}</span>"#)
+    }
+}
+
+struct NodeSummary {
+    summary: String,
+    end: NodeId,
+}
+
+enum SummaryType {
+    PlainText,
+    HTML,
+}
+
+/// Given a Node, traverse the tree until a node has more than one child.
+/// Returns a summary of the form "key1=v1/v2, key2=v1/v2/v3, key3=v1"
+/// and the id of the last node in the summary
+fn summarise_nodes(qube: &Qube, node_id: &NodeId, summary_type: SummaryType) -> NodeSummary {
+    let mut node_id = *node_id;
+    let mut summary_vec = vec![];
+    loop {
+        let node = &qube[node_id];
+        let summary = match summary_type {
+            SummaryType::PlainText => node.summary(&qube),
+            SummaryType::HTML => node.html_summary(&qube),
+        };
+        summary_vec.push(summary);
+
+        // Bail out if the node has anothing other than 1 child.
+        match node.has_exactly_one_child() {
+            Some(n) => node_id = n,
+            None => break,
+        };
+    }
+    NodeSummary {
+        summary: summary_vec.join(", "),
+        end: node_id,
+    }
+}
+
+fn qube_to_tree(qube: &Qube, node_id: &NodeId, prefix: &str, depth: usize) -> String {
+    let NodeSummary {
+        summary,
+        end: node_id,
+    } = summarise_nodes(qube, node_id, SummaryType::PlainText);
+
+    let mut output: Vec<String> = Vec::new();
+
+    if depth <= 0 {
+        return format!("{} - ...\n", summary);
+    } else {
+        output.push(format!("{}\n", summary));
+    }
+
+    let node = &qube[node_id];
+    for (position, child_id) in node.children().with_position() {
+        let (connector, extension) = match position {
+            Position::Last | Position::Only => ("└── ", "    "),
+            _ => ("├── ", "│   "),
+        };
+        output.extend([
+            prefix.to_string(),
+            connector.to_string(),
+            qube_to_tree(qube, child_id, &format!("{prefix}{extension}"), depth - 1),
+        ]);
+    }
+
+    output.join("")
+}
+
+fn qube_to_html(qube: &Qube, node_id: &NodeId, prefix: &str, depth: usize) -> String {
+    let NodeSummary {
+        summary,
+        end: node_id,
+    } = summarise_nodes(qube, node_id, SummaryType::PlainText);
+
+    let node = &qube[node_id];
+    let mut output: Vec<String> = Vec::new();
+
+    let open = if depth > 0 { "open" } else { "" };
+    output.push(format!(
+        r#"<details {open}><summary class="qubed-level">{summary}</summary>"#
+    ));
+
+    for (position, child_id) in node.children().with_position() {
+        let (connector, extension) = match position {
+            Position::Last | Position::Only => ("└── ", "    "),
+            _ => ("├── ", "│   "),
+        };
+        output.extend([
+            prefix.to_string(),
+            connector.to_string(),
+            qube_to_tree(qube, child_id, &format!("{prefix}{extension}"), depth - 1),
+        ]);
+    }
+
+    output.join("")
+}
+
+impl Qube {
+    /// Return a string version of the Qube in the format
+    /// root
+    /// ├── class=od, expver=0001/0002, param=1/2
+    /// └── class=rd, param=1/2/3
+    pub fn string_tree(&self) -> String {
+        qube_to_tree(&self, &self.root, "", 5)
+    }
+
+    /// Return an HTML version of the Qube which renders like this
+    /// root
+    /// ├── class=od, expver=0001/0002, param=1/2
+    /// └── class=rd, param=1/2/3
+    /// But under the hood children are represented with a details/summary tag and each key=value is a span
+    /// CSS and JS functionality is bundled inside.
+    pub fn html_tree(&self) -> String {
+        qube_to_html(&self, &self.root, "", 5)
+    }
+}
--- a/src/rust/lib.rs
+++ b/src/rust/lib.rs
@ -1,97 +1,235 @@
 #![allow(unused_imports)]
-// #![allow(dead_code)]
-// #![allow(unused_variables)]
-

 use pyo3::prelude::*;
 use pyo3::wrap_pyfunction;
 use pyo3::types::{PyDict, PyInt, PyList, PyString};
+use python_interface::QubeError;
+use std::collections::HashMap;
+use std::iter;
+use pyo3::prelude::*;
+use std::hash::Hash;
+use std::rc::Rc;

-#[pyfunction]
-fn hello(_py: Python, name: &str) -> PyResult<String> {
-    Ok(format!("Hello, {}!", name))
+use lasso::{Rodeo, Spur};
+use std::num::NonZero;
+use std::ops;
+
+mod serialisation;
+mod python_interface;
+mod formatters;
+mod set_operations;
+
+// This data structure uses the Newtype Index Pattern
+// See https://matklad.github.io/2018/06/04/newtype-index-pattern.html
+// See also https://github.com/nrc/r4cppp/blob/master/graphs/README.md#rcrefcellnode for a discussion of other approaches to trees and graphs in rust.
+// https://smallcultfollowing.com/babysteps/blog/2015/04/06/modeling-graphs-in-rust-using-vector-indices/
+
+// Index types use struct Id(NonZero<usize>)
+// This reserves 0 as a special value which allows Option<Id(NonZero<usize>)> to be the same size as usize.
+
+#[derive(Debug, Copy, Clone, PartialEq, PartialOrd, Ord, Eq, Hash)]
+pub(crate) struct NodeId(NonZero<usize>);
+
+// Allow node indices to index directly into Qubes:
+impl ops::Index<NodeId> for Qube {
+    type Output = Node;
+
+    fn index(&self, index: NodeId) -> &Node {
+        &self.nodes[index.0.get() - 1]
+    }
 }

+impl ops::IndexMut<NodeId> for Qube {
+    fn index_mut(&mut self, index: NodeId) -> &mut Node {
+        &mut self.nodes[index.0.get() - 1]
+    }
+}
+
+impl ops::Index<StringId> for Qube {
+    type Output = str;
+
+    fn index(&self, index: StringId) -> &str {
+        &self.strings[index]
+    }
+}
+
+impl NodeId {
+    pub fn new(value: usize) -> Option<NodeId> {
+        NonZero::new(value).map(NodeId)
+    }
+}
+
+#[derive(Debug, Copy, Clone, PartialEq, PartialOrd, Ord, Eq, Hash)]
+struct StringId(lasso::Spur);
+
+impl ops::Index<StringId> for lasso::Rodeo {
+    type Output = str;
+
+    fn index(&self, index: StringId) -> &str {
+        &self[index.0]
+    }
+}
+
+#[derive(Debug, Clone)]
+pub(crate) struct Node {
+    pub key: StringId,
+    pub metadata: HashMap<StringId, Vec<String>>,
+    pub parent: Option<NodeId>, // If not present, it's the root node
+    pub values: Vec<StringId>,
+    pub children: HashMap<StringId, Vec<NodeId>>,
+}
+
+impl Node {
+    fn new_root(q: &mut Qube) -> Node {
+        Node {
+            key: q.get_or_intern("root"),
+            metadata: HashMap::new(),
+            parent: None,
+            values: vec![],
+            children: HashMap::new(),
+        }
+    }
+
+    fn children(&self) -> impl Iterator<Item = &NodeId> {
+        self.children.values().flatten()
+    }
+
+    fn is_root(&self) -> bool {
+        self.parent.is_none()
+    }
+
+    /// Because children are stored grouped by key
+    /// determining the number of children quickly takes a little effort.
+    /// This is a fast method for the special case of checking if a Node has exactly one child.
+    /// Returns Ok(NodeId) if there is one child else None
+    fn has_exactly_one_child(&self) -> Option<NodeId> {
+        if self.children.len() != 1 {return None}
+        let Some(value_group) = self.children.values().next() else {return None};
+        let [node_id] = &value_group.as_slice() else {return None};
+        Some(*node_id)
+    }
+
+    fn n_children(&self) -> usize {
+        self.children
+            .values()
+            .map(|v| v.len())
+            .sum()
+    }
+
+    fn keys<'a>(&'a self, q: &'a Qube) -> impl Iterator<Item = &'a str> {
+        self.children.keys()
+        .map(|s| {&q[*s]})
+    }
+}
+
+#[derive(Debug, Clone)]
+#[pyclass(subclass, dict)]
+pub struct Qube {
+    pub root: NodeId,
+    nodes: Vec<Node>,
+    strings: Rodeo,
+}
+
+impl Qube {
+    pub fn new() -> Self {
+        let mut q = Self {
+            root: NodeId::new(1).unwrap(),
+            nodes: Vec::new(),
+            strings: Rodeo::default(),
+        };
+
+        let root = Node::new_root(&mut q);
+        q.nodes.push(root);
+        q
+    }
+
+    fn get_or_intern(&mut self, val: &str) -> StringId {
+        StringId(self.strings.get_or_intern(val))
+    }
+
+    pub(crate) fn add_node(&mut self, parent: NodeId, key: &str, values: impl IntoIterator<Item = impl AsRef<str>>) -> NodeId {
+        let key_id = self.get_or_intern(key);
+        let values = values.into_iter().map(|val| self.get_or_intern(val.as_ref())).collect();
+
+        // Create the node object
+        let node = Node {
+            key: key_id,
+            metadata: HashMap::new(),
+            values: values,
+            parent: Some(parent),
+            children: HashMap::new(),
+        };
+
+        // Insert it into the Qube arena and determine its id
+        self.nodes.push(node);
+        let node_id = NodeId::new(self.nodes.len()).unwrap();
+
+        // Add a reference to this node's id to the parents list of children.
+        let parent_node = &mut self[parent];
+        let key_group = parent_node.children.entry(key_id).or_insert(Vec::new());
+        key_group.push(node_id);
+
+        node_id
+    }
+
+    fn print(&self, node_id: Option<NodeId>) -> String {
+        let node_id: NodeId = node_id.unwrap_or(self.root);
+        let node = &self[node_id];
+        node.summary(&self)
+    }
+
+    fn get_node_ref(&self, id: NodeId) -> NodeRef {
+        let node = &self[id];
+        NodeRef { id: id, node: &node, qube: &self }
+    }
+
+    pub fn get_string_id(&self, s: &str) -> Option<StringId> {
+        self.strings.get(s)
+        .map(|id| StringId(id))
+    }
+}
+
+
 #[pymodule]
-fn rust(m: &Bound<'_, PyModule>) -> PyResult<()> {
-    m.add_function(wrap_pyfunction!(hello, m)?).unwrap();
+fn rust(py: Python<'_>, m: &Bound<'_, PyModule>) -> PyResult<()> {
+    m.add_class::<Qube>()?;
+    m.add("QubeError", py.get_type::<python_interface::QubeError>())?;
    Ok(())
 }


-// use rsfdb::listiterator::KeyValueLevel;
-// use rsfdb::request::Request;
-// use rsfdb::FDB;
+pub struct NodeRef<'a> {
+    pub id: NodeId,
+    pub node: &'a Node,
+    pub qube: &'a Qube,
+}

-// use serde_json::{json, Value};
-// use std::time::Instant;
+impl<'a> NodeRef<'a> {
+    pub fn keys(&self) -> impl Iterator<Item = &str> {
+        self.node.keys(self.qube)
+    }
+
+    fn flat_children(&'a self) -> impl Iterator<Item = Self> {
+        self.node.children
+        .values()
+        .flatten()
+        .map(|id| {
+            NodeRef { id: *id, node: &self.qube[*id], qube: self.qube }
+        })
+    }
+
+    fn children_by_key(&'a self, key: &str) -> impl Iterator<Item = Self> {
+        let id = self.qube.get_string_id(key);
+        let children = id
+            .map(|i| self.node.children.get(&i))
+            .flatten();
+
+        children.map(
+            |ids| ids.into_iter().map(
+                |id| {
+                NodeRef { id: *id, node: &self.qube[*id], qube: self.qube }
+        })).into_iter().flatten()
+    }


