Hardcoded Stats Audit · 2026-05-20

Single source of truth: src/lib/winning-strategy.ts (STRATEGY_PROOF, WINNING_STRATEGY), re-exported via src/config/strategy-proof.ts (STRATEGY_PROOF, STRATEGY_CIS, sharpeDisclosure, strategyHeadline, disclosureCopy).

Scope: src/app/methodologie/**, src/app/performance/**, src/app/docs/**, src/app/use-cases/**, src/components/{methodologie,landing,portfolio}/**, src/lib/i18n/**. Blog body and docs/method-review/** markdown audits out of scope (cf 91-blog-stale-stats audit).

Severity legend

• HIGH: visible on /performance, /methodologie, landing, use-cases (public).
• MED: dev surface (/docs API examples), feature-flagged stubs, secondary copy.
• LOW: code comment, dev playground.

Findings

file:line	current value	STRATEGY_PROOF	drift	sev	status
Ch06Backtest.tsx:559-561 (was)	hardcoded "1.21 / +0.42 / +45.8 % / -14.1 % / 57.1 % / [-0.57, +4.08] / 2025-01-01 to 2026-05-19"	oosResults.{sharpeAnnualized,sharpeDeflated,maxDDPct,hitRatePct,sharpeCI95Lo/Hi,testWindow,calmar}	matches today, future drift	HIGH	FIXED (templated from STRATEGY_PROOF.oosResults)
Ch06Backtest.tsx:262 / ChDisclosure.tsx:108,114 (was)	"signalScore ≥ 50"	WINNING_STRATEGY.minScore = 40 (v3 scoring)	Y, +10 pts	HIGH	FIXED (replaced with "≥ 40")
ChDisclosure.tsx:198-199 (was)	"44,3 % WR, −0,72 % T+90" (full-universe quote)	crossMarketUniverse.{winRatePct:46.8, meanReturnT90Pct:0.78} AND srMarkets.FR.{wr:44.3, mean:-0.25}	Y on mean (-0.72 vs -0.25); 44.3 WR matches FR-only, not cross-market	HIGH	FIXED (uses crossMarketUniverse via existing universeWinRate/universeAvgT90)
performance/page.tsx:189-196 (was)	fallback strings "1.21" / "+46.2%" / "-14.1" / "2025-01-01 to 2026-05-19"	STRATEGY_PROOF.oosResults (always populated)	dead branch but drift risk	HIGH	FIXED (fallback now reads STRATEGY_PROOF.oosResults; CAGR fallback dropped to "·")
use-cases/quant-fund/page.tsx:89-90 (was)	"win rate à T+90 d'environ 67 % pour les signaux à score ≥ 65"	no such bucket in STRATEGY_PROOF	Y — unfounded claim	HIGH	FIXED (now cites winRate=77, avgReturn=13.2, minScore=40, n=196)
performance/page.tsx:704,714	"score composite ≥ 50 + insider de haut rang ou cluster"	WINNING_STRATEGY.minScore=40 (signal engine) vs harness ≥50 in lib/performance-data.ts:373,402	independent backtest cohort	MED	LEFT (intentional decoupling; ≥50 is the "Recommended Sigma" backtest filter, not the live engine threshold)
Ch06Backtest.tsx:77-86 EQUITY_PATH + +25% label	hand-drawn SVG curve, "+25%" anchor text	n/a — illustrative	n/a	LOW	LEFT (decorative; CAGR callout already reads STRATEGY_PROOF.avgReturn)
WalkForwardAnimator.tsx:26-32	full WF_STEPS stub (sharpe 1.21/1.44/0.87/1.63/1.19/1.38/0.94)	none	n/a, behind SHOW_WALK_FORWARD_ANIMATOR=false	MED	LEFT (feature-flagged; TODO already documents public/data/walk-forward-sharpe.json swap)
FormulaTypewriter.tsx:31	"+17.2 pts" sample	illustrative	n/a	LOW	LEFT
LiveScoringPlayground.tsx:438	(returnT90/20)*weights.returnT90 formula	illustrative	n/a	LOW	LEFT (toy widget)
performance/page.tsx:881	"+11 % CAGR" copy from "average T+90 compounded over ~4 cycles/year"	could compute from avgReturn	mild drift (13.2/4≈11 holds)	LOW	LEFT (illustrative arithmetic, not a stat citation)
docs/page.tsx:1074	example JSON "value": 0.42	unrelated to oos sharpeDeflated=0.42	coincidental	LOW	LEFT
PITLeakCheckGate.tsx:60-61	"k factors = 583,200" / "583 200"	hard literal in copy; matches GRID_SEARCH_TRIALS const in winning-strategy.ts:779	matches today	LOW	LEFT (could export GRID_SEARCH_TRIALS but the const is also a copy literal)
Ch08Strategie.tsx:212-213,474-475	"583 200 backtests" / "N=583,200" in copy	same	matches	LOW	LEFT (same reason)

Flags for Simon

1. use-cases/quant-fund cohort claim was fabricated. "67 % T+90 at score ≥ 65" did not exist anywhere in STRATEGY_PROOF, crossMarketUniverse, srMarkets, nor any audit doc. Replaced with the legitimate filtered-subset numbers. If a real backtest at score-bucket ≥ 65 exists, surface it explicitly in STRATEGY_PROOF (e.g. add scoreBuckets: Record<60|65|70|...>) before re-claiming it.

