Skip to content
CSA Loom — the Microsoft Fabric experience for Azure tenants where Fabric isn't yet available: lakehouses, warehouses, notebooks, semantic models, Activator rules, Data Agents, across Commercial, GCC, GCC-High, and DoD IL5

Fabric-parity loop — running progress

Live log of the multi-agent fabric-parity-loop workflow. Most recent at top.

2026-05-26 — Build Phase 1 complete — vaporware cleared + D/C-grade upgrades shipped

After the catalog phase wrapped, ran the prioritized Build sequence end-to-end:

F-fixes (vaporware cleared) - data-product — Cosmos-state editor + Purview-pending MessageBar gate. Removed hardcoded customer-360 / alice@contoso / fixed bundle grid. - gql-graph — Run button now dispatches to /api/items/gql-graph/[id]/query; 3 backends (persist-only / fabric-graph / cosmos-gremlin-translate).

Quick wins (dead ribbon buttons wired) - cosmos-gremlin-graph Edges/Vertices, cross-item-copilot View registry + Session New/Refresh, data-product-instance Health column + ribbon Health button.

D-fixes - usql-job — deprecation MessageBar (ADLA EOL 2024-02-29) + heuristic U-SQL→PySpark translator. - vector-store — added Microsoft-recommended cosmos-nosql backend (now default). - ontology — Materialize as graph-model button creates a graph-model item from parsed class hierarchy. - plan — status badges (todo/doing/done/overdue) + progress meter. - map — Azure Maps Static-API tile preview when NEXT_PUBLIC_LOOM_AZURE_MAPS_KEY is set.

C-enhancements - Notebook Phase 2 — Data items pane + Lakehouse attach modal in left rail. - APIM Policy — Operation scope added (apis/{aid}/operations/{oid} resolver in backend + form fields in UI). - Shared pipeline DAG canvas — new lib/components/pipeline/pipeline-dag-view.tsx (read-only topological layout + activity-type color coding + success/failure/completion/skip edge legend); wired into adf-pipeline, synapse-pipeline, and data-pipeline editors as a "Graph" tab.

After these: zero F-grade items remain in the wiring-audit. Branch is ~30 commits ahead of origin/access-patterns-vpn-agw-fd.

Still deferred (truly multi-session work) - Pipeline DAG Phase 2 (drag-drop authoring, activity library palette, properties pane for the selected node) — currently read-only. - Foundry Agent Service runtime for operations-agent / data-agent (per their parity specs — 4-6 sessions each). - Purview Unified Catalog wiring for data-product — needs new lib/azure/purview-client.ts + Bicep module + role assignments. - Notebook Phase 3 (HistoryDrawer, per-cell edge toolbars, AI tools tab) and Phase 1B Monaco upgrade.

2026-05-26 — ✅ CATALOG PHASE COMPLETE — 85 / 85 UIs cataloged

All 85 Loom editors have a parity spec on disk under docs/fiab/<name>-parity-spec.md.

Wave-by-wave tally: - Wave 1 (previous session): 2 (notebook, lakehouse) - Wave 2: 10 (eventstream, eventhouse, dataflow, copy-job, warehouse, semantic-model, report, kql-database, kql-queryset, data-pipeline) - Wave 3: 13 (dashboard, paginated-report, scorecard, ml-experiment, ml-model, spark-job-definition, environment, graphql-api, user-data-function, kql-dashboard, activator, mirrored-database, dbt-job) - Wave 4: 21 (variable-library, plan, ontology, graph-model, map, operations-agent, data-agent, synapse-{ded,serverless,spark}-sql-pool, synapse-pipeline, adf-{pipeline,dataset,trigger}, databricks-{notebook,job,cluster,sql-warehouse}) - Wave 5: 21 (apim-{api,product,policy}, data-product, ai-foundry-{hub,project}, compute, dataset, prompt-flow, evaluation, content-safety, tracing, copilot-studio-{agent,knowledge,topic,action,channel,analytics}, ai-search-index, usql-job, ai-search-skillset) - Wave 6: 18 (copilot-template-library, powerplatform-environment, dataverse-table, power-app, power-automate-flow, power-page, ai-builder-model, azure-sql-{server,database,managed-instance}, sql-server-2025-vector-index, geo-{map,dataset,query,pipeline}, cosmos-gremlin-graph, cypher-graph, gql-graph, vector-store, data-product-{template,instance}, cross-item-copilot)

