feat(ui): Loss Chance % heuristic chip (PR 4e) by dackclup · Pull Request #72 · dackclup/quantrank

dackclup · 2026-05-14T18:11:19Z

Summary

PR 4e — the second user-facing feature of the Phase 4 UX trio (after PR 4d recommendation badge). Adds a deterministic 5-95% "Loss Chance" chip to every stock — answers "if I buy at the current price, what's my heuristic chance of loss?"

Tag target on merge: v1.1.0-rc2 per v1-to-v1-1-migration/PLAN.md sequencing.

Pre-implementation audit + calibration

Per user request "ตรวจสอบ 4e ว่าทุกอย่างถูกต้องแล้ว", validated PLAN against:

✅ Option D terminology locked
✅ Schema additive only (loss_chance_pct: float | None)
✅ Display position locked (after MoS bar)
✅ 5-band outlined-light + no dark: variants (lesson PR #70)
✅ Threshold-relative tests (lesson PR 4d)

⚠ Calibration delta — the PLAN draft weights yielded universe-median LC of 64 with bands skewed Sell-side. Re-calibrated against live S&P 500 production data:

Constant	PLAN draft	Locked
`BASELINE_PCT`	50	40
`MOS_SCALE`	0.4 (symmetric ±25)	0.35 (asymmetric: cap_neg=35, cap_pos=20)
`COMPOSITE_SCALE`	/5	/2.0
`DATA_QUALITY_PENALTY`	20	15
`ALTMAN_PENALTY`	15	12
`GOING_CONCERN_PENALTY`	10	8

Predicted distribution

Applied to commit eb861e12 production data — universe median 49 (target ~50 ✅):

Band	Count	Target
<25% (Low)	7.6%	~10%
25-40%	20.9%	~20% ✅
40-60% (Neutral)	43.8%	~30%
60-80%	26.3%	~25% ✅
>80% (High)	1.4%	~15% (S&P 500 mostly fully-priced)

Per-recommendation tier median:

Bullish: 24 (target 10-25 ✅)
Lean Bullish: 42 (target 25-40)
Neutral: 56 (target 40-60 ✅)
Cautious: 62 (target 60-90 ✅)

Sample tier-aligned outputs:

HST LC=24 (Strong Buy, +14% MoS)
CF LC=28 (Buy, +9% MoS, beneish_high disqualifies SB)
NVDA LC=57 (Sell, -271% MoS, asymmetric cap limits LC)
WBD/SBUX/SJM LC=80+ (Sell, deep overvaluation + flags)

Files (+674 LOC)

Backend (4 files):

compute/scoring/loss_chance.py (new) — pure heuristic combiner, 13 exported constants
compute/output/schemas.py — loss_chance_pct field on StockSummary + StockDetail
compute/main.py — wires derive_loss_chance() after derive_recommendation()
tests/test_scoring/test_loss_chance.py (new, 21 tests)

Frontend (4 files):

frontend/components/LossChanceBadge.tsx (new) — 5-band outlined-light × 3 sizes; em-dash placeholder when MoS missing; italic "heuristic" qualifier on detail page
frontend/lib/types.ts — loss_chance_pct field
frontend/lib/schema-snapshot.json — regenerated
frontend/components/RankingTable.tsx — new "Loss Chance" column after MoS; mobile-card row
frontend/app/stock/[ticker]/page.tsx — new metric column in header card next to MoS

Test plan

ruff check . → All passed
python -m pytest tests/ -m "not network" → 772 passed (was 751 + 21 new)
python -m compute.output.schema_check → in sync
npx tsc --noEmit → clean
npx next build → clean (502 SSG pages)
CI green on this PR
Post-merge weekly run: Loss Chance % field populates on all 502; distribution matches simulation; chip renders in correct band per stock

Notes

No new third-party deps
Schema additive — null for legacy data (badge renders em-dash)
Outlined-light + no dark: variants — matches lesson learned from PR ui(badge): unify recommendation badge with sector/MoS outlined-light pattern #70
Threshold-relative tests — re-tuning constants won't break tests
Disclaimer: chip text says "heuristic" not "probability" because the rubric is deterministic, not backtested. Backtest-calibrated probability requires Phase 5+ Triple-Barrier + Conformal Prediction work.