2. crossMarketUniverse.meanReturnT90Pct = 0.78 disagrees with srMarkets.FR.mean = -0.25. The ChDisclosure copy block previously said "−0,72 %" which matches neither value. The current refresh likely propagated crossMarketUniverse (cross-market positive) but kept the older FR-only narrative in one copy block. Verify which number Simon wants quoted in disclosure when the backtest is rerun.

3. signalScore ≥ 50 vs ≥ 40: two distinct gates exist in code:

   • WINNING_STRATEGY.minScore = 40 (live signal engine, scoring v3 threshold)
   • lib/performance-data.ts "Recommended Sigma" backtest filter ≥ 50 (and Ch01-style filters) Methodologie + ChDisclosure copy now say ≥ 40 (matching WINNING_STRATEGY); /performance "Recommended Sigma" still says ≥ 50 because that is the actual backtest harness threshold rendered. These should be aligned (either bump engine to 50, or rerun the recommended-Sigma backtest at 40) so users see one threshold, not two.

4. Performance hero fallback CAGR +46.2%: previous fallback exceeded the OOS Calmar-derived value (calmar × |maxDDPct| = 3.25 × 14.1 ≈ +45.8%). Fallback dropped; if monthlyPortfolio.meanPct ever goes null the surface degrades to "·" instead of a fake number.

5. exitRules.expectedHoldingDays = WINNING_STRATEGY.holdingDays = 90 but idealExitDays = 270 is shown on BUY cards. Confirm copy on /portfolio/DeclarationCard does not double-cite both as the same horizon (current code at DeclarationCard.tsx:330+ looks fine — uses idealExitDays only).

Files patched (5)

1. src/app/methodologie/_components/Ch06Backtest.tsx — OOS block templated from STRATEGY_PROOF.oosResults; signalScore ≥ 50 → ≥ 40.
2. src/app/methodologie/_components/ChDisclosure.tsx — "44,3 % WR, −0,72 % T+90" → templated from crossMarketUniverse; signalScore ≥ 50 → ≥ 40 (×2 locales).
3. src/app/performance/page.tsx — Hero fallback strings replaced with STRATEGY_PROOF.oosResults and "·".
4. src/app/use-cases/quant-fund/page.tsx — Fabricated 67% / score ≥ 65 claim replaced with STRATEGY_PROOF.winRate, avgReturn, filteredSubsetSize, WINNING_STRATEGY.minScore. Added STRATEGY_PROOF + WINNING_STRATEGY imports.

(Five surfaces touched; no other HIGH items found.)

Restant à faire (estimated effort)

• MED: swap WalkForwardAnimator.tsx stub for public/data/walk-forward-sharpe.json once harness 43 emits the file — ~30 min wiring + ~10 min visual QA.
• MED: align live engine threshold and "Recommended Sigma" backtest threshold (40 vs 50). Either rerun backtest at 40 (1h compute + audit refresh) or document the 50-threshold cohort explicitly in STRATEGY_PROOF.recommendedSigma as a separate sub-record (15 min code + audit cross-ref).
• LOW: export GRID_SEARCH_TRIALS (currently a module-local const) and consume in PITLeakCheckGate / Ch08Strategie copy — ~10 min cosmetic.

Verification

• npx tsc --noEmit — 0 errors after edits.
• No commits / pushes made.

---

Pass 2 · zones étendues (2026-05-20)

Scope extension beyond pass 1: full src/components/** (Hero, landing, methodologie, seo, portfolio, admin, email), src/lib/email*, src/lib/mcp/{tools,execute}.ts, src/app/api/v1/**, OpenAPI spec, sitemap + robots + hreflang surfaces, JSON-LD generator, CSS keyframes, and BlogArticle.bodyMdx (read-only via scripts/_audit-blog-pass2.ts, output /tmp/blog-stats-audit.csv).