Notable findings from the catalog: 1. data-product is F-grade vaporware — hardcoded productId='customer-360', owner 'alice@contoso', fixed bundle grid. "Publish to APIM" creates an APIM Product (wrong backend). Needs Purview Unified Catalog wiring. Reclassified in wiring-audit.md. 2. usql-job is unreachable — Azure Data Lake Analytics retired 2024-02-29. Recommendation: graceful deprecation + Spark-translation helper. 3. sql-server-2025-vector-index editor has a real DDL bug — DIMENSIONS is a column-level property (VECTOR(N)), not an index WITH option. 4. Copilot Studio 6-pack is real and wired through Dataverse msdyn_copilots / msdyn_knowledgesources / etc., gated by per-env Copilot Studio enablement. 5. AI Search Skillset / Indexer / DataSource items don't exist in registry — gap that blocks enterprise-search ingest. 6. APIM Policy operation scope is unwired — only Global/API/Product scopes; missing /apis/{aid}/operations/{oid}/policies/policy. 7. All four Geo items are Loom-native (no Fabric equivalent); specs pivot to Azure-native references (Azure Maps Studio, KQL geo_*, Synapse spatial extensions).

Next phase: with 85 specs in hand, Phase 2 (Build) can now be sequenced. Suggested ordering per the no-vaporware rule: 1. F-grade fixes (vaporware violations must clear first): data-product MessageBar gate + Purview wiring 2. D-grade upgrades: usql-job deprecation MessageBar, ontology/map/plan runtime wiring 3. C-grade enhancements: notebook Phase 2 (ExplorerPane / Lakehouse attach), apim-policy operation scope, the DAG canvas (shared by data-pipeline / adf-pipeline / synapse-pipeline) 4. A-grade polish: Monaco upgrades, multi-tab, query history, etc.

2026-05-26 — Catalog phase ~73% complete (62 / 85 UIs cataloged)

Today's full output: - Waves 1+2 (12 UIs) — closed previous session + data-pipeline added today - Wave 3 (13 UIs) — completed - Wave 4 (21 UIs) — all 6 agents complete - Wave 5 (21 UIs) — 4 of 5 agents complete; tracing + ai-foundry dataset still pending - Wave 6 (22 UIs) — 1 of 4 agents complete (geo 4-pack); Power Platform 7-pack, Azure SQL 4-pack, graphs+misc 7-pack still running

Build phase progress (alongside catalog): - ✓ Notebook Phase 1A — cell-based scaffold shipped: CodeCell + MarkdownCell + CellAdder; per-cell Run via Cosmos pendingRuns[runId] transient map; language picker (PySpark / Spark / SparkSQL / SparkR / Python / T-SQL); back-compat with legacy {code, lang} blob (commit 3e1b32b6) - ✓ Variable Library — extended from 4 to 9 variable types: + Integer, DateTime, Guid, ItemReference, ConnectionReference (in addition to existing String/Number/Boolean/SecretReference); added Description column; per-cell value validation regex per type

Remaining catalog work: tracing, ai-foundry-dataset (waiting on partial agent finish), plus 5 still-running agents covering: Copilot Studio 6-pack, Power Platform 7-pack (copilot-template-library + powerplatform-environment + dataverse-table + power-app + power-automate-flow + power-page + ai-builder-model), Azure SQL 4-pack, and graphs/misc 6-pack (cypher-graph + gql-graph + vector-store + data-product-template + data-product-instance + cross-item-copilot).