Next in Phase 4 UX trio: PR 4f (price-chart-enhancements) — depends on PR 4d's recommendation field for target-price-line conditional.

https://claude.ai/code/session_015649aRyi2bvciQYZVNACd2

Generated by Claude Code

Phase 4 UX feature #2 of the trio (after recommendation-badge / PR 4d). Adds a deterministic 5-95% "Loss Chance" chip to every stock — answers "if I buy at the current price, what's my heuristic chance of loss?" Locked decisions (per phase-4-kickoff-checklist/PLAN.md §1): - Terminology: Option D — "Loss Chance %" + small italic "heuristic" qualifier on detail page; tooltip qualifier on ranking row - Display: after MoS bar on overview table + detail header card - 5-band color: outlined-light pattern per frontend-design-system Rule 2 (emerald-50/-50/slate-100/red-50/-50, NO `dark:` variants — lesson PR #70) - Schema additive: `loss_chance_pct: float | None` on StockSummary + StockDetail Calibration (locked 2026-05-14 post-simulation on live S&P 500): The PLAN's draft weights (baseline 50, mos_scale 0.4, dq=20, altman=15) yielded a universe-median LC of 64 with bands skewed Sell-side. Locked constants: BASELINE_PCT 40 (not 50) MOS_SCALE 0.35 (asymmetric caps) MOS_NEG_CAP 35 (positive MoS push-down) MOS_POS_CAP 20 (negative MoS push-up) COMPOSITE_SCALE 2.0 (each 10 composite-pts → 5 pp) DATA_QUALITY_PENALTY 15 (was 20) ALTMAN_PENALTY 12 (was 15) GOING_CONCERN_PENALTY 8 (was 10) SLOAN_NSI_PENALTY 3 (was 5, per flag) BENEISH_DECHOW_PENALTY 3 (was 5) Predicted distribution on commit eb861e1 production data: Universe median: 49 (target ~50 ✅) <25%: 7.6% (target ~10%) 25-40%: 20.9% (target ~20%) 40-60%: 43.8% (target ~30%) 60-80%: 26.3% (target ~25%) >80%: 1.4% (target ~15% — S&P 500 mostly fully-priced) Per-recommendation tier median: Bullish: 24 (target 10-25 ✅) Lean Bullish: 42 (target 25-40, slightly above) Neutral: 56 (target 40-60 ✅) Cautious: 62 (target 60-90 ✅) Files (+450 LOC): - compute/scoring/loss_chance.py (new) — pure heuristic combiner, ~13 exported constants for downstream tuning, [5, 95] honest clip - compute/output/schemas.py — `loss_chance_pct: float | None` on both StockSummary + StockDetail (additive, None for legacy data) - compute/main.py — wire derive_loss_chance() in per-ticker loop after derive_recommendation() - tests/test_scoring/test_loss_chance.py (new, 21 tests) — covers 5-band gradient, MoS asymmetric caps, clipping, flag additivity, input shape tolerance, tier-aligned expectations. Threshold-relative so future tuning doesn't break tests - frontend/lib/types.ts — mirror loss_chance_pct field - frontend/lib/schema-snapshot.json — regenerated - frontend/components/LossChanceBadge.tsx (new) — 5-band outlined-light badge, 3 sizes (xs/sm/md), optional italic "heuristic" qualifier for detail-page contexts, em-dash placeholder when MoS missing - frontend/components/RankingTable.tsx — new "Loss Chance" column after MoS on desktop table; new row after MoS-and-fair-price on mobile card - frontend/app/stock/[ticker]/page.tsx — new metric column in header card next to MoS, with showQualifier (small italic "heuristic" text) - .claude/skills/phase-4/loss-chance/PLAN.md — marked ✅ Implemented; documented calibration delta vs draft rubric No new third-party deps. No external API surface change beyond the single additive Pydantic / TypeScript field. Verification ladder: - ruff check . → All passed - python -m pytest tests/ -m "not network" → 772 passed (was 751 + 21 new) - python -m compute.output.schema_check → in sync - npx tsc --noEmit → clean - npx next build → clean (502 SSG pages) Tag target on merge: v1.1.0-rc2 per v1-to-v1-1-migration/PLAN.md sequencing. Next: PR 4f (price-chart-enhancements). https://claude.ai/code/session_015649aRyi2bvciQYZVNACd2

vercel · 2026-05-14T18:11:24Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
quantrank	Ready	Preview, Comment	May 14, 2026 6:11pm