Findings table

file:line	current value	canonical (STRATEGY_PROOF)	drift	sev	status
src/lib/mcp/tools.ts:142 (was)	hardcoded Sharpe annualisé 1.21 / CI95 [-0.57, +4.08] / CAGR +45.8 % / max-DD -14.1 % / hits 57.1 % / +0.42 / n=196 / 77 % / +13.2 % / 1.87 / ≥ 40	STRATEGY_PROOF.oosResults.* + STRATEGY_PROOF.{winRate,avgReturn,sharpeCrossSectional,filteredSubsetSize} + WINNING_STRATEGY.minScore	matches today, future drift	HIGH	FIXED (templated via IIFE)
src/lib/mcp/execute.ts:260 (was)	same triplet as above in get_winning_strategy_signals runtime description	same	matches today, drift risk	HIGH	FIXED (templated via IIFE)
src/app/api/v1/strategy/winning/route.ts:46 (was)	"score v3 ≥ 40 ... Horizon T+90"	WINNING_STRATEGY.{minScore,holdingDays}	matches	HIGH	FIXED (template literal)
src/app/api/v1/strategy/winning/route.ts:1-16 (was)	header comment cited +13.2% avg, alpha +10.1, Sharpe 1.87, 0.40, null	STRATEGY_PROOF.*	comment-only, but factually drift-prone (0.40 is today's sharpeAnnualized in-sample)	MED	FIXED (rewrote to reference fields, not literals)
src/components/landing/LandingSigma.tsx:134-135	... ?? 1.21).toFixed(2) / ?? 0.42).toFixed(2) fallback strings	STRATEGY_PROOF.oosResults.{sharpeAnnualized,sharpeDeflated} (always populated)	matches; fallback unreachable in practice	MED	LEFT (dead fallback, never hit — Pass-1 style decision)
src/components/methodologie/WalkForwardAnimator.tsx:26-32	WF_STEPS stub (1.21/1.44/0.87/1.63/1.19/1.38/0.94)	none	n/a, behind SHOW_WALK_FORWARD_ANIMATOR=false (see Pass-1)	MED	LEFT (already documented Pass-1 line 29)
src/components/methodologie/FormulaTypewriter.tsx:31	"+17.2 pts" illustrative	n/a	n/a	LOW	LEFT (already Pass-1)
src/components/methodologie/LiveScoringPlayground.tsx:438	(SAMPLE.returnT90/20) * weights.returnT90	n/a	toy widget	LOW	LEFT (already Pass-1)
src/components/methodologie/PITLeakCheckGate.tsx:60	"k factors = 583,200"	matches GRID_SEARCH_TRIALS const	matches	LOW	LEFT (Pass-1 noted)
src/components/SignalWeightPlayground.tsx:75	comment: "avg 13.2 %, win rate 77 %, Sharpe ~1.87"	matches current STRATEGY_PROOF	comment only, drift risk if rebaselined	LOW	LEFT (code comment, allowed)
src/components/landing/LandingTrackRecord.tsx:35-37	seed = [0.58, 0.62, …] 23-point sparkline	n/a — "fake historical win rate" explicitly labelled (line 34) + "illustratif/illustrative" tag at line 161	n/a	LOW	LEFT (illustrative + disclosed)
src/components/landing/LandingHowItWorks.tsx:169-171	LVMH +18.4%, Schneider +12.1%, Hermès +9.7% hardcoded deltas	n/a — demo cards	n/a	LOW	LEFT (illustrative example data)
src/components/BacktestDashboard.tsx:599-600,1068-1069	"~55-60%" CAC40 baseline copy	external baseline, not Sigma proof	n/a	LOW	LEFT (market reference)
src/components/HoldingPeriodGuide.tsx:61,70	"T+60 et T+90" exit guidance	WINNING_STRATEGY.holdingDays=90, STRATEGY_PROOF.idealExitDays=270	T+60 is an explicit narrative warning window, not a stat citation	LOW	LEFT
src/lib/seo/jsonld.ts:71,97	wordCount field	read from BlogArticle.wordCount DB column	n/a	LOW	OK (already DB-sourced)
CSS keyframes (globals.css, landing.css)	no 57%/77%/45%/14%/13% keyframe stops matching stat literals	n/a	n/a	n/a	CLEAN
Email templates (src/lib/email/templates/*, src/lib/email.ts, src/lib/weekly-digest.ts, src/lib/newsletter-recap.ts)	no hardcoded headline stats	n/a	n/a	n/a	CLEAN
OpenAPI spec (src/lib/openapi-spec.ts, src/app/api/openapi.json)	no headline stats cited in descriptions	n/a	n/a	n/a	CLEAN
Sitemaps + robots.ts + hreflang	URLs only, no stat literals	n/a	n/a	n/a	CLEAN
src/components/seo/{EntityCrosslinks,RelatedArticles}.tsx	no stat literals	n/a	n/a	n/a	CLEAN
src/components/portfolio/{CsvImport,DemoBanner}.tsx	no stat literals	n/a	n/a	n/a	CLEAN
src/components/admin/SourceDebugPanel.tsx	no stat literals	n/a	n/a	n/a	CLEAN
src/components/StrategyProofHeadline.tsx	comment cites 1.87*	reads STRATEGY_PROOF everywhere in render path	n/a	n/a	CLEAN (template-driven)

Blog body MDX (read-only, NOT patched)

Audit: scripts/_audit-blog-pass2.ts → /tmp/blog-stats-audit.csv (111 lines, 110 unique tuples).