2026-05-26 — Wave 4 catalog launched (21 UIs in 6 parallel agents) + Notebook Phase 1A shipped

Wave 4 agents kicked off:

Agent UIs
a2433a6acd1482965 variable-library, plan
ae995b89188aa4e58 ontology, graph-model, map
a748958dbf3da3721 operations-agent, data-agent
a5984cea98c72082f synapse-dedicated-sql-pool, synapse-serverless-sql-pool, synapse-spark-pool
a898bc142c4bf4488 synapse-pipeline, adf-pipeline, adf-dataset, adf-trigger
abe32b64e154c734e databricks-notebook, databricks-job, databricks-cluster, databricks-sql-warehouse

Target after wave 4: 46 / 85 UIs cataloged.

Notebook editor Phase 1A landed (commit 3e1b32b6): - NotebookCell type + state migration in lib/types/notebook-cell.ts - lib/components/notebook/{code-cell,markdown-cell,cell-adder}.tsx — Fluent UI cell shells with per-cell Run, lang picker, move up/down, delete - Cell-based load/save with backward compat for legacy {code, lang} blob - Per-cell run dispatch via Cosmos pendingRuns[runId] transient map (each cell's source is cached at dispatch; poll endpoint reads from there when Livy session reaches idle) - Markdown minimal renderer (headings/bold/italic/code/lists/links) — no Monaco/react-markdown deps added yet - .gitignore carve-out for the previously-eaten runs/[runId]/ dynamic route segment

Honest scope: Phase 1A is the cell scaffold only. Phase 2 (ExplorerPane / Lakehouse attach), Phase 3 (HistoryDrawer / cell toolbars) still pending per notebook-parity-spec.md. The notebook editor now visibly looks like Fabric (cell-based) but doesn't yet have OneLake browsing or the Connect menu.

2026-05-26 — Wave 3 catalog launched (12 UIs in 6 parallel agents)

Wave 3 catalog kicked off — 6 general-purpose agents running in parallel, each owning 1-3 UIs:

Agent UIs
ab63f02fd92b935c8 dashboard, paginated-report, scorecard
a1402587cb2c6e0d7 ml-experiment, ml-model
a67c4b66eb7808477 spark-job-definition, environment
aa402e5fa7f4b0b6f graphql-api, user-data-function
a22495e60e0256876 kql-dashboard, activator
a588b40b0dd792de0 mirrored-database, dbt-job

After wave 3 lands: 24 / 85 UIs cataloged. Then Phase 2 (Build) waves.

Catalog format change: switched from read-only Explore agents to general-purpose agents so they can Write spec files directly (no more I-write-back-from-text). Faster end-to-end.

2026-05-26 — Wave 2 catalog COMPLETE (12 UIs cataloged total)

Specs written to docs/fiab/<name>-parity-spec.md for the next 10 Fabric UIs:

UI Agent Status
eventstream a8260c3697beb6c69 ✓
eventhouse a8260c3697beb6c69 ✓
dataflow a30c2872e59523af4 ✓
copy-job a30c2872e59523af4 ✓
warehouse a25d745464518b765 ✓ (rewritten — agent confused with eventhouse)
semantic-model a6112853cd6c023e5 ✓
report a6112853cd6c023e5 ✓
kql-database ad6a393dd34fd232c ✓
kql-queryset ad6a393dd34fd232c ✓
data-pipeline aff49f5c28912ff78 ✓ (manually written from agent text; was read-only)

Cumulative: 12/85 UIs cataloged. Notebook + Lakehouse from wave 1, plus the above 10.