Closes the silent-failure gap surfaced by Issue #287 PR A's Rule 18 instrumentation on cron Run #71 (368dccd, 2026-05-28 08:44 UTC). The PR #292 GOOG/GOOGL per-class XBRL share-override fix did not fire in production despite the code being correct on the runner. Root cause (edgar-debugger 2026-05-28 verdict): - PR #292 (e9aaab3, 04:22 UTC) landed the per-class XBRL override at compute/ingest/fundamentals.py:1043-1067 (Branch 3 of _build_snapshot). - Branch 3 only executes on live EDGAR fetch — fetch_fundamentals short-circuits at _is_fresh() (line 1292-1294) when cached parquet age by latest_filed_date < FUNDAMENTALS_REFETCH_DAYS = 45. - Earlier same-day cron 0ad1d57 (03:22 UTC, pre-PR-#292) wrote a stale aggregate parquet (GOOG shares_outstanding = 12.116B). - Cron Run #71 restored that parquet from the GitHub Actions cache; _is_fresh() returned True on latest_filed_date=2026-04-30 (28d < 45d), and Branch 3 never ran. - metadata.multi_class_per_class_attempt_count = 0 (PR #292 Rule 18 disambiguator working as designed — the smoking gun). - fundamentals_latency_p50_seconds = 0.0 (warm-cache replay confirmed universe-wide). Fix scope (6 files, YAML + paired test bump): - .github/workflows/compute-rankings.yml — 3 instances cache-v4- → cache-v5- (key + 2 restore-keys) + comment block expanded to cite Issue #288 follow-up + PR #292 + PR #269 + introduce a 2-trigger bump taxonomy (schema change OR value-correctness fix in live- fetch-only path). - .github/workflows/pre-merge-prod-sim.yml — mirror 3-string flip per the file's own "bump together if either changes" comment. Without this the simulate workflow would lose all 11 warm caches on every PR. - tests/test_workflow_cache_coverage.py — paired-test bump per the PR 4c.1 v3→v4 precedent. Function renamed test_workflow_cache_key_is_v4 → _v5; docstring rewritten to cite Issue #288 + PR #292 + the 3-trigger bump taxonomy. - CLAUDE.md §Phase status — drain stale "in flight" wording for PR #297 (now merged) + empirical-validation note for cron Run #71 + "in flight this PR" entry for the cache-v5 bump. - AGENTS.md open-issues list — update #287 (PR A merged via #297), #288 (fix in flight this PR), #289 (closed by PR #293). - PHASE_STATUS_INFLIGHT.md — full in-flight entry appended per PR #237 side-file convention. Why Option A (cache-key bump) over alternatives (per edgar-debugger): - Option B (targeted per-ticker invalidation): introduces cache-layer- knows-multi-class semantics + chicken-and-egg "detect stale aggregate from cached parquet" condition. - Option C (refactor override out of fetch path): cache hit triggers live SEC call (violates cache semantics) + FundamentalsSnapshot is frozen. - Option A: matches PR 4c.1 v3→v4 precedent exactly + zero compute/ change + guaranteed correctness on next cron. One-time cost: ~25-50 min cold cron on the immediately-following weekly run (full S&P 500 universe live re-fetch). Subsequent crons return to warm-cache ~5-10 min budget. No timeout-minutes impact — PR #297 just bumped to 195m which absorbs cold-cache reality. Verification on next cron Run #72: - metadata.multi_class_per_class_attempt_count = 2 (was 0) - metadata.multi_class_per_class_override_count = 2 - stocks/GOOG.json shares_outstanding ≈ 5.429B (Class C, was 12.116B) - stocks/GOOGL.json shares_outstanding ≈ 5.822B (Class A, was 12.116B) - stocks/GOOG.json market_cap ≈ $2.09T (was $4.66T) - stocks/GOOGL.json market_cap ≈ $2.59T (was $4.71T) - metadata.fundamentals_latency_p50_seconds > 0.0 (live fetch active) Adjacent findings deferred (NOT in this PR): - FOX / FOXA / NWS / NWSA: same multi_class_aggregate_shares_suspected annotate firing but they are on MULTI_CLASS_SHARE_ALLOWLIST (UNDERCOUNT path, PR #257). Decision on whether to add to overcount allowlist deferred to Q3 2026-08-19 quarterly cohort audit per methodology-scientist precedent (needs live XBRL probe). - OSAP wall-clock 347.1s on Run #71: cold OSAP download (cache > 31d mtime or evicted). Single observation; not a regression. Watch on next 2-3 crons. Hard constraints honored: - No compute / scoring / schema / valuation / Rule 16 / Top-5 invariant touched - No new defense flag · No new dep · No new env-var - YAML + paired-test diff (per quantrank-reviewer feedback on PR-title framing — original "YAML-only" was misleading) - Schema version UNCHANGED at 0.10.9-phase4.6 (no Pydantic / TS / snapshot change) Pre-push 3-reviewer gate: - phase-coordinator Mode B (sonnet): LOCKSTEP-SATISFIED — both CLAUDE.md + AGENTS.md substance touched, INFLIGHT entry well- shaped, branch in-sync with origin/main (no rebase needed) - quantrank-reviewer (opus): FIX-AND-RE-REVIEW → 2 FAIL + 4 WARN. Both FAILs fixed in this commit (tests/test_workflow_cache_coverage.py test pin + pre-merge-prod-sim.yml cache-v4 stragglers). WARN 1 (Issue #288 lifecycle) addressed via Reopens/Closes directives below. WARN 2 (AGENTS.md cron #69 cross-ref), WARN 3 (comment density) deferred — minor. Reopens #288 Closes #288 https://claude.ai/code/session_01AGU8d6pm4u2fQQ5cebg9qa Co-authored-by: Claude <noreply@anthropic.com>