-// use std::collections::HashMap;
-
-// pub mod tree;
-// use std::sync::Arc;
-// use std::sync::Mutex;
-// use tree::TreeNode;
-
-// #[pyclass(unsendable)]
-// pub struct PyFDB {
-//     pub fdb: FDB,
-// }
-
-// #[pymethods]
-// impl PyFDB {
-//     #[new]
-//     #[pyo3(signature = (fdb_config=None))]
-//     pub fn new(fdb_config: Option<&str>) -> PyResult<Self> {
-//         let fdb = FDB::new(fdb_config)
-//             .map_err(|e| PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(e.to_string()))?;
-//         Ok(PyFDB { fdb })
-//     }
-
-//     /// Traverse the FDB with the given request.
-//     pub fn traverse_fdb(
-//         &self,
-//         py: Python<'_>,
-//         request: HashMap<String, Vec<String>>,
-//     ) -> PyResult<PyObject> {
-//         let start_time = Instant::now();
-
-//         let list_request = Request::from_json(json!(request))
-//             .map_err(|e| PyErr::new::<pyo3::exceptions::PyValueError, _>(e.to_string()))?;
-
-//         // Use `fdb_guard` instead of `self.fdb`
-//         let list = self
-//             .fdb
-//             .list(&list_request, true, true)
-//             .map_err(|e| PyErr::new::<pyo3::exceptions::PyRuntimeError, _>(e.to_string()))?;
-
-//         let mut root = TreeNode::new(KeyValueLevel {
-//             key: "root".to_string(),
-//             value: "root".to_string(),
-//             level: 0,
-//         });
-
-//         for item in list {
-//             py.check_signals()?;
-
-//             if let Some(request) = &item.request {
-//                 root.insert(&request);
-//             }
-//         }
-
-//         let duration = start_time.elapsed();
-//         println!("Total runtime: {:?}", duration);
-
-//         let py_dict = root.to_py_dict(py)?;
-//         Ok(py_dict)
-//     }
-// }
-
-// use pyo3::prelude::*;
-
-// #[pymodule]
-// fn rust(m: &Bound<'_, PyModule>) -> PyResult<()> {
-//     m.add_class::<PyFDB>()?;
-//     Ok(())
-// }
+}
--- a/src/rust/python_interface.rs
+++ b/src/rust/python_interface.rs
@ -0,0 +1,179 @@
+use crate::{Node, NodeId, Qube, NodeRef};
+use pyo3::prelude::*;
+use pyo3::types::{PyList, PyType};
+use core::borrow;
+use std::ops::Deref;
+use std::cell::Ref;
+
+use crate::set_operations;
+use crate::serialisation;
+use itertools::Itertools;
+
+use pyo3::create_exception;
+
+create_exception!(qubed, QubeError, pyo3::exceptions::PyException);
+
+/// A reference to a particular node in a Qube
+#[pyclass]
+pub struct PyNodeRef {
+    id: NodeId,
+    qube: Py<Qube>, // see https://pyo3.rs/v0.23.1/types for a discussion of Py<T> and Bound<'py, T>
+}
+
+fn into_py_node_ref(node_ref: NodeRef, qube: Py<Qube>) -> PyNodeRef {
+    PyNodeRef {
+        id: node_ref.id,
+        qube: qube,
+    }
+}
+
+#[pymethods]
+impl PyNodeRef {
+    fn __repr__(&self, py: Python) -> PyResult<String> {
+        // Get the Py<Qube> reference, bind it to the GIL.
+        let qube = self.qube.bind(py);
+
+        fn repr_helper<'py>(node_id: NodeId, qube: &Bound<'py, Qube>) -> String {
+            let node = &qube.borrow()[node_id];
+            let key = &qube.borrow()[node.key];
+            let children = node
+                .children
+                .values()
+                .flatten()
+                .map(|child_id| repr_helper(child_id.clone(), qube))
+                .collect::<Vec<String>>()
+                .join(", ");
+
+            format!("Node({}, {})", key, children)
+        }
+
+        Ok(repr_helper(self.id, qube))
+    }
+
+    fn __str__(&self, py: Python) -> String {
+        let qube = self.qube.bind(py).borrow();
+        let node = &qube[self.id];
+        let key = &qube.strings[node.key];
+        format!("Node({})", key)
+    }
+
+    #[getter]
+    pub fn get_children(&self, py: Python) -> Vec<Self> {
+        let qube = self.qube.bind(py).borrow();
+        let node = &qube[self.id];
+        node.children
+            .values()
+            .flatten()
+            .map(|child_id| Self {
+                id: *child_id,
+                qube: self.qube.clone_ref(py),
+            })
+            .collect()
+    }
+}
+
+#[derive(FromPyObject)]
+pub enum OneOrMany<T> {
+    One(T),
+    Many(Vec<T>),
+}
+
+// Todo: Is there a way to rewrite this so that is doesn't allocate?
+// Perhaps by returning an iterator?
+impl<T> Into<Vec<T>> for OneOrMany<T> {
+    fn into(self) -> Vec<T> {
+        match self {
+            OneOrMany::One(v) => vec![v],
+            OneOrMany::Many(vs) => vs,
+        }
+    }
+}
+
+#[pymethods]
+impl Qube {
+    #[new]
+    pub fn py_new() -> Self {
+        Qube::new()
+    }
+
+    #[pyo3(name = "add_node")]
+    pub fn py_add_node(
+        slf: Bound<'_, Self>,
+        parent: PyRef<'_, PyNodeRef>,
+        key: &str,
+        values: OneOrMany<String>,
+    ) -> PyResult<PyNodeRef> {
+        // Check that the given parent is actually in this qube and not another one
+        if !parent.qube.bind(slf.py()).is(&slf) {
+            return Err(QubeError::new_err("Supplied parent node is not in the target qube."))
+        }
+
+        // massage values from T | Vec<T> into Vec<T>
+        let values: Vec<String> = values.into();
+        let mut q = slf.borrow_mut();
+        let node_id = q.add_node(parent.id, key, &values);
+        Ok(PyNodeRef { id: node_id, qube: slf.into()})
+    }
+
+    pub fn set_root(
+        slf: Bound<'_, Self>,
+        node: PyRef<'_, PyNodeRef>,
+    ) -> () {
+        let mut q = slf.borrow_mut();
+        q.root = node.id;
+    }
+
+    #[getter]
+    fn get_root(slf: Bound<'_, Self>) -> PyResult<PyNodeRef> {
+        Ok(PyNodeRef {
+            id: slf.borrow().root,
+            qube: slf.unbind(),
+        })
+    }
+
+    fn __repr__(&self) -> String {
+        // format!("{:?}", self)
+        let nodes_str: String = self.nodes.iter()
+        .enumerate()
+        .map(|(id, node)| {
+            format!("{{id: {}, key: {}, values: [{}], children: [{}]}}",
+            id+1,
+            &self[node.key],
+            node.values.iter().map(|s| &self[*s]).join(", "),
+            node.children().map(|n| n.0).join(", "),
+        )
+        }).join(", ");
+        format!("Qube {{root: {}, nodes: {}}}", self.root.0, nodes_str)
+    }
+
+    fn __str__<'py>(&self) -> String {
+        self.string_tree()
+    }
+
+    fn _repr_html_(&self) -> String {
+        self.html_tree()
+    }
+
+    #[pyo3(name = "print")]
+    fn py_print(&self) -> String {
+        self.print(Option::None)
+    }
+
+    #[getter]
+    pub fn get_children(slf: Bound<'_, Self>, py: Python) -> PyResult<Vec<PyNodeRef>> {
+        let root = PyNodeRef {
+            id: slf.borrow().root,
+            qube: slf.unbind(),
+        };
+        Ok(root.get_children(py))
+    }
+
+    #[staticmethod]
+    pub fn from_json(data: &str) -> Result<Self, serialisation::JSONError> {
+        serialisation::from_json(data)
+    }
+
+    pub fn __or__(slf: Bound<'_, Self>, other: Bound<'_, Qube>) -> Qube {
+    set_operations::set_operation(&slf.borrow(), &other.borrow(), set_operations::Op::Union)
+    }
+}
--- a/src/rust/serialisation/json.rs
+++ b/src/rust/serialisation/json.rs
@ -0,0 +1,80 @@
+use pyo3::exceptions::PyValueError;
+use pyo3::prelude::*;
+use serde::{Deserialize, Serialize};
+use serde_json::Value;
+use std::collections::HashMap;
+
+use crate::{Node, NodeId, Qube};
+
+// Use a newtype wrapper to allow us to implement auto conversion from serde_json::Error to PyErr
+// via a wrapper intermediate
+// see https://pyo3.rs/main/function/error-handling.html#foreign-rust-error-types
+pub struct JSONError(serde_json::Error);
+
+impl From<JSONError> for PyErr {
+    fn from(error: JSONError) -> Self {
+        PyValueError::new_err(format!("{}", error.0))
+    }
+}
+
+impl From<serde_json::Error> for JSONError {
+    fn from(other: serde_json::Error) -> Self {
+        Self(other)
+    }
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+#[serde(tag = "dtype")]
+enum Ranges {
+    Int64{values: Vec<(i64, i64)>}
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+#[serde(tag = "dtype", rename_all = "lowercase")]
+enum Enum {
+    Str{values: Vec<String>}
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+#[serde(tag = "type", rename_all = "lowercase")]
+enum Values {
+    Wildcard{},
+    Enum(Enum),
+    Range(Ranges)
+}
+
+#[derive(Serialize, Deserialize, Debug)]
+struct JSONQube {
+    key: String,
+    values: Values,
+    metadata: HashMap<String, String>,
+    children: Vec<JSONQube>,
+}
+
+fn add_nodes(qube: &mut Qube, parent: NodeId, nodes: &[JSONQube]) -> Vec<NodeId> {
+    nodes
+        .iter()
+        .map(|json_node| {
+            let values = match &json_node.values {
+                Values::Wildcard{} => &vec!["*"],
+                Values::Enum(Enum::Str{values}) => &values.iter().map(|s| s.as_str()).collect(),
+                Values::Range(_) => todo!(),
+            };
+            let node_id = qube.add_node(parent, &json_node.key, values);
+
+            //
+            add_nodes(qube, node_id, &json_node.children);
+            node_id
+        })
+        .collect()
+}
+
+pub fn from_json(data: &str) -> Result<Qube, JSONError> {
+    // Parse the string of data into serde_json::Value.
+    let json_qube: JSONQube = serde_json::from_str(data).expect("JSON parsing failed");
+
+    let mut qube = Qube::new();
+    let root = qube.root;
+    add_nodes(&mut qube, root, &json_qube.children);
+    Ok(qube)
+}
--- a/src/rust/serialisation/mod.rs
+++ b/src/rust/serialisation/mod.rs
@ -0,0 +1,2 @@
+mod json;
+pub use json::{from_json, JSONError};
--- a/src/rust/set_operations.rs
+++ b/src/rust/set_operations.rs
@ -0,0 +1,40 @@
+use crate::NodeRef;
+use crate::{Node, NodeId, Qube};
+use itertools::chain;
+use std::collections::HashSet;
+
+pub enum Op {
+    Union,
+    Intersection,
+    Difference,
+    SymmetricDifference,
+}
+
+fn op_to_venn_diagram(op: Op) -> (bool, bool, bool) {
+    use Op::*;
+    match op {
+        Union => (true, true, true),
+        Intersection => (false, true, false),
+        Difference => (true, false, false),
+        SymmetricDifference => (true, false, true),
+    }
+}
+
+pub fn set_operation<'a>(a: &'a Qube, b: &'a Qube, op: Op) -> Qube {
+    todo!()
+    // _set_operation(a.root_ref(), a.root_ref(), op)
+}
+
+// fn _set_operation<'a>(a: NodeRef, b: NodeRef, op: Op) -> Qube {
+//     let keys: HashSet<&str> = HashSet::from_iter(chain(a.keys(), b.keys()));
+
+//     for key in keys {
+//         let a = a.children_by_key(key)
+//     }
+
+//     todo!()
+// }
+
+pub fn set_operation_inplace<'a>(a: &'a mut Qube, b: &'a Qube, op: Op) -> &'a Qube {
+    a
+}
--- a/src/rust/tree.rs
+++ b/src/rust/tree.rs
@ -1,82 +0,0 @@
-// use pyo3::prelude::*;
-// use pyo3::types::PyDict;
-// use rsfdb::listiterator::KeyValueLevel;
-// use serde_json::Value;
-
-// #[derive(Debug)]
-// pub struct TreeNode {
-//     pub key: KeyValueLevel,
-//     pub children: Vec<TreeNode>,
-// }
-
-// impl TreeNode {
-//     pub fn new(key: KeyValueLevel) -> Self {
-//         TreeNode {
-//             key,
-//             children: Vec::new(),
-//         }
-//     }
-
-//     pub fn insert(&mut self, path: &[KeyValueLevel]) {
-//         if path.is_empty() {
-//             return;
-//         }
-
-//         let kvl = &path[0];
-
-//         // Check if a child with the same key and value exists
-//         if let Some(child) = self.children.iter_mut().find(|child| child.key == *kvl) {
-//             // Insert the remaining path into the existing child
-//             child.insert(&path[1..]);
-//         } else {
-//             // Create a new child node
-//             let mut new_child = TreeNode::new(kvl.clone());
-//             new_child.insert(&path[1..]);
-//             self.children.push(new_child);
-//         }
-//     }
-
-//     pub fn traverse<F>(&self, level: usize, callback: &F)
-//     where
-//         F: Fn(&TreeNode, usize),
-//     {
-//         callback(self, level);
-//         for child in &self.children {
-//             child.traverse(level + 1, callback);
-//         }
-//     }
-
-//     pub fn to_json(&self) -> Value {
-//         let formatted_key = format!("{}={}", self.key.key, self.key.value);
-
-//         let children_json: Value = if self.children.is_empty() {
-//             Value::Object(serde_json::Map::new())
-//         } else {
-//             Value::Object(
-//                 self.children
-//                     .iter()
-//                     .map(|child| {
-//                         (
-//                             format!("{}={}", child.key.key, child.key.value),
-//                             child.to_json(),
-//                         )
-//                     })
-//                     .collect(),
-//             )
-//         };
-
-//         // Combine the formatted key with children
-//         serde_json::json!({ formatted_key: children_json })
-//     }
-
-//     pub fn to_py_dict(&self, py: Python) -> PyResult<PyObject> {
-//         let py_dict = PyDict::new(py);
-
-//         for child in &self.children {
-//             let child_key = format!("{}={}", child.key.key, child.key.value);
-//             py_dict.set_item(child_key, child.to_py_dict(py)?)?;
-//         }
-
-//         Ok(py_dict.to_object(py))
-//     }
-// }
--- a/stac_server/main.py
+++ b/stac_server/main.py
@ -1,17 +1,21 @@
 import json
 import os
 from collections import defaultdict
-from pathlib import Path
-from typing import Any, Dict

