ml-model — parity gap (validator v2, 2026-05-26)¶
Loom URL: /items/ml-model/new Fabric reference: ai.azure.com → Models (registered model with versions, tags, lineage, deployments, sample inputs, evaluation cards) Loom screenshot: temp/parity/ml-model-loom.png
Phase 4¶
| Route | Status | Notes |
|---|---|---|
GET /api/items/ml-model (list) | 200 | 0 models |
GET /api/items/ml-model/new | 404 | {"ok":false,"error":"not found"} — Loom UI renders "Load failed — not found" MessageBar |
Same pattern as ml-experiment: /new doesn't make sense for read-only model registry.
Phase 3 — Fabric vs Loom¶
| Fabric element | Loom present? | Severity |
|---|---|---|
| Versions left tree | YES — but empty | — |
| Version detail (URI / type / created / tags / properties) | YES — flat table | — |
| Schema (inputs/outputs with type + shape) | NO | MAJOR |
| Sample inferencing widget (paste JSON → POST to endpoint) | NO | MAJOR |
| Lineage (training run that produced the model + dataset assets) | NO | MAJOR |
| Evaluation scorecards (eval runs tied to this model version) | NO | MAJOR |
| Deploy wizard (online endpoint / batch endpoint) | partial — "Real-time endpoint" ribbon button visible but dead | MAJOR |
| Apply (PREDICT) ribbon — dead button | dead | MAJOR (Fabric T-SQL PREDICT integration is a real feature) |
| Compare versions ribbon — dead button | dead | MAJOR |
| Tags / properties edit | NO — read only | MINOR |
Functional¶
- /new crashes with "Load failed — not found"
- 3 ribbon actions ("Compare versions", "Apply (PREDICT)", "Real-time endpoint") are all dead
Grade — F¶
/new is broken (read-only registry can't have /new). Three vaporware ribbon buttons. Grade F.
Recommendation: same as ml-experiment — should be a list view, not a create flow.