…rkers + bump pointers (#299) Closes today's 10-PR cycle (#286 / #290 / #291 / #292 / #293 / #294 / #295 / #296 / #297 / #298). Mirror of PR #286 (post-v1.4.0 cycle drain) for the post-cron-#71 cycle. Three stale `(in flight, 2026-05-28)` markers in PHASE_STATUS_INFLIGHT.md drained to `(merged 2026-05-28, <SHA>)`: - PR #295 (`2d2ec83e`) — Post-session housekeeping drain 6 INFLIGHT + bump pointers - PR #297 (`ecb60e64`) — Issue #287 PR A: durable timeout + cache canary + per-loop wall-clock Metadata (schema 0.10.8 → 0.10.9-phase4.6) - PR #298 (`030675e9`) — Issue #288 follow-up: cache-key bump v4 → v5 Bodies preserved (historical record). CLAUDE.md §Phase status — drained the "(In flight this PR — cache-v5)" qualifier (PR #298 merged) + added post-PR-#298 confirmation note + cron Run #71 production-verified pointer. AGENTS.md open-issues list — #288 status flipped "(fix in flight this PR)" → "(closed by PR #298 cache-v5 bump)" + clarified the silent- failure root-cause + Run #72 verification gate. Why this PR exists: without end-of-day drain, session N+1 reading CLAUDE.md / PHASE_STATUS_INFLIGHT.md would see 3 PRs still marked "in flight" despite them merging hours earlier — the same friction pattern PR #286 closed for the post-v1.4.0 cycle. Three same-day drains in one PR keeps the side-file disciplined. Scope (3 files, doc-only): - PHASE_STATUS_INFLIGHT.md — 3 header substitutions + this PR's own in-flight entry appended per PR #237 side-file convention - CLAUDE.md §Phase status pointer refresh - AGENTS.md open-issues list #288 status update Hard constraints honored: - No code / scoring / schema / valuation / Rule 16 / Top-5 invariant touched - No new defense flag · No new dep · No new env-var - Doc-only diff (Markdown only) - Schema version UNCHANGED at 0.10.9-phase4.6 (no Pydantic / TS / snapshot change) PHASE_STATUS_INFLIGHT.md side-file satisfies §Conventions "ship with every PR" lockstep per PR #237 convention. Same drain template as PR #286 (post-v1.4.0 cycle). https://claude.ai/code/session_01AGU8d6pm4u2fQQ5cebg9qa Co-authored-by: Claude <noreply@anthropic.com>