-import redis
+import requests
 import yaml
-from fastapi import FastAPI, Request
+from fastapi import Depends, FastAPI, HTTPException, Request
 from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import FileResponse
-from tree_traverser import CompressedTree
+from fastapi.responses import FileResponse, HTMLResponse
+from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
+from fastapi.staticfiles import StaticFiles
+from fastapi.templating import Jinja2Templates
+from frozendict import frozendict
+from qubed import Qube
+from qubed.tree_formatters import node_tree_to_html

 app = FastAPI()
+security = HTTPBearer()
 app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
@ -20,39 +24,97 @@ app.add_middleware(
    allow_headers=["*"],
 )

+app.mount("/static", StaticFiles(directory="static"), name="static")
+templates = Jinja2Templates(directory="templates")

-@app.get("/favicon.ico", include_in_schema=False)
-async def favicon():
-    return FileResponse("favicon.ico")
-
+qubes: dict[str, Qube] = {}
+# print("Getting climate and extremes dt data from github")
+# try:
+#     qubes["climate-dt"] = Qube.from_json(
+#         requests.get(
+#             "https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/climate_dt.json",
+#         timeout=3).json()
+#     )
+#     qubes["extremes-dt"] = Qube.from_json(
+#         requests.get(
+#             "https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/extremes_dt.json",
+#         timeout=3).json()
+#     )
+#     mars_language = yaml.safe_load(
+#         requests.get(
+#             "https://github.com/ecmwf/qubed/raw/refs/heads/main/config/climate-dt/language.yaml",
+#         timeout=3).content
+#     )
+# except:
+qubes["climate-dt"] = Qube.empty()
+qubes["extremes-dt"] = Qube.empty()
+mars_language = {}

 if "LOCAL_CACHE" in os.environ:
-    print("Getting data from local file")
+    print("Getting climate and extremes dt data from local files")
+    with open("../tests/example_qubes/climate_dt.json") as f:
+        qubes["climate-dt"] = Qube.from_json(json.load(f))

-    base = Path(os.environ["LOCAL_CACHE"])
-    with open(base / "compressed_tree.json", "r") as f:
-        json_tree = f.read()
+    with open("../tests/example_qubes/extremes_dt.json") as f:
+        qubes["climate-dt"] = qubes["climate-dt"] | Qube.from_json(json.load(f))

-    with open(base / "language.yaml", "r") as f:
+    with open("../tests/example_qubes/od.json") as f:
+        qubes["climate-dt"] = qubes["climate-dt"] | Qube.from_json(json.load(f))
+
+    with open("../config/language/language.yaml", "r") as f:
        mars_language = yaml.safe_load(f)["_field"]

+    with open("../config/language/paramids.yaml", "r") as f:
+        params = yaml.safe_load(f)
 else:
-    print("Getting cache from redis")
-    r = redis.Redis(host="redis", port=6379, db=0)
-    json_tree = r.get("compressed_catalog")
-    assert json_tree, "No compressed tree found in redis"
-    mars_language = json.loads(r.get("mars_language"))
+    print("Getting climate and extremes dt data from github")
+    qubes["climate-dt"] = Qube.from_json(
+        requests.get(
+            "https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/climate_dt.json",
+            timeout=1,
+        ).json()
+    )
+    qubes["extremes-dt"] = Qube.from_json(
+        requests.get(
+            "https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/extremes_dt.json",
+            timeout=1,
+        ).json()
+    )

-print("Loading tree from json")
-c_tree = CompressedTree.from_json(json.loads(json_tree))
+    qubes["od"] = Qube.from_json(
+        requests.get(
+            "https://github.com/ecmwf/qubed/raw/refs/heads/main/tests/example_qubes/od.json",
+            timeout=1,
+        ).json()
+    )
+    qubes["climate-dt"] = qubes["climate-dt"] | qubes["extremes-dt"] | qubes["od"]
+    mars_language = yaml.safe_load(
+        requests.get(
+            "https://github.com/ecmwf/qubed/raw/refs/heads/main/config/climate-dt/language.yaml",
+            timeout=3,
+        ).content
+    )["_field"]

-print("Partialy decompressing tree, shoud be able to skip this step in future.")
-tree = c_tree.reconstruct_compressed_ecmwf_style()
+if "API_KEY" in os.environ:
+    api_key = os.environ["API_KEY"]
+else:
+    with open("api_key.secret", "r") as f:
+        api_key = f.read()

 print("Ready to serve requests!")


-def request_to_dict(request: Request) -> Dict[str, Any]:
+def validate_key(key: str):
+    if key not in qubes:
+        raise HTTPException(status_code=404, detail=f"Qube {key} not found")
+    return key
+
+
+async def get_body_json(request: Request):
+    return await request.json()
+
+
+def parse_request(request: Request) -> dict[str, str | list[str]]:
    # Convert query parameters to dictionary format
    request_dict = dict(request.query_params)
    for key, value in request_dict.items():
@ -63,116 +125,68 @@ def request_to_dict(request: Request) -> Dict[str, Any]:
    return request_dict


-def match_against_cache(request, tree):
-    if not tree:
-        return {"_END_": {}}
-    matches = {}
-    for k, subtree in tree.items():
-        if len(k.split("=")) != 2:
-            raise ValueError(f"Key {k} is not in the correct format")
-        key, values = k.split("=")
-        values = set(values.split(","))
-        if key in request:
-            if isinstance(request[key], list):
-                matching_values = ",".join(
-                    request_value
-                    for request_value in request[key]
-                    if request_value in values
-                )
-                if matching_values:
-                    matches[f"{key}={matching_values}"] = match_against_cache(
-                        request, subtree
-                    )
-            elif request[key] in values:
-                matches[f"{key}={request[key]}"] = match_against_cache(request, subtree)
-
-    if not matches:
-        return {k: {} for k in tree.keys()}
-    return matches
+def validate_api_key(credentials: HTTPAuthorizationCredentials = Depends(security)):
+    if credentials.credentials != api_key:
+        raise HTTPException(status_code=403, detail="Incorrect API Key")
+    return credentials


-def max_tree_depth(tree):
-    "Figure out the maximum depth of a tree"
-    if not tree:
-        return 0
-    return 1 + max(max_tree_depth(v) for v in tree.values())
+@app.get("/favicon.ico", include_in_schema=False)
+async def favicon():
+    return FileResponse("favicon.ico")


-def prune_short_branches(tree, depth=None):
-    if depth is None:
-        depth = max_tree_depth(tree)
-    return {
-        k: prune_short_branches(v, depth - 1)
-        for k, v in tree.items()
-        if max_tree_depth(v) == depth - 1
-    }
+@app.get("/", response_class=HTMLResponse)
+async def read_root(request: Request):
+    return templates.TemplateResponse(
+        "index.html",
+        {
+            "request": request,
+            "config": {
+                "message": "Hello from the dev server!",
+            },
+            "api_url": os.environ.get("API_URL", "/api/v1/"),
+        },
+    )


-def get_paths_to_leaves(tree):
-    for k, v in tree.items():
-        if not v:
-            yield [
-                k,
-            ]
-        else:
-            for leaf in get_paths_to_leaves(v):
-                yield [
-                    k,
-                ] + leaf
+@app.get("/api/v1/keys/")
+async def keys():
+    return list(qubes.keys())


-def get_leaves(tree):
-    for k, v in tree.items():
-        if not v:
-            yield k
-        else:
-            for leaf in get_leaves(v):
-                yield leaf
+@app.get("/api/v1/get/{key}/")
+async def get(
+    key: str = Depends(validate_key),
+    request: dict[str, str | list[str]] = Depends(parse_request),
+):
+    return qubes[key].to_json()


-@app.get("/api/tree")
-async def get_tree(request: Request):
-    request_dict = request_to_dict(request)
-    print(c_tree.multi_match(request_dict))
-    return c_tree.multi_match(request_dict)
+@app.post("/api/v1/union/{key}/")
+async def union(
+    key: str,
+    credentials: HTTPAuthorizationCredentials = Depends(validate_api_key),
+    body_json=Depends(get_body_json),
+):
+    if key not in qubes:
+        qubes[key] = Qube.empty()
+
+    q = Qube.from_json(body_json)
+    qubes[key] = qubes[key] | q
+    return qubes[key].to_json()


-@app.get("/api/match")
-async def get_match(request: Request):
-    # Convert query parameters to dictionary format
-    request_dict = request_to_dict(request)
-
-    # Run the schema matching logic
-    match_tree = match_against_cache(request_dict, tree)
-
-    # Prune the tree to only include branches that are as deep as the deepest match
-    # This means if you don't choose a certain branch at some point
-    # the UI won't keep nagging you to choose a value for that branch
-    match_tree = prune_short_branches(match_tree)
-
-    return match_tree
-
-
-@app.get("/api/paths")
-async def api_paths(request: Request):
-    request_dict = request_to_dict(request)
-    match_tree = match_against_cache(request_dict, tree)
-    match_tree = prune_short_branches(match_tree)
-    paths = get_paths_to_leaves(match_tree)
-
-    # deduplicate leaves based on the key
+def follow_query(request: dict[str, str | list[str]], qube: Qube):
+    s = qube.select(request, mode="next_level", consume=False)
    by_path = defaultdict(lambda: {"paths": set(), "values": set()})
-    for p in paths:
-        if p[-1] == "_END_":
-            continue
-        key, values = p[-1].split("=")
-        values = values.split(",")
-        path = tuple(p[:-1])

-        by_path[key]["values"].update(values)
-        by_path[key]["paths"].add(tuple(path))
+    for request, node in s.leaf_nodes():
+        if not node.metadata.get("is_leaf", True):
+            by_path[node.key]["values"].update(node.values.values)
+            by_path[node.key]["paths"].add(frozendict(request))

-    return [
+    return s, [
        {
            "paths": list(v["paths"]),
            "key": key,
@ -182,49 +196,140 @@ async def api_paths(request: Request):
    ]


-@app.get("/api/stac")
-async def get_STAC(request: Request):
-    request_dict = request_to_dict(request)
-    paths = await api_paths(request)
+@app.get("/api/v1/select/{key}/")
+async def select(
+    key: str = Depends(validate_key),
+    request: dict[str, str | list[str]] = Depends(parse_request),
+):
+    q = qubes[key].select(request)
+    return q.to_json()
+
+
+@app.get("/api/v1/query/{key}")
+async def query(
+    key: str = Depends(validate_key),
+    request: dict[str, str | list[str]] = Depends(parse_request),
+):
+    qube, paths = follow_query(request, qubes[key])
+    return paths
+
+
+@app.get("/api/v1/basicstac/{key}/{filters:path}")
+async def basic_stac(filters: str, key: str = Depends(validate_key)):
+    pairs = filters.strip("/").split("/")
+    request = dict(p.split("=") for p in pairs if "=" in p)
+
+    qube, _ = follow_query(request, qubes[key])
+
+    def make_link(child_request):
+        """Take a MARS Key and information about which paths matched up to this point and use it to make a STAC Link"""
+        kvs = [f"{key}={value}" for key, value in child_request.items()]
+        href = f"/api/v1/basicstac/{key}/{'/'.join(kvs)}"
+        last_key, last_value = list(child_request.items())[-1]
+
+        return {
+            "title": f"{last_key}={last_value}",
+            "href": href,
+            "rel": "child",
+            "type": "application/json",
+        }
+
+    # Format the response as a STAC collection
+    (this_key, this_value), *_ = (
+        list(request.items())[-1] if request else ("root", "root"),
+        None,
+    )
+    key_info = mars_language.get(this_key, {})
+    try:
+        values_info = dict(key_info.get("values", {}))
+        value_info = values_info.get(
+            this_value, f"No info found for value `{this_value}` found."
+        )
+    except ValueError:
+        value_info = f"No info found for value `{this_value}` found."
+
+    if this_key == "root":
+        value_info = "The root node"
+    # key_desc = key_info.get(
+    #     "description", f"No description for `key` {this_key} found."
+    # )
+    print(this_key, this_value)
+
+    print(this_key, key_info)
+    stac_collection = {
+        "type": "Catalog",
+        "stac_version": "1.0.0",
+        "id": "root"
+        if not request
+        else "/".join(f"{k}={v}" for k, v in request.items()),
+        "title": f"{this_key}={this_value}",
+        "description": value_info,
+        "links": [make_link(leaf) for leaf in qube.leaves()],
+        # "debug": {
+        #     "qube": str(qube),
+        # },
+    }
+
+    return stac_collection
+
+
+@app.get("/api/v1/stac/{key}/")
+async def get_STAC(
+    key: str = Depends(validate_key),
+    request: dict[str, str | list[str]] = Depends(parse_request),
+):
+    qube, paths = follow_query(request, qubes[key])
+    kvs = [
+        f"{k}={','.join(v)}" if isinstance(v, list) else f"{k}={v}"
+        for k, v in request.items()
+    ]
+    request_params = "&".join(kvs)

    def make_link(key_name, paths, values):
        """Take a MARS Key and information about which paths matched up to this point and use it to make a STAC Link"""
-        path = paths[0]
-        href_template = f"/stac?{'&'.join(path)}{'&' if path else ''}{key_name}={{}}"
-        optional = [False]
-        # optional_str = (
-        #     "Yes"
-        #     if all(optional) and len(optional) > 0
-        #     else ("Sometimes" if any(optional) else "No")
-        # )
-        values_from_mars_language = mars_language.get(key_name, {}).get("values", [])
+        href_template = f"/stac?{request_params}{'&' if request_params else ''}{key_name}={{{key_name}}}"

