Session 7e: Grade adapter, normalize consolidation, ARCH-2 banners

docs/SYSTEM-MANIFEST.md

@@ -411,24 +411,50 @@ No circular imports detected.
 ### ARCH — Architecture
 
 - **[ARCH-1] Dual grading paths.** Severity: Medium. Status:
-  **DEFERRED in Session 7d** — the legacy `propAnalyzer → grader →
-  UnifiedOddsProvider` chain returns a 4-letter grade + 0-100
-  confidence + kill-conditions + reasoning.steps shape; the new
-  `engine1` returns an 11-step grade + 0-1 confidence + factors. Both
-  shapes are consumed by different surfaces (GradeCard + DemoScan vs
-  the orchestrator's grade_history writer). Unification requires
-  frontend + backend coordination — out of scope for the audit-fix
-  session. Migration plan: (1) decide which shape the UI keeps; (2)
-  write an adapter from the other shape; (3) rewire one route at a
-  time, starting with `/api/analyze/prop`. Session 7d added a
-  DEPRECATED banner to `src/services/grader.js` to signal that new
-  features should land in engine1.js.
+  **PARTIAL in Session 7e** — Step 1 of the migration (the adapter)
+  is complete and tested. Steps 2-6 deferred per ESCAPE HATCH.
+    Step 1 ✓ — `src/utils/gradeAdapter.js` translates engine1 output
+                  into the legacy shape `DemoScan` reads. 26 unit
+                  tests covering grade collapse, confidence math,
+                  kill-condition mapping, reasoning fallback, partial
+                  input safety.
+    Step 2-6 deferred — legacy `analyzeProp` and `engine1.gradeProp()`
+                  have incompatible INPUT contracts: the legacy
+                  analyzer takes a raw prop and self-fetches data
+                  (odds, stats, opponent, spread, kill conditions);
+                  engine1 takes a pre-computed feature vector and is
+                  a pure grading function. To rewire any route to
+                  engine1 we'd first need to extract an
+                  orchestrator-lite preprocessor — that's the kind
+                  of architectural change 7c/7d's no-restructure
+                  rule blocks. The reasoning.summary string-level
+                  parity is also currently lost (concrete numbers
+                  in legacy vs abstract factor labels in engine1).
+  Migration roadmap (now actionable for the next session):
+    (a) Extract `gradingOrchestrator.gradeProp()` into a standalone
+        `computeFeaturesForProp({player, stat_type, line, direction})`
+        that lifts the player-roster + feature-fetch + trap-detect +
+        consistency-score logic into a reusable callable.
+    (b) Wrap engine1 + gradeAdapter behind a thin
+        `analyzeViaEngine1(prop)` helper.
+    (c) Rewire one route at a time — `/api/analyze/prop` first
+        (lowest risk: single prop, public, well-tested), then
+        `/api/analyze/batch`, then `/api/scan/parlay`, then
+        `/api/bets/*`.
+    (d) Remove `grader.js` + `propAnalyzer.js` + the legacy book
+        adapters only after every consumer has migrated and a soak
+        period.
+  Session 7d's DEPRECATED banner on `src/services/grader.js` stays
+  put.
 
 - **[ARCH-2] Two circuit-breaker / rate-limiter modules.** Severity:
-  Low (documented in 6a). `src/services/{circuitBreaker.js,
-  rateLimiter.js}` (keyed registry, legacy) coexist with
-  `src/utils/rateLimiter.js` (factory, new). Consolidation is purely
-  cosmetic — both work. Scope: half-session. Status: open.
+  Low. Status: **DOCUMENTED in Session 7e.** Both legacy modules now
+  carry banners listing their three callers:
+    - `src/services/UnifiedOddsProvider.js`
+    - `src/services/adapters/ESPNAdapter.js`
+    - `src/services/adapters/PinnacleAdapter.js`
+  Full removal blocks on ARCH-1 Step 6 — the legacy book adapters
+  retire together with the legacy grading path.
 
 ### SEC — Security
 
@@ -466,10 +492,12 @@ No circular imports detected.
 ### DUP — Duplicates
 
 - **[DUP-1] `normalizeName()` exists twice.** Severity: Low. Status:
-  **DOCUMENTED in Session 7d.** Both implementations now carry
-  cross-reference comments explaining the divergence (script keeps
-  digits for legacy roster fields; trap detector strips them). Full
-  consolidation deferred until the script can drop its digit support.
+  **FIXED in Session 7e.** Consolidated into `src/utils/normalize.js`
+  with a `keepDigits` option (default false). `trapDetection.js`
+  imports the default; `scripts/populate-player-ids.js` wraps with
+  `keepDigits: true` via `normalizeRosterName`. 9 unit tests cover
+  accent strip, suffix removal, digit options, idempotency, null
+  input.
 
 - **[DUP-2] `oddsToImplied` etc. only live in `src/utils/odds.js`.**
   Confirmed not duplicated despite the `oddsNormalizer.js` neighbor —