… instrumentation Methodology-scientist Mode B Q2 follow-up deferred from PR #294 (sector- CoE flip, 2026-05-28 05:39 UTC). Adds `Metadata.value_trap_risk_delta_by_sector: dict[str, int] | None` so Q3 2026-08-19 quarterly cohort audit has visible per-sector shape evidence — not just the aggregate `value_trap_risk_count_*_sector_coe` scalars that landed in PR #204. Schema PATCH bump 0.10.9 → 0.10.10-phase4.6 (additive Metadata-only). Methodology context (Damodaran 2019 Ch. 8.4 §"Industry Beta"): After `USE_SECTOR_COE = True` per-sector Ke replaces the flat 10% baseline at SECTOR_COST_OF_EQUITY (11 GICS sectors, Ke 6%-12%). Directional predictions: - Lower-Ke sectors (Utilities ~6-7% / Real Estate ~7-8% / Consumer Staples ~7-8%): ROE ≥ Ke threshold relaxed → fewer RIM-skipped → POSITIVE delta (sector DROPPED flags) - Higher-Ke sectors (Information Technology ~11-12% / Energy ~10-12%): ROE ≥ Ke tightened → more RIM-skipped → NEGATIVE delta - Neutral sectors (6 GICS sectors at ~9-11%): small delta near zero Cron #69 + Run #71 universe-wide already confirmed the aggregate: 132 → 109 (−23 tickers, −17.4%). This PR breaks the −23 down by sector. Scope (10 files, additive only): - compute/output/schemas.py — new value_trap_risk_delta_by_sector field with full docstring (methodology-scientist verdict + Damodaran 2019 anchor + direction semantics) - frontend/lib/types.ts — mirror TS field as Record<string, number> | null - frontend/lib/schema-snapshot.json — regenerated via --update-snapshot - compute/config.py — SCHEMA_VERSION = "0.10.10-phase4.6" - compute/main.py — 3 surgical edits mirroring existing scalar dual-counter pattern (init two dict[str, int] counters / per-sector increment co-located with the existing scalar bump in both branches / delta computation in Metadata constructor) - tests/test_config.py — schema version pin bump + docstring rewrite - tests/test_output/test_value_trap_delta_by_sector_schema.py (NEW) — 2 active GREEN schema-contract tests (mirror test_wall_clock_schema.py pattern from PR #297) - CLAUDE.md — §Phase status pointer block refresh - AGENTS.md — open-issues #67 status: flip landed + per-sector follow-up in flight this PR - PHASE_STATUS_INFLIGHT.md — full in-flight entry per PR #237 side-file convention Implementation note: Per-sector dict construction uses `sorted(set(without) | set(with))` for stable key ordering; `.get(sec, 0)` fallback handles sectors appearing in only one path; `{} or None` falls back to None when both dicts are empty (test-mode universe). Co-located with the existing scalar bump in both `_rim_flat` (flat-Ke) and `_rim_sector` (sector-Ke) branches at the same `value_trap_risk_roe_below_cost_of_equity` reason guard — scalar and dict always stay in lockstep. Verification ladder: - ruff check . PASS - python -m compute.output.schema_check PASS (triple in sync 0.10.10) - pytest tests/test_config.py -v 11/11 PASS (pin held) - python -m pytest tests/test_output/ 2/2 NEW PASS - Full offline suite via test-engineer 1367 → 1369 (+2 NEW) Pre-push 3-reviewer gate: - schema-sentinel (sonnet) PASS (52 fields, triple aligned, PATCH bump correct, snapshot alphabetical ordering held) - test-engineer (sonnet) GREEN (2/2 new tests pass, 1367 → 1369, 0 regressions, 0 skipped stubs) - quantrank-reviewer (opus) READY-TO-PUSH (0 FAIL, 4 WARN all pre-existing PR-#297-era drift, defer to next housekeeping PR — incl. SKILL.md/PHASE_STATUS.md schema-table tops still on 0.10.8) Empirical validation gate (post-merge, next cron Run #72): - metadata.value_trap_risk_delta_by_sector populates as non-null dict - Damodaran shape directionally correct: Util/Real Estate/Staples POSITIVE, Information Technology/Energy NEGATIVE - sum(delta.values()) == without_sector_coe_count - with_sector_coe_count (= 23 per Run #71 universe-wide; matches within rounding) Note: per-sector accumulation runs in the Step 8 per-ticker loop, INDEPENDENT of cache-v5 cache busting (PR #298). Field populates on next cron regardless of warm/cold fetch path. Hard constraints honored: - No new defense flag · No scoring formula change · No Rule 16 / Top-5 violation - Additive-only schema change (PATCH bump) - Field nullable per Rule 18 graceful-degradation - Phase 4.5e PR 5 (cluster weight promotion) gate-data UNCHANGED — independent track Methodology decision: methodology-scientist verdict NOT re-requested — this is the EXACT field shape Mode B Q2 verdict from PR #294 explicitly authorized. Future re-trigger only if post-merge cron shows sector breakdown contradicting Damodaran prediction OR Q3 2026-08-19 audit reads ≥ 6 crons of data and per-sector decay pattern needs interpretation. https://claude.ai/code/session_01AGU8d6pm4u2fQQ5cebg9qa