-        # values = [v[0] if isinstance(v, list) else v for v in values_from_mars_language]
+        print(f"{key_name = }")
+        if key_name == "param":
+            print(params)
+            values_from_mars_language = params
+            value_descriptions = [
+                max(params.get(int(v), [""]), key=len) for v in values
+            ]
+            print(value_descriptions)
+        else:
+            values_from_mars_language = mars_language.get(key_name, {}).get(
+                "values", []
+            )

-        if all(isinstance(v, list) for v in values_from_mars_language):
-            value_descriptions_dict = {
-                k: v[-1]
-                for v in values_from_mars_language
-                if len(v) > 1
-                for k in v[:-1]
-            }
-            value_descriptions = [value_descriptions_dict.get(v, "") for v in values]
-            if not any(value_descriptions):
-                value_descriptions = None
+            if all(isinstance(v, list) for v in values_from_mars_language):
+                value_descriptions_dict = {
+                    k: v[-1]
+                    for v in values_from_mars_language
+                    if len(v) > 1
+                    for k in v[:-1]
+                }
+                value_descriptions = [
+                    value_descriptions_dict.get(v, "") for v in values
+                ]
+                if not any(value_descriptions):
+                    value_descriptions = None

        return {
            "title": key_name,
-            "generalized_datacube:href_template": href_template,
+            "uriTemplate": href_template,
            "rel": "child",
            "type": "application/json",
-            "generalized_datacube:dimension": {
-                "type": mars_language.get(key_name, {}).get("type", ""),
-                "description": mars_language.get(key_name, {}).get("description", ""),
-                "values": values,
-                "value_descriptions": value_descriptions,
-                "optional": any(optional),
-                "multiple": True,
-                "paths": paths,
+            "variables": {
+                key_name: {
+                    "type": "string",
+                    "description": mars_language.get(key_name, {}).get(
+                        "description", ""
+                    ),
+                    "enum": values,
+                    "value_descriptions": value_descriptions,
+                    # "paths": paths,
+                }
            },
        }

@ -242,20 +347,29 @@ async def get_STAC(request: Request):
            "description": mars_language.get(key, {}).get("description", ""),
            "value_descriptions": value_descriptions(key, values),
        }
-        for key, values in request_dict.items()
+        for key, values in request.items()
    }

    # Format the response as a STAC collection
    stac_collection = {
-        "type": "Collection",
+        "type": "Catalog",
        "stac_version": "1.0.0",
-        "id": "partial-matches",
+        "id": "root" if not request else "/stac?" + request_params,
        "description": "STAC collection representing potential children of this request",
        "links": [make_link(p["key"], p["paths"], p["values"]) for p in paths],
        "debug": {
-            "request": request_dict,
+            # "request": request,
            "descriptions": descriptions,
-            "paths": paths,
+            # "paths": paths,
+            "qube": node_tree_to_html(
+                qube.compress(),
+                collapse=True,
+                depth=10,
+                include_css=False,
+                include_js=False,
+                max_summary_length=200,
+                css_id="qube",
+            ),
        },
    }

--- a/stac_server/requirements.txt
+++ b/stac_server/requirements.txt
@ -1,3 +1,5 @@
 fastapi[standard]
 pe
 redis
+frozendict
+requests
--- a/stac_server/run.sh
+++ b/stac_server/run.sh
@ -1,3 +1,3 @@
 parent_path=$( cd "$(dirname "${BASH_SOURCE[0]}")" ; pwd -P )
 cd "$parent_path"
-LOCAL_CACHE=../config/climate-dt fastapi dev ./main.py --port 8124 --reload
+LOCAL_CACHE=True fastapi dev ./main.py --port 8124 --reload
--- a/stac_server/run_prod.sh
+++ b/stac_server/run_prod.sh
@ -0,0 +1,3 @@
+parent_path=$( cd "$(dirname "${BASH_SOURCE[0]}")" ; pwd -P )
+cd "$parent_path"
+sudo LOCAL_CACHE=True ../../.venv/bin/fastapi dev ./main.py --port 80 --host=0.0.0.0 --reload
--- a/web_query_builder/static/app.js
+++ b/web_query_builder/static/app.js
@ -138,39 +138,23 @@ async function createCatalogItem(link, itemsContainer) {
    // Update the item div with real content
    itemDiv.classList.remove("loading");

-    const dimension = link["generalized_datacube:dimension"];
+    const variables = link["variables"];
+    const key = Object.keys(variables)[0];
+    const variable = variables[key];

    // add data-key attribute to the itemDiv
    itemDiv.dataset.key = link.title;
-    itemDiv.dataset.keyType = dimension.type;
+    itemDiv.dataset.keyType = variable.type;

    itemDiv.innerHTML = `
      <h3 class="item-title">${link.title || "No title available"}</h3>
      <p class="item-type">Key Type: ${itemDiv.dataset.keyType || "Unknown"}</p>
-      <!-- <p class="item-type">Paths: ${dimension.paths}</p> -->
-      <p class="item-type">Optional: ${dimension.optional ? "Yes" : "No"}</p>
      <p class="item-description">${
-        dimension.description
-          ? dimension.description.slice(0, 100)
-          : "No description available"
-      }...</p>
+        variable.description ? variable.description.slice(0, 100) : ""
+      }</p>
    `;

-    // if (dimension.type === "date" || dimension.type === "time") {
-    //   // Render a date picker for the "date" key
-    //   const picker = `<input type="${link.title}" name="${link.title}">`;
-    //   //convert picker to HTML node
-    //   const pickerNode = document
-    //     .createRange()
-    //     .createContextualFragment(picker);
-    //   itemDiv.appendChild(pickerNode);
-    // }
-    // Otherwise create a scrollable list with checkboxes for values if available
-    if (
-      //   dimension.type === "enum" &&
-      dimension.values &&
-      dimension.values.length > 0
-    ) {
+    if (variable.enum && variable.enum.length > 0) {
      const listContainer = renderCheckboxList(link);
      itemDiv.appendChild(listContainer);
    } else {
@ -185,14 +169,15 @@ async function createCatalogItem(link, itemsContainer) {
 }

 function renderCheckboxList(link) {
-  const dimension = link["generalized_datacube:dimension"];
-  const value_descriptions = dimension.value_descriptions || [];
+  const variables = link["variables"];
+  const key = Object.keys(variables)[0];
+  const variable = variables[key];
+  const value_descriptions = variable.value_descriptions || [];

  const listContainerHTML = `
      <div class="item-list-container">
-        <label class="list-label">Select one or more values:</label>
        <div class="scrollable-list">
-          ${dimension.values
+          ${variable.enum
            .map((value, index) => {
              const labelText = value_descriptions[index]
                ? `${value} - ${value_descriptions[index]}`
@ -201,7 +186,7 @@ function renderCheckboxList(link) {
                <div class="checkbox-container">
                  <label class="checkbox-label">
                  <input type="checkbox" class="item-checkbox" value="${value}" ${
-                dimension.values.length === 1 ? "checked" : ""
+                variable.enum.length === 1 ? "checked" : ""
              }>
                  ${labelText}
                  </label>
@ -268,8 +253,10 @@ function renderRawSTACResponse(catalog) {
  itemDetails.textContent = JSON.stringify(just_stac, null, 2);

  const debug_container = document.getElementById("debug");
-  // create new object without debug key
  debug_container.textContent = JSON.stringify(catalog.debug, null, 2);
+
+  const qube_container = document.getElementById("qube");
+  qube_container.innerHTML = catalog.debug.qube;
 }

 // Fetch STAC catalog and display items
@ -293,6 +280,7 @@ async function fetchCatalog(request, stacUrl) {
    // Highlight the request and raw STAC
    hljs.highlightElement(document.getElementById("raw-stac"));
    hljs.highlightElement(document.getElementById("debug"));
+    hljs.highlightElement(document.getElementById("example-python"));
  } catch (error) {
    console.error("Error fetching STAC catalog:", error);
  }
--- a/stac_server/static/qube_styles.css
+++ b/stac_server/static/qube_styles.css
@ -0,0 +1,50 @@
+pre#qube {
+    font-family: monospace;
+    white-space: pre;
+    font-family: SFMono-Regular,Menlo,Monaco,Consolas,Liberation Mono,Courier New,Courier,monospace;
+    font-size: 12px;
+    line-height: 1.4;
+
+    details {
+        margin-left: 0;
+    }
+
+    .qubed-level a {
+        margin-left: 10px;
+        text-decoration: none;
+    }
+
+    summary {
+        list-style: none;
+        cursor: pointer;
+        text-overflow: ellipsis;
+        overflow: hidden;
+        text-wrap: nowrap;
+        display: block;
+    }
+
+    span.qubed-node:hover {
+        background-color: #f0f0f0;
+    }
+
+    details > summary::after {
+        content: ' ▲';
+    }
+
+    details:not([open]) > summary::after {
+        content: " ▼";
+    }
+
+    .qubed-level {
+        text-overflow: ellipsis;
+        overflow: hidden;
+        text-wrap: nowrap;
+        display: block;
+    }
+
+    summary::-webkit-details-marker {
+      display: none;
+      content: "";
+    }
+
+}
--- a/web_query_builder/static/styles.css
+++ b/web_query_builder/static/styles.css
@ -2,6 +2,9 @@ html,
 body {
    min-height: 100vh;
    height: 100%;
+
+    --accent-color: #003399;
+    --background-grey: #f4f4f4;
 }

 body {
@ -23,7 +26,7 @@ body {
    width: 30%;
    padding: 10px;
    overflow-y: scroll;
-    background-color: #f4f4f4;
+    background-color: var(--background-grey);
    border-right: 1px solid #ddd;
 }

@ -45,7 +48,9 @@ body {
 }

 .sidebar-header button {
-    width: 10em;
+    width: 7em;
+    height: 2em;
+    padding: 0;
 }

 canvas {
@ -63,6 +68,7 @@ canvas {
    margin-bottom: 10px;
    border-radius: 5px;
    transition: background-color 0.2s ease;
+    box-shadow: 0 2px 5px rgba(0, 0, 0, 0.1);
 }

 .item-title {
@ -91,10 +97,8 @@ canvas {
 }

 .item.selected {
-    background-color: #d4e9ff;
-    /* Lighter blue for selection */
-    border-color: #003399;
-    /* Keep the original ECMWF blue for the border */
+    background-color: var(--background-grey);
+    border-color: var(--accent-color);
 }

 summary h2 {
@ -117,7 +121,7 @@ button {
    /* Padding around button text */
    margin: 0 5px;
    /* Margin between buttons */
-    background-color: #003399;
+    background-color: var(--accent-color);
    /* ECMWF blue */
    color: white;
    /* White text color */
@ -138,7 +142,6 @@ button:hover {

 .item-list-container {
    margin-top: 20px;
-    margin-bottom: 20px;
 }

 .scrollable-list {
@ -148,7 +151,6 @@ button:hover {
    border: 1px solid #ccc;
    border-radius: 4px;
    background-color: #fff;
-    box-shadow: 0 2px 5px rgba(0, 0, 0, 0.1);
 }

 .checkbox-container {
@ -168,14 +170,14 @@ button:hover {
 }

 .checkbox-container:hover .checkbox-label {
-    color: #003399;
+    color: var(--accent-color);
 }

 .list-label {
    font-weight: bold;
    margin-bottom: 0.5em;
    display: block;
-    color: #003399;
+    color: var(--accent-color);
 }

 span.key,
@ -209,3 +211,7 @@ span.value:hover {
        width: 100%;
    }
 }
+
+details h2 {
+    font-size: medium;
+}
--- a/web_query_builder/templates/index.html
+++ b/web_query_builder/templates/index.html
@ -5,6 +5,7 @@
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>ECMWF DestinE STAC Viewer</title>
    <link rel="stylesheet" href="/static/styles.css" />
+    <link rel="stylesheet" href="/static/qube_styles.css" />
    <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/styles/github.min.css">
    <script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/highlight.min.js"></script>
    <script src="https://cdnjs.cloudflare.com/ajax/libs/highlight.js/11.9.0/languages/json.min.js"></script>
@ -38,8 +39,25 @@
 }
            </code></pre>

+            <!-- Container to show the current tree -->
+            <h2>Currently Selected Tree</h2></summary>
+            <p>This shows the data <a href="https://qubed.readthedocs.io/en/latest/quickstart.html">qube</a> that matches with the current query. The leaves are the next set if available selections you can make. </p>
+            <pre id = "qube"></pre>
+
+            <details>
+                <summary><h2>Example Qubed Code</h2></summary>
+                See the <a href="https://qubed.readthedocs.io/en/latest/">Qubed documentation</a> for more details.
+                <pre><code id="example-python" class="language-python">
+# pip install qubed requests
+import requests
+from qubed import Qube
+qube = Qube.from_json(requests.get("{{ api_url }}select/climate-dt/?{{request.url.query}}").json())
+qube.print()
+                </code></pre>
+            </details>
+
            <!-- Container fo the raw STAC response -->
-            <details open>
+            <details>
                <summary><h2>Raw STAC Response</h2></summary>
                <p>See the <a href="https://github.com/ecmwf-projects/catalogs/blob/main/structured_stac.md">STAC Extension Proposal</a> for more details on the format.</p>
                <pre class="json-pre"><code id="raw-stac" class="language-json"></code></pre>
@ -54,7 +72,7 @@
    </div>

    <script>
-        window.API_URL = "{{ api_url }}";
+        window.API_URL = "{{ api_url }}stac/climate-dt/";
    </script>
    <script src="/static/app.js"></script>
 </body>
--- a/structured_stac.md
+++ b/structured_stac.md
@ -7,18 +7,39 @@
 - **Extension [Maturity Classification](https://github.com/radiantearth/stac-spec/tree/master/extensions/README.md#extension-maturity):** Proposal
 - **Owner**: @TomHodson

-This STAC extension allows for represention of [generalised datacubes][gen_datacubes].
+This STAC extension borrows the [Draft OGC Records API](https://docs.ogc.org/DRAFTS/20-004.html), specifically the [templated links section](https://docs.ogc.org/DRAFTS/20-004.html#sc_templated_links_with_variables) to give STAC the ability to index very large datasets that conform to a generalised datacube model.

