docs: update STATUS.md and CHECKPOINT.md for Session 18b

Session 18b: Azure AI Foundry multi-model routing (7 models, dual API
pattern, lazy init fix). Known issue: catalog models return markdown
instead of JSON — to fix in Session 19.

Co-Authored-By: Claude Opus 4.6 <[email protected]>

Author: mechramc

CHECKPOINT.md

@@ -9,20 +9,36 @@
 ## Last Updated By
 - **Tool**: Claude Code
 - **Date**: 2026-02-17
-- **Session**: 18
+- **Session**: 18b
 
 ## Current State
-- **Phase**: Demo Preparation — All features + MS visibility polish implemented
-- **Last completed task**: Session 18 Microsoft visibility features (6 features for hackathon wow factor)
-- **Next task**: E2E re-test all flows, demo recording (7 workflows)
+- **Phase**: Demo Preparation — Multi-model routing implemented, catalog model parsing issue to fix
+- **Last completed task**: Session 18b Azure AI Foundry multi-model routing (7 models across 2 providers)
+- **Next task**: Fix catalog model JSON parsing (Phi-4/Llama return markdown not JSON), then E2E test + demo
 - **Branch**: `main`
-- **Repo is green**: YES (full build passes — 6/6 turbo tasks, 0 lint errors, all tests green)
+- **Repo is green**: YES (full build passes — 6/6 turbo tasks, 0 lint errors)
 - **CI/CD**: All changes committed and pushed
+- **Known issue**: Catalog models (Phi-4, Llama-3.3-70B) return markdown instead of JSON — `JSON.parse()` fails with "Unexpected token '#'"
 - **Live API**: `https://blueflame-api-dev.blackfield-ff30bbff.centralus.azurecontainerapps.io`
 - **Live Web**: `https://blueflame-web-dev.blackfield-ff30bbff.centralus.azurecontainerapps.io`
 - **Licensing**: BSL 1.1 (source-available, Murai Labs commercial ownership)
 
-## What Just Happened (Sessions 10–18)
+## What Just Happened (Sessions 10–18b)
+
+### Session 18b: Azure AI Foundry Multi-Model Routing
+
+Upgraded Azure OpenAI resource to Azure AI Foundry. Deployed 3 new models (Phi-4, Llama-3.3-70B-Instruct, o3-mini) alongside existing gpt-4o and gpt-4o-mini. Updated σ-router to use 7 models across 2 active providers.
+
+**Key changes:**
+1. **Model registry** — New routing table: Phi-4 (Routine tier), Llama-3.3-70B (Standard Builder/Fixer), o3-mini (Complex Verifier/Planner), gpt-4o/gpt-4o-mini (Standard/Complex), Claude Sonnet 4.5 (Complex when Anthropic key set)
+2. **Lazy initialization** — Fixed ESM import hoisting bug: `initDefaults()` was running at module load before `dotenv.config()`, making env vars empty. Now uses `ensureInitialized()` pattern.
+3. **Dual API pattern** — OpenAI models use `{endpoint}/openai/deployments/{name}` + api-version query; catalog models use `{resource}/openai/v1/` with NO api-version (model in request body). New helpers: `getAzureBaseURL()`, `getAzureDefaultQuery()`, `isOpenAIModel()`.
+4. **All 7 agents + provider updated** — builder, verifier, designer, planner, explainer, fixer, spec-generator, azure-openai provider all use new helpers.
+5. **Biome CI fix** — Changed lint script to `--diagnostic-level=error` so warnings don't block CI. Fixed 4 SVG accessibility errors and 2 missing hook deps.
+
+**Known issue:** Catalog models (Phi-4, Llama) return markdown instead of JSON when asked for structured output. Need to add explicit JSON format instructions to system prompts or use `response_format: { type: "json_object" }`.
+
+**Commits:** 6 commits pushed to main (lint fix, model registry, lazy init, api-version updates, agent URL fixes).
 
 ### Session 18: Microsoft Visibility Features (Hackathon Polish)
 
@@ -266,13 +282,15 @@ Resolved all 13 integration gaps identified in the gap analysis. Every phase ver
 - **Session 16**: Spec viewer on run dashboard (API response, SpecViewerPanel, toggle)
 - **Session 17**: Post-execution deployment workflow + Session 16 UX fixes committed
 - **Session 18**: Microsoft visibility features (6 features for hackathon wow factor)
+- **Session 18b**: Azure AI Foundry multi-model routing (7 models, dual API pattern, lazy init fix)
 
 ## What To Pick Up Next
 
 ### Immediate (Session 19)
-1. **E2E test all flows** — Spec→Plan→Execute, SCR, failure→fix→approve
-2. **Demo recording** — 7 workflow demonstrations (WF1-WF7)
-3. **Submission package** — README (done), architecture diagram, demo video
+1. **Fix catalog model JSON parsing** — Phi-4/Llama return markdown not JSON. Options: (a) add `response_format: { type: "json_object" }` to API calls, (b) add stronger "respond ONLY in JSON" instructions to system prompts, (c) add a markdown-to-JSON fallback parser. Option (a) is preferred if catalog models support it.
+2. **E2E test all flows** — Spec→Plan→Execute, SCR, failure→fix→approve
+3. **Demo recording** — 7 workflow demonstrations (WF1-WF7)
+4. **Submission package** — README (done), architecture diagram, demo video
 