…ta instrumentation (#300) Methodology-scientist Mode B Q2 follow-up deferred from PR #294 (sector- CoE flip, 2026-05-28 05:39 UTC). Adds `Metadata.value_trap_risk_delta_by_sector: dict[str, int] | None` so Q3 2026-08-19 quarterly cohort audit has visible per-sector shape evidence — not just the aggregate `value_trap_risk_count_*_sector_coe` scalars that landed in PR #204. Schema PATCH bump 0.10.9 → 0.10.10-phase4.6 (additive Metadata-only). Methodology context (Damodaran 2019 Ch. 8.4 §"Industry Beta"): After `USE_SECTOR_COE = True` per-sector Ke replaces the flat 10% baseline at SECTOR_COST_OF_EQUITY (11 GICS sectors, Ke 6%-12%). Directional predictions: - Lower-Ke sectors (Utilities ~6-7% / Real Estate ~7-8% / Consumer Staples ~7-8%): ROE ≥ Ke threshold relaxed → fewer RIM-skipped → POSITIVE delta (sector DROPPED flags) - Higher-Ke sectors (Information Technology ~11-12% / Energy ~10-12%): ROE ≥ Ke tightened → more RIM-skipped → NEGATIVE delta - Neutral sectors (6 GICS sectors at ~9-11%): small delta near zero Cron #69 + Run #71 universe-wide already confirmed the aggregate: 132 → 109 (−23 tickers, −17.4%). This PR breaks the −23 down by sector. Scope (10 files, additive only): - compute/output/schemas.py — new value_trap_risk_delta_by_sector field with full docstring (methodology-scientist verdict + Damodaran 2019 anchor + direction semantics) - frontend/lib/types.ts — mirror TS field as Record<string, number> | null - frontend/lib/schema-snapshot.json — regenerated via --update-snapshot - compute/config.py — SCHEMA_VERSION = "0.10.10-phase4.6" - compute/main.py — 3 surgical edits mirroring existing scalar dual-counter pattern (init two dict[str, int] counters / per-sector increment co-located with the existing scalar bump in both branches / delta computation in Metadata constructor) - tests/test_config.py — schema version pin bump + docstring rewrite - tests/test_output/test_value_trap_delta_by_sector_schema.py (NEW) — 2 active GREEN schema-contract tests (mirror test_wall_clock_schema.py pattern from PR #297) - CLAUDE.md — §Phase status pointer block refresh - AGENTS.md — open-issues #67 status: flip landed + per-sector follow-up in flight this PR - PHASE_STATUS_INFLIGHT.md — full in-flight entry per PR #237 side-file convention Implementation note: Per-sector dict construction uses `sorted(set(without) | set(with))` for stable key ordering; `.get(sec, 0)` fallback handles sectors appearing in only one path; `{} or None` falls back to None when both dicts are empty (test-mode universe). Co-located with the existing scalar bump in both `_rim_flat` (flat-Ke) and `_rim_sector` (sector-Ke) branches at the same `value_trap_risk_roe_below_cost_of_equity` reason guard — scalar and dict always stay in lockstep. Verification ladder: - ruff check . PASS - python -m compute.output.schema_check PASS (triple in sync 0.10.10) - pytest tests/test_config.py -v 11/11 PASS (pin held) - python -m pytest tests/test_output/ 2/2 NEW PASS - Full offline suite via test-engineer 1367 → 1369 (+2 NEW) Pre-push 3-reviewer gate: - schema-sentinel (sonnet) PASS (52 fields, triple aligned, PATCH bump correct, snapshot alphabetical ordering held) - test-engineer (sonnet) GREEN (2/2 new tests pass, 1367 → 1369, 0 regressions, 0 skipped stubs) - quantrank-reviewer (opus) READY-TO-PUSH (0 FAIL, 4 WARN all pre-existing PR-#297-era drift, defer to next housekeeping PR — incl. SKILL.md/PHASE_STATUS.md schema-table tops still on 0.10.8) Empirical validation gate (post-merge, next cron Run #72): - metadata.value_trap_risk_delta_by_sector populates as non-null dict - Damodaran shape directionally correct: Util/Real Estate/Staples POSITIVE, Information Technology/Energy NEGATIVE - sum(delta.values()) == without_sector_coe_count - with_sector_coe_count (= 23 per Run #71 universe-wide; matches within rounding) Note: per-sector accumulation runs in the Step 8 per-ticker loop, INDEPENDENT of cache-v5 cache busting (PR #298). Field populates on next cron regardless of warm/cold fetch path. Hard constraints honored: - No new defense flag · No scoring formula change · No Rule 16 / Top-5 violation - Additive-only schema change (PATCH bump) - Field nullable per Rule 18 graceful-degradation - Phase 4.5e PR 5 (cluster weight promotion) gate-data UNCHANGED — independent track Methodology decision: methodology-scientist verdict NOT re-requested — this is the EXACT field shape Mode B Q2 verdict from PR #294 explicitly authorized. Future re-trigger only if post-merge cron shows sector breakdown contradicting Damodaran prediction OR Q3 2026-08-19 audit reads ≥ 6 crons of data and per-sector decay pattern needs interpretation. https://claude.ai/code/session_01AGU8d6pm4u2fQQ5cebg9qa Co-authored-by: Claude <noreply@anthropic.com>