-A datacube has a fixed set of dimensions `[a, b, c..]` , each of which have a fixed span `{a: ["temp","rainfall"], b : [1-7], c:[True, False]}` such that we can access data by indexing, i.e providing a value for each axis, `a="rainfall", b=1, ...`.  A generalised datacubes allow the dimensions to change during indexing, so choosing `a="rainfall"` might yield a different set of axes from `a="temp"`.
+A typical datacube has a fixed set of dimensions `[a, b, c..]` , each of which have a fixed span `{a: ["temp","rainfall"], b : [1-7], c:[True, False]}` such that we can access data by indexing, i.e providing a value for each axis, `a="rainfall", b=1, ...`.  A generalized datacube, by our defintion, allow the dimensions to change during indexing, so choosing `a="rainfall"` might yield a different set of axes from `a="temp"`.

-The [STAC Datacube][datacube_extension] extension serves the needs of datacubes that appear in STAC as Items or Collections, i.e as leaves in the tree. This extension instead focussing on allowing STAC to serve as an interface to dynamically explore the branches of generalised datacubes. It does this by adding additional metadata to the children of Catalog entries.
+The [STAC Datacube][datacube_extension] extension serves the needs of datacubes that appear in STAC as Items or Collections, i.e as leaves in the tree. This extension instead focussing on allowing STAC to serve as an interface to dynamically explore the branches of generalised datacubes. It does this by adding additional metadata from the OGC Records standard to the children of Catalog entries.

-We take the *Dimension Objects* defined by the [Datacube Extension][datacube_extension] and add them to [Link objects][link_objects] under the key `generalized_datacube:dimension`. This enables a single Link Object to represent a whole axis and its allowed values. Since `href` must now be constructed dynamically, we rempve it and add a `generalized_datacube:href_template` attribute to communicate how to construct the URLs corresponding to particular choice of value or values.
+In practice, what this proposal does is:

-In order to support more complex slicing operations in which multiple indices may be selected for a given dimensions we also add additional optional attributes to all *Dimension Objects*, these are:
+1. For child items that represent many distinct children, replace `"links":` with `"linkTemplates":` in the Catalog entry. (Following the example of the OGC Records API.)
+2. For each `rel: Child` object in `linkTemplates`:

-* `optional` : bool whether this dimension can be skipped.
-* `multiple` : boo wether multiple values can be selected for this key.
+    a. Add a `variables` key following the OGC Records API whose values is a dictionary with entries like 
+
+    ```json
+    "format": {
+            "type": "string",
+            "enum": [
+                "application/vnd.google-earth.kml+xml",
+                "application/vnd.google-earth.kmz",
+                "image/png",
+                "image/jpeg",
+                "image/gif",
+                "image/png; mode=8bit",
+                "application/x-pdf",
+                "image/svg+xml",
+                "image/tiff"
+            ]
+            }
+    ```
+
+    b. Add a "uriTemplate" key that specifies how to contruct the resulting URL: i.e `http://hostname.tld/app/index.html?class=od&format={format}`
+
+This enables a child object to represent a whole axis and its allowed values. Since `href` must now be constructed dynamically, we rempve it and add a `generalized_datacube:href_template` attribute to communicate how to construct the URLs corresponding to particular choice of value or values.

 [gen_datacubes]: https://github.com/ecmwf/datacube-spec
 [link_objects]: https://github.com/radiantearth/stac-spec/blob/master/commons/links.md#link-object
@ -34,19 +55,20 @@ A typical `Catalog` entry with this extension:
  "id": "rainfall",
  "stac_version": "1.0.0",
  "description": "ECMWF's Operational Data Archive",
-  "links": [
+  "linkTemplates": [
    {
-      "title": "Expver - Experiment Version",
-      "generalized_datacube:href_template": "http://136.156.129.226/app/index.html?class=od&expver={}",
      "rel": "child",
+      "title": "Expver - Experiment Version",
+      "uriTemplate": "http://hostname.tld/app/index.html?class=od&expver={expver}",
      "type": "application/json",
-      "generalized_datacube:dimension" : {
-        "type" : "enum",
-        "description": "Experiment version, 0001 selects operational data.",
-        "values" : ["0001", "xxxx"],
-        "value_descriptions" : ["Operational Data", "Experimental Data"],
-        "optional" : false,
-        "multiple": true,
+      "variables" : {
+        "expver" : {
+            "description": "Experiment version, 0001 selects operational data.",
+            "type" : "string",
+            "enum" : ["0001", "xxxx"],
+            "value_descriptions" : ["Operational Data", "Experimental Data"],
+            "optional" : false,
+        }
      }
      ""

@ -70,120 +92,19 @@ The fields in the table below can be used in these parts of STAC documents:
 - [ ] Assets (for both Collections and Items, incl. Item Asset Definitions in Collections)
 - [x] Links

-| Field Name           | Type                      | Description                                              |
-| -------------------- | ------------------------- | -------------------------------------------------------- |
-| axis:dimension       | Dimension Object          | Whether the axis is an enum, date range, time range etc  |
-| axis:href_template   | string                    | Whether the axis is an enum, date range, time range etc  |
+| Field Name           | Type                      | Description                                                                                                           |
+| -------------------- | ------------------------- | --------------------------------------------------------------------------------------------------------------------- |
+| uriTemplate          | URI Template              | Of the form "http://hostname.tld/app/index.html?class=od&expver={expver}", follows OGC Records Spec for uriTemplates  |
+| variables            |                           |                                      |