 ### What's Deferred (OK to skip)
 - **S16-004: Azure SignalR migration** — Socket.IO works; migration is mechanical
@@ -304,7 +322,7 @@ None — all changes committed and pushed.
 | Container Registry | `blueflamecr.azurecr.io` | Active |
 | Log Analytics | `blueflame-logs-dev` | Active |
 | App Insights | Connected | Active |
-| OpenAI | `blueflame-openai-dev` (gpt-4o deployed) | Active |
+| AI Foundry | `blueflame-openai-dev` (gpt-4o, gpt-4o-mini, o3-mini, Phi-4, Llama-3.3-70B) | Active |
 
 ## Key Files Reference
 - `Blueflame-Spec-v3-ACAR.md` — Source of truth
@@ -346,5 +364,7 @@ None — all changes committed and pushed.
 - ACR admin credentials are persistent; GHCR tokens are ephemeral (don't use GHCR)
 - On Windows/MSYS: use `MSYS_NO_PATHCONV=1` prefix for az CLI commands with `/` paths
 - Express route ordering: static routes before catch-all `/:id` routes
+- Catalog models (Phi-4, Llama) use `/openai/v1/` path with NO api-version; OpenAI models use `/openai/deployments/{name}` with api-version
+- ESM import hoisting: never call functions that read `process.env` at module load time — use lazy initialization
 - Authorize endpoint requires `Blueflame_Authorizer` role (dev mode: set `X-Dev-Role` header)
 - **Licensing**: BSL 1.1 — treat as commercially owned, not open source

docs/STATUS.md

@@ -212,11 +212,23 @@
 | AzureServiceUsagePanel on run dashboard | UI | P0 | **DONE** |
 | Landing page: 6 MS-branded feature cards | UI | P0 | **DONE** |
 
+### Session 18b — Azure AI Foundry Multi-Model Routing
+| Task | System | Priority | Status |
+|------|--------|----------|--------|
+| Deploy Phi-4, Llama-3.3-70B, o3-mini to Foundry | S12 | P0 | **DONE** |
+| Update model registry routing table (7 models) | S12 | P0 | **DONE** |
+| Lazy init model registry (ESM hoisting fix) | S12 | P0 | **DONE** |
+| Fix API version per model type (OpenAI vs catalog) | S12 | P0 | **DONE** |
+| Update all 7 agents + provider to use getAzureDefaultQuery | S12 | P0 | **DONE** |
+| Biome lint CI fix (diagnostic-level=error) | CI/CD | P0 | **DONE** |
+| Catalog model JSON parsing (Phi-4 returns markdown) | S12 | P0 | **KNOWN ISSUE** |
+
 ### Final — Demo + Submit
 | Task | Priority | Status |
 |------|----------|--------|
 | S16-004: Azure SignalR migration | P1 | Deferred (Socket.IO adequate) |
 | S16-005: Application Insights SDK | P1 | Deferred (OTel spans cover this) |
+| Fix catalog model response parsing (markdown→JSON) | P0 | Not started |
 | End-to-end testing (Spec→Plan→Execute flow) | P0 | In progress |
 | Demo recording (7 workflows) | P0 | Not started |
 | Submission package | P0 | Not started |
@@ -235,8 +247,9 @@
 - **Documentation**: README, STATUS, CHECKPOINT, PRD, spec, CHANGELOG updated
 - **Demo/Submit**: 0/2
 - **Session 18 MS Visibility**: 6/6 (services strip, infra selection, Azure branding, toasts, usage panel, landing page)
-- **Grand Total**: 119/124 complete (2 deferred, 2 demo remaining, 1 skipped)
-- **Critical path**: E2E testing → demo recording → submission package
+- **Session 18b Foundry Multi-Model**: 6/7 (deploy, registry, lazy init, api-version, agents, lint — 1 known issue: catalog JSON parsing)
+- **Grand Total**: 125/131 complete (2 deferred, 3 demo remaining, 1 known issue)
+- **Critical path**: Fix catalog model JSON parsing → E2E testing → demo recording → submission package
 
 ## Enterprise Streams Overview
 
@@ -273,6 +286,7 @@
 | 2026-02-12 | Defer AppInsights SDK (S16-005) | OTel spans already provide instrumentation |
 
 ## Blockers
+- **Catalog model JSON parsing**: Phi-4 (and likely Llama) returns markdown instead of JSON — `JSON.parse()` fails with "Unexpected token '#'". Need to add structured output instructions or response format enforcement for catalog models.
 - **E2E flow partially tested**: User tested Spec→Plan→Execute and SCR flows — found and fixed 11 UX/backend bugs in Session 14
 - **Demo recording not started**: All features implemented + polished; need to record 7 workflow demos