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

content-safety — parity gap (validator v2, 2026-05-26)

Loom URL: /items/content-safety/new Fabric reference: ai.azure.com — Content Safety Studio (Try it: text/image · Blocklists · Jailbreak detection · Custom categories · Prompt Shields · Protected material) Loom screenshot: temp/parity/content-safety-loom.png

Phase 4

Route Status Notes
GET /api/items/content-safety 200 Returns default thresholds: hate:4, selfHarm:4, sexual:4, violence:4
POST /api/items/content-safety text+image wired but not exercised —

Page shows two cards: Text moderation (Textarea + Analyze button), Image moderation (file input + Analyze button). Pre-text caption: "Default category set + severity thresholds. Custom blocklists land in v2.6." which is at least an honest deferral note.

Phase 3 — Fabric vs Loom

Fabric (Content Safety Studio) element Loom present? Severity
Per-category severity sliders (0–6 for Hate, Sexual, Self-harm, Violence) NO — thresholds are server-defaults only, no UI to change MAJOR
Blocklist editor (terms · regex · wildcard · case-sensitivity) NO — explicit "v2.6" deferral but no MessageBar in UI surface to set expectations MAJOR
Custom category wizard (train on labelled examples) NO MAJOR
Prompt Shields (jailbreak / indirect attack detection) NO MAJOR
Protected Material detection NO MAJOR
Text moderation try-it YES — Textarea + Analyze —
Image moderation try-it YES — file input + Analyze —
Result rendering: severity badges per category NO — Loom dumps JSON in a <pre> MINOR (BLOCKER per build-phase contract #4 — output rendering should be structured)
Batch / bulk testing NO MINOR
API key view + rate-limit info NO MINOR

Functional

  • Analyze Text / Analyze Image buttons fire real POSTs to BFF (BFF returns 200 with policies for GET; POST not exercised live but route is registered)
  • Result rendering is JSON.stringify in a <pre> block — that fails the structured-output rule

Grade — D

Read endpoint is real. Try-it text+image work. But the editor doesn't surface ANY of the Content Safety Studio editing surfaces (sliders, blocklists, Prompt Shields, custom categories) — it's a try-it widget over a no-config policy. The "v2.6 deferral" is honest in the caption but the buttons that LOOK like they configure thresholds don't exist. Grade D.