 ### Additional Field Information

-#### axis:dimension
+#### uriTemplate
+Todo


-
-### Dimension Object
-
-The dimension object reuses all those [defined by the datacube extension](https://github.com/stac-extensions/datacube#dimension-object), currently those are Horizontal Spatial Raster, Vertical Spatial, Temporal Dimension, Spatial Vector Dimension, Additional Dimension. They are reproduced below for reference.
-
-These dimension objects are defined in addition:
-
-### Enum Dimension Object
-
-
-| Field Name       | Type              | Description                                                  |
-| ---------------- | ----------------- | ------------------------------------------------------------ |
-| type             | string            | **REQUIRED.**  `enum`.                                       |
-| description      | string            | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| values           | \[number\|string] | An ordered list of all values, especially useful for [nominal](https://en.wikipedia.org/wiki/Level_of_measurement#Nominal_level) values. |
-| value_descriptions           | \[string] | Optionally provide a human readable description for each value.  Useful if the values are codes that have defined meanings. |
-| step             | number\|null      | If the dimension consists of [interval](https://en.wikipedia.org/wiki/Level_of_measurement#Interval_scale) values, the space between the values. Use `null` for irregularly spaced steps. |
-| unit             | string            | The unit of measurement for the data, preferably compliant to [UDUNITS-2](https://ncics.org/portfolio/other-resources/udunits2/) units (singular). |
-| reference_system | string            | The reference system for the data.                           |
-
-An Enum Dimension Object MUST specify `values`.
-
-Dimension objects degined by the datacube extension:
-
-### Horizontal Spatial Raster Dimension Object
-
-A spatial raster dimension in one of the horizontal (x or y) directions.
-
-| Field Name       | Type           | Description                                                  |
-| ---------------- | -------------- | ------------------------------------------------------------ |
-| type             | string         | **REQUIRED.** Type of the dimension, always `spatial`.       |
-| axis             | string         | **REQUIRED.** Axis of the spatial raster dimension (`x`, `y`).      |
-| description      | string         | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| extent           | \[number]      | **REQUIRED.** Extent (lower and upper bounds) of the dimension as two-element array. Open intervals with `null` are not allowed. |
-| values           | \[number]      | Optionally, an ordered list of all values.                   |
-| step             | number\|null   | The space between the values. Use `null` for irregularly spaced steps. |
-| reference_system | string\|number\|object | The spatial reference system for the data, specified as [numerical EPSG code](http://www.epsg-registry.org/), [WKT2 (ISO 19162) string](http://docs.opengeospatial.org/is/18-010r7/18-010r7.html) or [PROJJSON object](https://proj.org/specifications/projjson.html). Defaults to EPSG code 4326. |
-
-### Vertical Spatial Dimension Object
-
-A spatial dimension in vertical (z) direction.
-
-| Field Name       | Type             | Description                                                  |
-| ---------------- | ---------------- | ------------------------------------------------------------ |
-| type             | string           | **REQUIRED.** Type of the dimension, always `spatial`.       |
-| axis             | string           | **REQUIRED.** Axis of the spatial dimension, always `z`.     |
-| description      | string           | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| extent           | \[number\|null\]   | If the dimension consists of [ordinal](https://en.wikipedia.org/wiki/Level_of_measurement#Ordinal_scale) values, the extent (lower and upper bounds) of the values as two-element array. Use `null` for open intervals. |
-| values           | \[number\|string\] | An ordered list of all values, especially useful for [nominal](https://en.wikipedia.org/wiki/Level_of_measurement#Nominal_level) values. |
-| step             | number\|null     | If the dimension consists of [interval](https://en.wikipedia.org/wiki/Level_of_measurement#Interval_scale) values, the space between the values. Use `null` for irregularly spaced steps. |
-| unit             | string           | The unit of measurement for the data, preferably compliant to [UDUNITS-2](https://ncics.org/portfolio/other-resources/udunits2/) units (singular). |
-| reference_system | string\|number\|object | The spatial reference system for the data, specified as [numerical EPSG code](http://www.epsg-registry.org/), [WKT2 (ISO 19162) string](http://docs.opengeospatial.org/is/18-010r7/18-010r7.html) or [PROJJSON object](https://proj.org/specifications/projjson.html). Defaults to EPSG code 4326. |
-
-A Vertical Spatial Dimension Object MUST specify an `extent` or `values`. It MAY specify both.
-
-### Temporal Dimension Object
-
-A temporal dimension based on the ISO 8601 standard. The temporal reference system for the data is expected to be ISO 8601 compliant
-(Gregorian calendar / UTC). Data not compliant with ISO 8601 can be represented as an *Additional Dimension Object* with `type` set to `temporal`.
-
-| Field Name | Type            | Description                                                  |
-| ---------- | --------------- | ------------------------------------------------------------ |
-| type       | string          | **REQUIRED.** Type of the dimension, always `temporal`.      |
-| description | string         | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| extent     | \[string\|null] | **REQUIRED.** Extent (lower and upper bounds) of the dimension as two-element array. The dates and/or times must be strings compliant to [ISO 8601](https://en.wikipedia.org/wiki/ISO_8601). `null` is allowed for open date ranges. |
-| values     | \[string]       | If the dimension consists of an ordered list of specific values they can be listed here. The dates and/or times must be strings compliant to [ISO 8601](https://en.wikipedia.org/wiki/ISO_8601). |
-| step       | string\|null    | The space between the temporal instances as [ISO 8601 duration](https://en.wikipedia.org/wiki/ISO_8601#Durations), e.g. `P1D`. Use `null` for irregularly spaced steps. |
-
-### Spatial Vector Dimension Object
-
-A vector dimension that defines a spatial dimension based on geometries.
-
-| Field Name       | Type           | Description                                                  |
-| ---------------- | -------------- | ------------------------------------------------------------ |
-| type             | string         | **REQUIRED.** Type of the dimension, always `geometry`.    |
-| axes             | \[string]      | Axes of the vector dimension as an ordered set of `x`, `y` and `z`. Defaults to `x` and `y`. |
-| description      | string         | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| bbox             | \[number]      | **REQUIRED.** A single bounding box of the geometries as defined for [STAC Collections](https://github.com/radiantearth/stac-spec/blob/master/collection-spec/collection-spec.md#spatial-extent-object), but not nested. |
-| values           | \[string\]     | Optionally, a representation of the geometries. This could be a list of WKT strings or other identifiers. |
-| geometry_types   | \[[GeoJSON Types](https://www.rfc-editor.org/rfc/rfc7946#section-1.4)] | A set of geometry types. If not present, mixed geometry types must be assumed. |
-| reference_system | string\|number\|object | The spatial reference system for the data, specified as [numerical EPSG code](http://www.epsg-registry.org/), [WKT2 (ISO 19162) string](http://docs.opengeospatial.org/is/18-010r7/18-010r7.html) or [PROJJSON object](https://proj.org/specifications/projjson.html). Defaults to EPSG code 4326. |
-
-For a general explanation what a vector datacube and a vector dimension is, please read the article "[Vector Data Cubes](https://r-spatial.org/r/2022/09/12/vdc.html)".
-
-### Additional Dimension Object
-
-An additional dimension that is not `spatial`, but may be `temporal` if the data is not compliant with ISO 8601 (see below).
-
-| Field Name       | Type              | Description                                                  |
-| ---------------- | ----------------- | ------------------------------------------------------------ |
-| type             | string            | **REQUIRED.** Custom type of the dimension, never `spatial` or `geometry`. |
-| description      | string            | Detailed multi-line description to explain the dimension. [CommonMark 0.29](http://commonmark.org/) syntax MAY be used for rich text representation. |
-| extent           | \[number\|null]   | If the dimension consists of [ordinal](https://en.wikipedia.org/wiki/Level_of_measurement#Ordinal_scale) values, the extent (lower and upper bounds) of the values as two-element array. Use `null` for open intervals. |
-| values           | \[number\|string] | An ordered list of all values, especially useful for [nominal](https://en.wikipedia.org/wiki/Level_of_measurement#Nominal_level) values. |
-| step             | number\|null      | If the dimension consists of [interval](https://en.wikipedia.org/wiki/Level_of_measurement#Interval_scale) values, the space between the values. Use `null` for irregularly spaced steps. |
-| unit             | string            | The unit of measurement for the data, preferably compliant to [UDUNITS-2](https://ncics.org/portfolio/other-resources/udunits2/) units (singular). |
-| reference_system | string            | The reference system for the data.                           |
-
-An Additional Dimension Object MUST specify an `extent` or `values`. It MAY specify both.
-
-Note on "Additional Dimension" with type `temporal`:
-You can distinguish the "Temporal Dimension" from an "Additional Dimension" by checking whether the extent exists and contains strings.
-So if the `type` equals `temporal` and `extent` is an array of strings/null, then you have a "Temporal Dimension",
-otherwise you have an "Additional Dimension".
+#### variables
+Todo
--- a/test_scripts/rust.py
+++ b/test_scripts/rust.py
@ -0,0 +1,81 @@
+from __future__ import annotations
+
+from datetime import datetime
+from typing import Sequence
+
+from qubed.rust import Qube as rsQube
+
+# q = pyQube.from_tree("""
+# root, class=d1
+# ├── dataset=another-value, generation=1/2/3
+# └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+# """)
+# json_str = json.dumps(q.to_json())
+# rust_qube = Qube.from_json(json_str)
+# # print(repr(rust_qube))
+
+# # print(json_str)
+
+# expected = """root, class=d1
+# ├── dataset=another-value, generation=1/2/3
+# └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+# """
+# assert repr(rust_qube) == expected
+# # print(rs_qube._repr_html_())
+
+# print(q | q)
+
+value = str | int | float | datetime
+
+
+class Qube(rsQube):
+    @classmethod
+    def empty(cls):
+        q = cls()
+        print(f"empty called {cls = } {q = }")
+        return q
+
+    @classmethod
+    def from_datacube(cls, datacube: dict[str, value | Sequence[value]]) -> Qube:
+        qube = cls.empty()
+        (key, values), *key_vals = list(datacube.items())
+        node = qube.add_node(qube.root, key, values)
+        for key, values in key_vals:
+            node = qube.add_node(parent=node, key=key, values=values)
+
+        return qube
+
+    @classmethod
+    def from_dict(cls, d: dict) -> Qube:
+        q = cls.empty()
+
+        def from_dict(parent, d: dict):
+            for k, children in d.items():
+                key, values = k.split("=")
+                values = values.split("/")
+
+                node = q.add_node(
+                    parent=parent,
+                    key=key,
+                    values=values,
+                )
+                from_dict(parent=node, d=children)
+
+        from_dict(q.root, d)
+        return q
+
+
+q = Qube.from_datacube({"a": ["4"], "b": "test", "c": ["1", "2", "3"]})
+
+print(q)
+print(repr(q))
+
+q = Qube.from_dict(
+    {
+        "a=2/3": {"b=1": {}},
+        "a2=a/b": {"b2=1/2": {}},
+    }
+)
+
+print(q)
+print(repr(q))
--- a/test_scripts/update_dts.py
+++ b/test_scripts/update_dts.py
@ -0,0 +1,99 @@
+# Example script for ingesting data from an fdb into a qube
+# Notes
+# Uses fdb --compact
+# Splits by data in order to avoid out of memory problems with fdb --compact
+# Does a bit of processing like removing "year" and "month" keys
+# Might want to add datatypes and reordering of keys there too
+
+import json
+import subprocess
+from datetime import datetime, timedelta
+from time import time
+
+import psutil
+from qubed import Qube
+from tqdm import tqdm
+import requests
+
+process = psutil.Process()
+
+CHUNK_SIZE = timedelta(days=60)
+FILEPATH = "tests/example_qubes/full_dt.json"
+API = "https://qubed.lumi.apps.dte.destination-earth.eu/api/v1"
+
+with open("config/api.secret", "r") as f:
+    secret = f.read()
+
+def ecmwf_date(d):
+    return d.strftime("%Y%m%d")
+
+
+start_date = datetime.now() - timedelta(days=120)
+# start_date = datetime(1990, 1, 1)
+# end_date = datetime.now()
+end_date = datetime(2026, 1, 1)
+
+current_span = [end_date - CHUNK_SIZE, end_date]
+
+try:
+    qube = Qube.load(FILEPATH)
+except:
+    qube = Qube.empty()
+
+while current_span[0] > start_date:
+    for config in ["config/config-climate-dt.yaml", "config/config-extremes-dt.yaml"]:
+        t0 = time()
+        start, end = map(ecmwf_date, current_span)
+        print(f"Doing {config} {current_span[0].date()} - {current_span[1].date()}")
+        print(f"Current memory usage: {process.memory_info().rss / 1e9:.2g}GB")
+        print(f"{qube.n_nodes = }, {qube.n_leaves = },")
+
+        subqube = Qube.empty()
+        command = [
+            f"fdb list --compact --config {config} --minimum-keys=date class=d1,date={start}/{end}"
+        ]
+        try:
+            p = subprocess.run(
+                command,
+                text=True,
+                shell=True,
+                stderr=subprocess.PIPE,
+                stdout=subprocess.PIPE,
+                check=True,
+            )
+        except Exception as e:
+            print(f"Failed for {current_span} {e}")
+            continue
+
+        print("Got compact list")
+        for i, line in tqdm(enumerate(list(p.stdout.split("\n")))):
+            if not line.startswith("retrieve,class="):
+                continue
+
+            def split(t):
+                return t[0], t[1].split("/")
+
+            # Could do datatypes here
+            request = dict(split(v.split("=")) for v in line.strip().split(",")[1:])
+            request.pop("year", None)
+            request.pop("month", None)
+            # Could do things like date = year + month + day
+            q = Qube.from_datacube(request)
+            subqube = subqube | q
+        print("added to qube")
+
+        qube = qube | subqube
+        subqube.print(depth=2)
+        print(f"{subqube.n_nodes = }, {subqube.n_leaves = },")
+
+        requests.post(
+                API + "/union/climate-dt/",
+                headers = {"Authorization" : f"Bearer {secret}"},
+                json = subqube.to_json())
+
+        current_span = [current_span[0] - CHUNK_SIZE, current_span[0]]
+        print(
+            f"Did that taking {(time() - t0) / CHUNK_SIZE.days:2g} seconds per day ingested, total {(time() - t0):2g}s"
+        )
+    with open(FILEPATH, "w") as f:
+        json.dump(qube.to_json(), f)
--- a/tests/example_qubes/cads.json
+++ b/tests/example_qubes/cads.json
--- a/tests/example_qubes/climate_dt.json
+++ b/tests/example_qubes/climate_dt.json
--- a/tests/example_qubes/climate_dt_old_schema.json
+++ b/tests/example_qubes/climate_dt_old_schema.json
--- a/tests/example_qubes/extremes_dt.json
+++ b/tests/example_qubes/extremes_dt.json
--- a/tests/example_qubes/od.json
+++ b/tests/example_qubes/od.json
--- a/tests/test_basic_operations.py
+++ b/tests/test_basic_operations.py
@ -1,36 +1,40 @@
 from qubed import Qube

-d = {
-    "class=od": {
-        "expver=0001": {"param=1": {}, "param=2": {}},
-        "expver=0002": {"param=1": {}, "param=2": {}},
-    },
-    "class=rd": {
-        "expver=0001": {"param=1": {}, "param=2": {}, "param=3": {}},
-        "expver=0002": {"param=1": {}, "param=2": {}},
-    },
-}
-q = Qube.from_dict(d)
-
-
-def test_eq():
-    r = Qube.from_dict(d)
-    assert q == r
+q = Qube.from_tree("""
+root
+├── class=od
+│   ├── expver=0001
+│   │   ├── param=1
+│   │   └── param=2
+│   └── expver=0002
+│       ├── param=1
+│       └── param=2
+└── class=rd
+    ├── expver=0001
+    │   ├── param=1
+    │   ├── param=2
+    │   └── param=3
+    └── expver=0002
+        ├── param=1
+        └── param=2
+""")


 def test_getitem():
-    assert q["class", "od"] == Qube.from_dict(
-        {
-            "expver=0001": {"param=1": {}, "param=2": {}},
-            "expver=0002": {"param=1": {}, "param=2": {}},
-        }
-    )
-    assert q["class", "od"]["expver", "0001"] == Qube.from_dict(
-        {
-            "param=1": {},
-            "param=2": {},
-        }
-    )
+    assert q["class", "od"] == Qube.from_tree("""
+root
+├── expver=0001
+│   ├── param=1
+│   └── param=2
+└── expver=0002
+    ├── param=1
+    └── param=2
+""")
+
+    assert q["class", "od"]["expver", "0001"] == Qube.from_tree("""
+root
+├── param=1
+└── param=2""")


 def test_n_leaves():
--- a/tests/test_compression.py
+++ b/tests/test_compression.py
@ -15,19 +15,62 @@ def test_smoke():
        }
    )

-    # root
-    # ├── class=od, expver=0001/0002, param=1/2
-    # └── class=rd
-    #     ├── expver=0001, param=1/2/3
-    #     └── expver=0002, param=1/2
-    ct = Qube.from_dict(
+    ct = Qube.from_tree("""
+    root
+    ├── class=od, expver=0001/0002, param=1/2
+    └── class=rd
+        ├── expver=0001, param=1/2/3
+        └── expver=0002, param=1/2
+                        """)
+
+    assert q.compress() == ct
+
+
+def test_2():
+    qube = Qube.from_dict(
        {
-            "class=od": {"expver=0001/0002": {"param=1/2": {}}},
-            "class=rd": {
-                "expver=0001": {"param=1/2/3": {}},
-                "expver=0002": {"param=1/2": {}},
-            },
+            "class=d1": {
+                "generation=1": {
+                    "date=20240728": {"time=0600": {"param=8/78/79": {}}},
+                    "date=20240828": {"time=0600": {"param=8/78/79": {}}},
+                    "date=20240928": {"time=0600": {"param=8/78/79": {}}},
+                }
+            }
        }
    )