2026-05-26 — Workflow scaffold + first parallel catalog run

Workflow infrastructure shipped: - .claude/workflows/fabric-parity-loop.md — 3-agent pipeline design - .claude/commands/fabric-parity-loop.md — slash command orchestrator - docs/fiab/fabric-parity-tasks.json — 15-UI prioritized task list

First parallel catalog run (Phase 1 only, Phase 2 + 3 pending):

UI Agent Status Output
notebook af6f80e466901eecf (Explore) ✓ complete Validated existing notebook-parity-spec.md; added 3 ribbon items (AutoML / Pipeline / VS Code), "Ask Copilot" on cell toolbar, execution count badge [N]
lakehouse a4f93461062e0e80c (Explore) ✓ complete New lakehouse-parity-spec.md written — auto-paired SQL endpoint pattern confirmed, ribbon (Open notebook / Add to data agent / Manage OneLake security / Update all variables), 6 real bronze tables discovered

Build phase queued for next session: - /fabric-parity-loop notebook → cell-based editor rewrite, language picker, OneLake explorer panel - /fabric-parity-loop lakehouse → auto-paired SQL endpoint + ribbon + data grid + Open-in-notebook flow

Validate phase runs immediately after each Build via the verify-app subagent.

Known limitations the catalog agents surfaced

  1. The Playwright MCP can't reliably navigate Fabric's portaled overlays (modals close before screenshot, kebab menus dismiss on focus loss). Specs are written from snapshot-tree inspection instead of pixel-perfect screenshots.
  2. The Explore agent is read-only, so it documents findings in markdown rather than committing screenshots to git. Build agents read the markdown.
  3. Fabric workspaces with F-capacity (like casino-fabric-poc F64) have ALL Fabric items enabled; specs derived from there are upper bound. Some items (like AutoML, VS Code integration) may not have full Loom equivalents in v1.

Cumulative Loom shipping log (background)

Loom v Released Key changes
v3.18 2026-05-26 /api/cosmos-items fix for editor hydration bug + tab strip Fabric parity
v3.19 2026-05-26 Dataverse-scope tokens route through MSAL Web App SP
v3.20 2026-05-26 Power Pages schema fix + Copilot Studio gate + AppUser bootstrap
v3.21–22 2026-05-26 /api/loom/workspaces + /api/loom/compute-targets + 4-editor Fabric→Loom swap
v3.23–24 2026-05-26 Async notebook Run dispatch (beats FD 30s timeout)
v3.25 2026-05-26 data-pipeline → ADF redirect, dataflow + mirrored to Cosmos, bicep Spark pool

2026-05-26 (later) — v2 validator batch (39 editors)

Session: source-grade Phase 3 + Phase 4 onClick audit (live Phase 4 blocked by MFA expiry).

Output: docs/fiab/parity-gap/_validation-summary-2026-05-26.md plus per-family gap docs: - powerplatform-editors.md (6 editors) - azure-sql-editors.md (4) - geo-editors.md (5 including phase4 map) - graph-vector-editors.md (4) - fabric-iq-editors.md (4) - data-engineering-misc-editors.md (6) - api-data-product-editors.md (5) - stream-analytics-job.md (1 — replaces usql-job) - bi-rti-editors.md (6)

Grade distribution: 0 A · 4 B · 18 C · 16 D · 1 F (stream-analytics-job — source C, but slug 404s in deployed bundle).

B-grade editors (the wins): - mirrored-database — real 8-source-type create wizard, real Fabric REST - variable-library — 9 types + 4 value sets + per-type validation - data-product-template — gallery + detail + Instantiate end-to-end - data-product-instance — Health refresh wired ✓

Structural blockers (each affects most of the catalog): 1. No Monaco — every code/query/text editor is <textarea> (caps catalog at C until fixed) 2. Dead ribbon labels — ~120 ribbon buttons across the catalog have no onClick 3. No live output rendering — most query editors show <pre>{JSON}</pre> instead of a result grid 4. Read-only Power Platform / Power BI editors — browse + open-in-other-tool pattern, no in-Loom edit

Deploy required: stream-analytics-job is in source + registry but NOT in deployed loom-console--0000075 bundle. Today users opening it hit 404.