• 110 unique (slug, locale, pattern, literal) tuples flagged across 104 published articles.
• 94 marked drift=Y, 16 drift=N. By locale: FR 56 drift / 8 stable · EN 38 drift / 8 stable.
• Drift breakdown by pattern: avgReturn 13.2% 31 · winRate 77% 29 · OOS Sharpe 1.21 19 · T+90 horizon 10 · filteredSubsetSize n=196 4 · sharpeDeflated 0.42 1.
• IMPORTANT caveat: the drift=Y count includes legitimate per-cohort numbers (e.g. women-insiders n=128, closely-associated-PDMR n=111 85.4%, sector-ETF 40.37%). The CSV preserves the 60-char context window on each row so Simon can triage legitimate alternate cohort vs stale Sigma headline misquoted.
• The dominant systematic stale block is still the V12 triplet -1.70 / 41.82 / -0.08 (see doc 91), unchanged by this pass: 34 category-A EN articles. Action plan remains the Hybrid B/A in doc 91.
• No bodyMdx row was modified in this pass (per instructions).

Files patched in pass 2 (4)

1. src/lib/mcp/tools.ts — get_winning_strategy_signals description templated from STRATEGY_PROOF.oosResults + headline fields + WINNING_STRATEGY.minScore. Added STRATEGY_PROOF/WINNING_STRATEGY imports and a header note banning hardcoded literals.
2. src/lib/mcp/execute.ts — runtime strategy.description returned by get_winning_strategy_signals templated the same way.
3. src/app/api/v1/strategy/winning/route.ts (line 46) — public REST description templated from WINNING_STRATEGY.{minScore,holdingDays}.
4. src/app/api/v1/strategy/winning/route.ts (header comment) — rewrote to reference STRATEGY_PROOF / WINNING_STRATEGY field names instead of brittle literals (0.40, 1.87, +10.1, +13.2).

Justifications for MED/LOW left untouched

• LandingSigma.tsx fallback string ?? 1.21 / ?? 0.42 is dead code (OOS struct is always populated at module init); changing it would touch a third file for zero behavioural win.
• WalkForwardAnimator.tsx stub: feature-flagged off (SHOW_WALK_FORWARD_ANIMATOR=false) and already TODO'd in Pass-1 row.
• FormulaTypewriter.tsx, LiveScoringPlayground.tsx, BacktestDashboard.tsx tooltips, LandingHowItWorks.tsx LVMH/Schneider/Hermès deltas, LandingTrackRecord.tsx sparkline seed: illustrative content explicitly labelled as such in the same component (or in surrounding section copy).
• SignalWeightPlayground.tsx:75 and winning-strategy.ts:21-31: docstring/JSDoc comments. CLAUDE.md em-dash + emoji policies explicitly allow literals in code comments.
• PITLeakCheckGate.tsx:60 (583,200) and Ch08Strategie.tsx mirror GRID_SEARCH_TRIALS: same const, identical literal — exporting and templating was already triaged as cosmetic-only in Pass-1.

Verification

• npx tsc --noEmit — 0 errors after Pass-2 edits.
• npm run lint:emdash — OK.
• npm run lint:emoji — OK (463 files scanned).
• No commits / pushes made.

Honest confidence score

Combined Pass-1 + Pass-2 estimate for the rendered surface (TSX, MCP, REST, emails, sitemaps, JSON-LD) being legacy-stat-free: ~92 %. Remaining 8 %:

• WalkForwardAnimator stub (behind flag) and the 4 illustrative-data callouts: ~2 %.
• Comment/docstring hardcoded literals that would become misleading after a STRATEGY_PROOF refresh (winning-strategy.ts header, SignalWeightPlayground.tsx:75, route.ts was the worst, now fixed): ~2 %.
• Decoupled threshold drift (Recommended Sigma ≥ 50 vs live engine ≥ 40) still unresolved (Pass-1 flag 3): ~2 %.
• Blog body MDX: ~35 % of EN published articles still recite V12 triplet — not counted in the rendered-surface estimate above but is the dominant remaining liability for any reader who clicks past the headline pages: ~2 %.

For the blog body MDX surface specifically, confidence is ~64 % (104 articles scanned, 37 carry the V12 triplet, but the headline pages do not consume those bodies). Plan: ship doc 91 remediation.