-    assert q.compress() == ct
+    target = Qube.from_datacube(
+        {
+            "class": "d1",
+            "generation": "1",
+            "date": ["20240728", "20240828", "20240928"],
+            "time": "0600",
+            "param": ["8", "78", "79"],
+        }
+    )
+    assert qube.compress() == target
+
+
+def test_removal_compression():
+    qube = Qube.from_dict(
+        {
+            "class=d1": {
+                "generation=1": {
+                    "month=07": {"date=20240728": {"time=0600": {"param=8/78/79": {}}}},
+                    "month=08": {"date=20240828": {"time=0600": {"param=8/78/79": {}}}},
+                    "month=09": {"date=20240928": {"time=0600": {"param=8/78/79": {}}}},
+                }
+            }
+        }
+    )
+
+    target = Qube.from_datacube(
+        {
+            "class": "d1",
+            "generation": "1",
+            "date": ["20240728", "20240828", "20240928"],
+            "time": "0600",
+            "param": ["8", "78", "79"],
+        }
+    )
+    assert qube.remove_by_key(["month"]) == target
--- a/tests/test_conversions.py
+++ b/tests/test_conversions.py
@ -2,7 +2,7 @@ from qubed import Qube


 def test_json_round_trip():
-    u = Qube.from_dict(
+    from_dict = Qube.from_dict(
        {
            "class=d1": {
                "dataset=climate-dt/weather-dt": {
@ -14,5 +14,54 @@ def test_json_round_trip():
            }
        }
    )
-    json = u.to_json()
-    assert Qube.from_json(json) == u
+
+    from_tree = Qube.from_tree("""
+    root, class=d1
+    ├── dataset=another-value, generation=1/2/3
+    └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+    """)
+
+    from_json = Qube.from_json(
+        {
+            "key": "root",
+            "values": ["root"],
+            "metadata": {},
+            "children": [
+                {
+                    "key": "class",
+                    "values": ["d1"],
+                    "metadata": {},
+                    "children": [
+                        {
+                            "key": "dataset",
+                            "values": ["another-value"],
+                            "metadata": {},
+                            "children": [
+                                {
+                                    "key": "generation",
+                                    "values": ["1", "2", "3"],
+                                    "metadata": {},
+                                    "children": [],
+                                }
+                            ],
+                        },
+                        {
+                            "key": "dataset",
+                            "values": ["climate-dt", "weather-dt"],
+                            "metadata": {},
+                            "children": [
+                                {
+                                    "key": "generation",
+                                    "values": ["1", "2", "3", "4"],
+                                    "metadata": {},
+                                    "children": [],
+                                }
+                            ],
+                        },
+                    ],
+                }
+            ],
+        }
+    )
+    assert from_tree == from_json
+    assert from_tree == from_dict
--- a/tests/test_creation.py
+++ b/tests/test_creation.py
@ -0,0 +1,56 @@
+from qubed import Qube
+
+
+def test_simple():
+    q = Qube.from_tree("""
+    root, frequency=6:00:00
+    ├── levtype=pl, param=t, levelist=850, threshold=-2/-4/-8/2/4/8
+    └── levtype=sfc
+        ├── param=10u/10v, threshold=10/15
+        ├── param=2t, threshold=273.15
+        └── param=tp, threshold=0.1/1/10/100/20/25/5/50
+    """)
+
+    r = Qube.from_dict(
+        {
+            "frequency=6:00:00": {
+                "levtype=pl": {
+                    "param=t": {"levelist=850": {"threshold=-8/-4/-2/2/4/8": {}}}
+                },
+                "levtype=sfc": {
+                    "param=10u/10v": {"threshold=10/15": {}},
+                    "param=2t": {"threshold=273.15": {}},
+                    "param=tp": {"threshold=0.1/1/5/10/20/25/50/100": {}},
+                },
+            },
+        }
+    )
+
+    assert q == r
+
+
+def test_simple_2():
+    models = Qube.from_datacube(
+        dict(
+            param="10u/10v/2d/2t/cp/msl/skt/sp/tcw/tp".split("/"),
+            threshold="*",
+            levtype="sfc",
+            frequency="6:00:00",
+        )
+    ) | Qube.from_datacube(
+        dict(
+            param="q/t/u/v/w/z".split("/"),
+            threshold="*",
+            levtype="pl",
+            level="50/100/150/200/250/300/400/500/600/700/850".split("/"),
+            frequency="6:00:00",
+        )
+    )
+
+    models2 = Qube.from_tree("""
+    models
+    ├── param=10u/10v/2d/2t/cp/msl/skt/sp/tcw/tp, threshold=*, levtype=sfc, frequency=6:00:00
+    └── param=q/t/u/v/w/z, threshold=*, levtype=pl, level=100/150/200/250/300/400/50/500/600/700/850, frequency=6:00:00
+    """)
+
+    assert models == models2
--- a/tests/test_formatters.py
+++ b/tests/test_formatters.py
@ -20,14 +20,6 @@ root
    └── expver=0002, param=1/2
 """.strip()

-as_html = """
-<details open data-path="root"><summary class="qubed-node">root</summary><span class="qubed-node leaf" data-path="class=od,expver=0001/0002,param=1/2">├── class=od, expver=0001/0002, param=1/2</span><details open data-path="class=rd"><summary class="qubed-node">└── class=rd</summary><span class="qubed-node leaf" data-path="expver=0001,param=1/2/3">    ├── expver=0001, param=1/2/3</span><span class="qubed-node leaf" data-path="expver=0002,param=1/2">    └── expver=0002, param=1/2</span></details></details>
-""".strip()
-

 def test_string():
    assert str(q).strip() == as_string
-
-
-def test_html():
-    assert as_html in q._repr_html_()
--- a/tests/test_iteration.py
+++ b/tests/test_iteration.py
@ -16,3 +16,29 @@ def test_iter_leaves_simple():
    ]

    assert set(make_hashable(q.leaves())) == set(make_hashable(entries))
+
+
+def test_datacubes():
+    q = Qube.from_tree("""
+    root, class=d1
+    ├── date=19920101/19930101/19940101, params=1/2/3
+    └── date=19950101
+        ├── level=1/2/3, params=1/2/3/4
+        └── params=1/2/3/4
+    """)
+    assert len(list(q.datacubes())) == 3
+
+    assert list(q.datacubes()) == [
+        {
+            "class": ["d1"],
+            "date": ["19920101", "19930101", "19940101"],
+            "params": ["1", "2", "3"],
+        },
+        {
+            "class": ["d1"],
+            "date": ["19950101"],
+            "level": ["1", "2", "3"],
+            "params": ["1", "2", "3", "4"],
+        },
+        {"class": ["d1"], "date": ["19950101"], "params": ["1", "2", "3", "4"]},
+    ]
--- a/tests/test_metadata.py
+++ b/tests/test_metadata.py
@ -0,0 +1,98 @@
+# from frozendict import frozendict
+# from qubed import Qube
+
+
+# def make_set(entries):
+#     return set((frozendict(a), frozendict(b)) for a, b in entries)
+
+# def construction():
+#     q = Qube.from_nodes(
+#         {
+#             "class": dict(values=["od", "rd"]),
+#             "expver": dict(values=[1, 2]),
+#             "stream": dict(
+#                 values=["a", "b", "c"], metadata=dict(number=list(range(12)))
+#             ),
+#         }
+#     )
+#     assert make_set(q.leaves_with_metadata()) == make_set([
+#  ({'class': 'od', 'expver': 1, 'stream': 'a'}, {'number': 0}),
+#  ({'class': 'od', 'expver': 1, 'stream': 'b'}, {'number': 1}),
+#  ({'class': 'od', 'expver': 1, 'stream': 'c'}, {'number': 2}),
+#  ({'class': 'od', 'expver': 2, 'stream': 'a'}, {'number': 3}),
+#  ({'class': 'od', 'expver': 2, 'stream': 'b'}, {'number': 4}),
+#  ({'class': 'od', 'expver': 2, 'stream': 'c'}, {'number': 5}),
+#  ({'class': 'rd', 'expver': 1, 'stream': 'a'}, {'number': 6}),
+#  ({'class': 'rd', 'expver': 1, 'stream': 'b'}, {'number': 7}),
+#  ({'class': 'rd', 'expver': 1, 'stream': 'c'}, {'number': 8}),
+#  ({'class': 'rd', 'expver': 2, 'stream': 'a'}, {'number': 9}),
+#  ({'class': 'rd', 'expver': 2, 'stream': 'b'}, {'number': 10}),
+#  ({'class': 'rd', 'expver': 2, 'stream': 'c'}, {'number': 11})])
+
+
+# def test_simple_union():
+#     q = Qube.from_nodes(
+#         {
+#             "class": dict(values=["od", "rd"]),
+#             "expver": dict(values=[1, 2]),
+#             "stream": dict(
+#                 values=["a", "b", "c"], metadata=dict(number=list(range(12)))
+#             ),
+#         }
+#     )
+
+#     r = Qube.from_nodes(
+#         {
+#             "class": dict(values=["xd"]),
+#             "expver": dict(values=[1, 2]),
+#             "stream": dict(
+#                 values=["a", "b", "c"], metadata=dict(number=list(range(12, 18)))
+#             ),
+#         }
+#     )
+
+#     expected_union = Qube.from_nodes(
+#         {
+#             "class": dict(values=["od", "rd", "xd"]),
+#             "expver": dict(values=[1, 2]),
+#             "stream": dict(
+#                 values=["a", "b", "c"], metadata=dict(number=list(range(18)))
+#             ),
+#         }
+#     )
+
+#     union = q | r
+
+#     assert union == expected_union
+#     assert make_set(expected_union.leaves_with_metadata()) == make_set(
+#         union.leaves_with_metadata()
+#     )
+
+
+# def test_construction_from_fdb():
+#     import json
+#     paths = {}
+#     current_path = None
+#     i = 0
+
+#     qube = Qube.empty()
+#     with open("tests/data/climate_dt_paths.json") as f:
+#         for l in f.readlines():
+#             i += 1
+#             j = json.loads(l)
+#             if "type" in j and j["type"] == "path":
+#                 paths[j["i"]] = j["path"]
+
+#             else:
+#                 request = j.pop("keys")
+#                 metadata = j
+#                 # print(request, metadata)
+
+#                 q = Qube.from_nodes({
+#                     key : dict(values = [value])
+#                     for key, value in request.items()
+#                 }).add_metadata(**metadata)
+
+#                 qube = qube | q
+
+#                 if i > 100: break
--- a/tests/test_protobuf.py
+++ b/tests/test_protobuf.py
@ -0,0 +1,12 @@
+from qubed import Qube
+
+
+def test_protobuf_simple():
+    q = Qube.from_tree("""
+    root, class=d1
+    ├── dataset=another-value, generation=1/2/3
+    └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+    """)
+    wire = q.to_protobuf()
+    round_trip = Qube.from_protobuf(wire)
+    assert round_trip == q
--- a/tests/test_rust.py
+++ b/tests/test_rust.py
@ -1,5 +1,21 @@
-from qubed.rust import hello
+from __future__ import annotations


-def test_hello():
-    assert hello("World") == "Hello, World!"
+from qubed.rust import Qube as Qube
+
+
+# def test_from_json():
+#     q = pyQube.from_tree("""
+#     root, class=d1
+#     ├── dataset=another-value, generation=1/2/3
+#     └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+#     """)
+#     json_str = json.dumps(q.to_json())
+#     rust_qube = Qube.from_json(json_str)
+#     print(repr(rust_qube))
+
+#     expected = """root, class=d1
+#     ├── dataset=another-value, generation=1/2/3
+#     └── dataset=climate-dt/weather-dt, generation=1/2/3/4
+#     """
+#     assert repr(rust_qube) == expected
--- a/tests/test_selection.py
+++ b/tests/test_selection.py
@ -0,0 +1,56 @@
+from qubed import Qube
+
+q = Qube.from_tree("""
+root
+├── class=od, expver=0001/0002, param=1/2
+└── class=rd, param=1/2/3
+""")
+
+
+def test_consumption():
+    assert q.select({"expver": "0001"}, consume=True) == Qube.from_tree(
+        "root, class=od, expver=0001, param=1/2"
+    )
+
+
+def test_consumption_off():
+    expected = Qube.from_tree("""
+root
+├── class=od, expver=0001, param=1/2
+└── class=rd, param=1/2/3
+""")
+    assert q.select({"expver": "0001"}, consume=False) == expected
+
+
+def test_function_input_to_select():
+    q = Qube.from_tree("""
+    root, frequency=6:00:00
+    ├── levtype=pl, param=t, levelist=850, threshold=-2/-4/-8/2/4/8
+    └── levtype=sfc
+        ├── param=10u/10v, threshold=10/15
+        ├── param=2t, threshold=273.15
+        └── param=tp, threshold=0.1/1/10/100/20/25/5/50
+    """).convert_dtypes(
+        {
+            "threshold": float,
+        }
+    )
+
+    r = q.select(
+        {
+            "threshold": lambda t: t > 5,
+        }
+    )
+
+    assert r == Qube.from_tree("""
+    root, frequency=6:00:00
+    ├── levtype=pl, param=t, levelist=850, threshold=8
+    └── levtype=sfc
+        ├── param=10u/10v, threshold=10/15
+        ├── param=2t, threshold=273.15
+        └── param=tp, threshold=10/100/20/25/50
+    """).convert_dtypes(
+        {
+            "threshold": float,
+        }
+    )
--- a/tests/test_set_operations.py
+++ b/tests/test_set_operations.py
@ -1,6 +1,127 @@
 from qubed import Qube


+def set_operation_testcase(name, testcase):
+    q1 = Qube.from_tree(testcase["q1"])
+    q2 = Qube.from_tree(testcase["q2"])
+    assert q1 | q2 == Qube.from_tree(testcase["union"]), (
+        f"Case: {name} Op: Union\n {q1 = }\n {q2 = }\n {q1 | q2 = }\n expected = {testcase['union']}\n"
+    )
+    assert q1 & q2 == Qube.from_tree(testcase["intersection"]), (
+        f"Case: {name} Op: Intersection\n {q1 = }\n {q2 = }\n {q1 - q2 = }\n expected = {testcase['intersection']}\n"
+    )
+    assert q1 - q2 == Qube.from_tree(testcase["difference"]), (
+        f"Case: {name} Op: Difference\n {q1 = }\n {q2 = }\n {q1 - q2 = }\n expected = {testcase['difference']}\n"
+    )
+
+
+# These are a bunch of testcases where q1 and q2 are specified and then their union/intersection/difference are checked
+# Generate them with code like:
+# q1 = Qube.from_tree("root, frequency=*, levtype=*, param=*, levelist=*, domain=a/b/c/d")
+# q2 = Qube.from_tree("root, frequency=*, levtype=*, param=*, domain=a/b/c/d")
+
+# test = {
+# "q1": str(q1),
+# "q2": str(q2),
+# "union": str(q1 | q2),
+# "intersection": str(q1 & q2),
+# "difference": str(q1 - q2),
+# }
+# BUT MANUALLY CHECK THE OUTPUT BEFORE ADDING IT AS A TEST CASE!
+
+
+testcases = {
+    "Simplest case, only leaves differ": {
+        "q1": "root, a=1, b=1, c=1",
+        "q2": "root, a=1, b=1, c=2",
+        "union": "root, a=1, b=1, c=1/2",
+        "intersection": "root",
+        "difference": "root, a=1, b=1, c=1",
+    },
+    "Some overlap but also each tree has unique items": {
+        "q1": "root, a=1, b=1, c=1/2/3",
+        "q2": "root, a=1, b=1, c=2/3/4",
+        "union": "root, a=1, b=1, c=1/2/3/4",
+        "intersection": "root, a=1, b=1, c=2/3",
+        "difference": "root, a=1, b=1, c=1",
+    },
+    "Overlap at two levels": {
+        "q1": "root, a=1, b=1/2, c=1/2/3",
+        "q2": "root, a=1, b=2/3, c=2/3/4",
+        "union": """
+        root, a=1
+        ├── b=1, c=1/2/3
+        ├── b=2, c=1/2/3/4
+        └── b=3, c=2/3/4
+    """,
+        "intersection": "root, a=1, b=2, c=2/3",
+        "difference": """
+        root, a=1
+        ├── b=1, c=1/2/3
+        └── b=2, c=1""",
+    },
+    "Simple difference": {
+        "q1": "root, a=1, b=1, c=1/2/3",
+        "q2": "root, a=1, b=1, c=2",
+        "union": "root, a=1, b=1, c=1/2/3",
+        "intersection": "root, a=1, b=1, c=2",
+        "difference": "root, a=1, b=1, c=1/3",
+    },
+    "Check that we can merge even if the divergence point is higher": {
+        "q1": "root, a=1, b=1, c=1",
+        "q2": "root, a=2, b=1, c=1",
+        "union": "root, a=1/2, b=1, c=1",
+        "intersection": "root",
+        "difference": "root, a=1, b=1, c=1",
+    },
+    "Two equal qubes": {
+        "q1": "root, a=1, b=1, c=1",
+        "q2": "root, a=1, b=1, c=1",
+        "union": "root, a=1, b=1, c=1",
+        "intersection": "root, a=1, b=1, c=1",
+        "difference": "root",
+    },
+    "Two qubes that don't compress on their own but the union does": {
+        "q1": """
+        root
+        ├── a=1/3, b=1
+        └── a=2, b=1/2
+        """,
+        "q2": "root, a=1/3, b=2",
+        "union": "root, a=1/2/3, b=1/2",
+        "intersection": "root",
+        "difference": """
+        root
+        ├── a=1/3, b=1
+        └── a=2, b=1/2
+        """,
+    },
+    "With wildcards": {
+        "q1": "root, frequency=*, levtype=*, param=*, levelist=*, domain=a/b/c/d",
+        "q2": "root, frequency=*, levtype=*, param=*, domain=a/b/c/d",
+        "union": """
+        root, frequency=*, levtype=*, param=*
+        ├── domain=a/b/c/d
+        └── levelist=*, domain=a/b/c/d
+    """,
+        "intersection": "root",
+        "difference": "root, frequency=*, levtype=*, param=*, levelist=*, domain=a/b/c/d",
+    },
+    "Merging wildcard groups": {
+        "q1": "root, levtype=pl, param=q, levelist=100/1000, quantile=*",
+        "q2": "root, levtype=pl, param=t, levelist=100/1000, quantile=*",
+        "union": "root, levtype=pl, param=q/t, levelist=100/1000, quantile=*",
+        "intersection": "root",
+        "difference": "root, levtype=pl, param=q, levelist=100/1000, quantile=*",
+    },
+}
+
+
+def test_cases():
+    for name, case in testcases.items():
+        set_operation_testcase(name, case)
+
+
 def test_leaf_conservation():
    q = Qube.from_dict(
        {
--- a/tests/test_wildcard.py
+++ b/tests/test_wildcard.py
@ -0,0 +1,44 @@
+from qubed import Qube
+
+q = Qube.from_tree("""
+root
+├── class=od, expver=0001/0002, param=1/2
+└── class=rd
+    ├── expver=0001, param=1/2/3
+    └── expver=0002, param=1/2
+""")
+
+wild_datacube = {
+    "class": "*",
+    "expver": "*",
+    "param": "1",
+}
+
+
+def test_wildcard_creation():
+    Qube.from_datacube(wild_datacube)
+
+
+def test_intersection():
+    wild_qube = Qube.from_datacube(wild_datacube)
+    intersection = q & wild_qube
+    assert intersection == Qube.from_dict(
+        {
+            "class=od/rd": {
+                "expver=0001/0002": {"param=1": {}},
+            },
+        }
+    )
+
+
+def test_wildcard_union():
+    q1 = Qube.from_tree(
+        "root, frequency=*, levtype=*, param=*, levelist=*, domain=a/b/c/d"
+    )
+    q2 = Qube.from_tree("root, frequency=*, levtype=*, param=*, domain=a/b/c/d")
+    expected = Qube.from_tree("""
+    root, frequency=*, levtype=*, param=*
+    ├── domain=a/b/c/d
+    └── levelist=*, domain=a/b/c/d
+    """)
+    assert (q1 | q2) == expected
--- a/web_query_builder/.env
+++ b/web_query_builder/.env
@ -1 +0,0 @@
-API_HOST=localhost:8124
--- a/web_query_builder/app.py
+++ b/web_query_builder/app.py
@ -1,33 +0,0 @@
-import os
-
-from flask import (
-    Flask,
-    render_template,
-    request,
-)
-from flask_cors import CORS
-from werkzeug.middleware.proxy_fix import ProxyFix
-
-app = Flask(__name__)
-CORS(app, resources={r"/api/*": {"origins": "*"}})
-
-# This is required because when running in k8s the flask server sits behind a TLS proxy
-# So flask speaks http while the client speaks https
-# Client <-- https ---> Proxy <---- http ---> Flask server
-# For the Oauth flow, flask needs to provide a callback url and it needs to use the right scheme=https
-# This line tells flask to look at HTTP headers set by the TLS proxy to figure out what the original
-# Traffic looked like.
-# See https://flask.palletsprojects.com/en/3.0.x/deploying/proxy_fix/
-app.wsgi_app = ProxyFix(app.wsgi_app, x_for=1, x_proto=1, x_host=1, x_prefix=1)
-
-config = {}
-
-
-@app.route("/")
-def index():
-    return render_template(
-        "index.html",
-        request=request,
-        config=config,
-        api_url=os.environ.get("API_URL", "/api/stac"),
-    )
--- a/web_query_builder/requirements.txt
+++ b/web_query_builder/requirements.txt
@ -1,8 +0,0 @@
-flask==3
-pyyaml
-flask_dance
-python-dotenv
-flask-login
-flask-cors
-cachetools
-uvicorn
--- a/web_query_builder/run.sh
+++ b/web_query_builder/run.sh
@ -1,2 +0,0 @@
-export API_URL="http://127.0.0.1:8124/api/stac"
-flask run --debug --port=5006
Author	SHA1	Message	Date
Tom	165bf5aca2	Tests passing checkpoint	2025-06-03 14:57:27 +02:00
Tom	aaafa28dfb	A bit more on the rust backend	2025-05-29 17:09:17 +02:00
Tom Hodson	3328a0375b	Fix update script a bit	2025-05-23 16:45:37 +00:00
Tom	ba2c67d812	Create example ingestion script	2025-05-23 10:55:32 +01:00
Tom	04b4ee24eb	Silence protobuf warning	2025-05-22 17:26:58 +01:00
Tom	7069b70dd4	remove prints	2025-05-22 14:42:49 +01:00
Tom	90ea736c43	flesh out rust implementation	2025-05-22 14:40:44 +01:00
Tom	959dac332d	Start writing rust backend	2025-05-19 10:20:12 +01:00
Tom	97c5abc38b	Update image link	2025-05-14 10:33:38 +01:00
Tom	1188733034	Update re	2025-05-14 10:21:48 +01:00
Tom	35bb8f0edd	Massive rewrite	2025-05-14 10:14:02 +01:00
Tom	ed4a9055fa	fix bug add testcases	2025-05-12 14:40:16 +01:00
Tom	110046b251	progress on metadata	2025-05-09 17:25:00 +01:00
Tom	a85b700084	Merge branch 'main' into metadata	2025-05-07 15:47:40 +01:00
Tom	271d06c65a	Update banner.svg	2025-05-06 15:16:34 +01:00
Tom	7c28c7023b	Update README.md	2025-05-06 15:16:22 +01:00
Tom	4924fdb804	Add banner	2025-05-06 10:19:11 +01:00
Tom	d246dae54d	Update climate_dt.json	2025-04-30 14:17:15 +02:00
Tom	07f9a24daa	Add require_match argument to select	2025-04-30 14:05:42 +02:00
Tom	b13a06a0cc	Update .gitignore	2025-04-30 14:05:42 +02:00
Tom	87c57ec2cc	add selection test	2025-04-30 14:05:42 +02:00
Tom	a957d26da7	Update README.md	2025-04-30 11:06:48 +02:00
Tom	80b0408722	Update README ulrs	2025-04-24 10:59:42 +01:00
Tom	79e9f83c8c	Add note about live server to docs	2025-04-24 10:54:11 +01:00
Tom	fff00ca6f1	Add test cads.json	2025-04-24 10:30:32 +01:00
Tom	fa646aee77	cosmetics	2025-04-24 10:28:52 +01:00
Tom	e04c0dd3bc	Add tests	2025-04-23 14:40:09 +01:00
Tom	4e777f295d	More work on metadata	2025-04-23 14:40:09 +01:00
Tom	1259ff08b6	first attempt	2025-04-23 14:38:33 +01:00
Tom	7b36a76154	Fix all of mypy's complaints.	2025-04-23 12:43:49 +01:00
Tom	10106ba6d8	Fix link templates	2025-04-17 09:31:51 +01:00
Tom Hodson	2fa99d775c	fix mars language	2025-04-16 16:19:34 +00:00
Tom	b5c2681f63	Quick and dirty simple stac endpoint	2025-04-16 17:05:40 +01:00
Tom	dfc61caa38	Moves installation -> development	2025-04-16 15:20:38 +01:00
Tom	a502cb6ab2	Rejig quickstart a bit	2025-04-16 15:13:06 +01:00
Tom Hodson	ca944521f9	fix	2025-04-15 14:04:52 +00:00
Tom Hodson	6ec4b044b8	fix urls	2025-04-15 14:02:01 +00:00
Tom Hodson	251bec14fc	update webapp	2025-04-15 13:50:03 +00:00
Tom	c3556ce6fa	Fix example code url	2025-04-15 14:43:37 +01:00
Tom	bf47401e6c	Add Qube.load	2025-04-12 16:06:29 +02:00
Tom	11014b07ea	Update climate_dt.json	2025-04-03 14:03:44 +01:00
Tom	b6a27fdadf	Update Qube.py	2025-04-03 14:02:50 +01:00
Tom	70b1fd65e5	add remove_by_key and improve compression	2025-04-02 17:51:20 +01:00
Tom	2e36db4268	Create od.json	2025-03-31 19:11:42 +01:00
Tom	6039a3a494	update cmd app	2025-03-31 19:11:15 +01:00
Tom	7ef930bc1c	Update climate_dt.json	2025-03-31 18:45:40 +01:00
Tom	79983f85a1	clean up chart	2025-03-31 17:24:20 +01:00
Tom	ab2f8cf3f3	Integrate web_query_builder and stac server backend	2025-03-31 16:36:04 +01:00
Tom Hodson	4502a942cb	integrate stac-server and web app	2025-03-31 12:48:13 +00:00
Tom Hodson	3017185950	stac_server	2025-03-31 12:40:23 +00:00
Tom	cd26905261	Update structured_stac.md	2025-03-31 10:37:56 +01:00
Tom	0d3c8248b0	update chart	2025-03-31 10:10:05 +01:00
Tom	b2aba5dd42	update build	2025-03-28 18:08:31 +00:00
Tom	79b97fd813	Update stac server and frontend	2025-03-28 17:50:29 +00:00
Tom	275831d186	Update climate dt schema	2025-03-28 16:32:46 +00:00
Tom	57877e1e0c	Update extremes_dt.json	2025-03-28 09:52:29 +00:00
Tom	cf9db41dc4	add extremes_dt test data	2025-03-28 09:31:26 +00:00
Tom	8f1735c076	Update .gitignore	2025-03-28 09:29:13 +00:00
Tom	39f348244d	Better error message for from_tree	2025-03-27 18:30:39 +00:00
Tom	2884f9fff8	update docs	2025-03-27 18:30:12 +00:00
Tom	df5360f29a	Add convert_dtypes and selection with functions	2025-03-27 16:02:58 +00:00
Tom	d2f3165fe8	Remove print statement	2025-03-25 15:07:06 +00:00
Tom	6b98f7b7a9	Add creation from tree representation	2025-03-25 15:01:23 +00:00
Tom	9beaaa2e10	make consumption by selection off by default	2025-03-24 15:30:34 +00:00
Tom	06c84fb20e	Fix selection bug to require that arguments be consumed by a branch	2025-03-24 15:28:06 +00:00
Tom	c31467fb04	Update fiab.md	2025-03-04 19:18:42 +01:00
Tom	6648502bf4	Add experimental wildcard value	2025-03-04 19:02:55 +01:00
				`@ -0,0 +1 @@`
				`kubectl -n qubed logs deployment/stac-server`