quantrank-reviewer (opus) flagged 3 non-blocking WARNs on the initial PR 6 commit. 2 addressed inline; 1 deferred to follow-up. - WARN-2 FIXED: `_NEGATION_REGEX` BEFORE branch gains an inline NOTE block explaining `no` is intentionally BEFORE-only (post- mention "10b5-1 plan, no shares sold" reads as affirmative disclosure + non-negation use of `no` → FP risk too high). AFTER branch carries a cross-reference comment so the asymmetry is self-documenting at the regex. - WARN-3 FIXED: new `test_M3_negation_patterns_each_appear_in_compiled_regex` drift-detector — every token in `_NEGATION_PATTERNS` must appear somewhere in `_NEGATION_REGEX.pattern` source string. Catches drift in both directions (add to frozenset without updating regex; remove from regex without updating frozenset). Carve-outs handle the `cancelled` ↔ `canceled` collapse to `cancell?ed` and the multi-word `not in effect` `\s+` split. Tests 33 → 34; full suite 1399 → 1400. - WARN-1 DEFERRED in `PHASE_STATUS_INFLIGHT.md` PR #303 entry: `_NEGATION_REGEX` anchor matches only 2 of upstream `detect_10b5_1_plan`'s 6 substrings. Bias direction remains safe (over-includes legit trades in cohort, never under-excludes). Fix path noted; gated on cron Run #72+ empirical data showing whether the gap is material. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>

…al footgun #1) (#303) * feat(scoring): Phase 4.5e PR 6 — Form-4 10b5-1 negation guard (residual footgun #1) Closes the residual of footgun #1 from `compute/scoring/form4_signals.py` module docstring + the PR 4-eq Mode B verdict (2026-05-23, pre-approved "harden detect_10b5_1_plan with a negation guard against FP matches on phrases like '10b5-1 plan terminated'"). PR 6 implements the engineering of the approved mitigation. **Architecture** — post-detector wrapper: - `compute/scoring/form4_insider.py` gains `_NEGATION_PATTERNS` (11 tokens) + `_NEGATION_REGEX` (compiled bidirectional regex; case-insensitive; accepts both `10b5-1` and `10b-5-1` spellings; ±5 word-token window) + `_has_negation()` helper + thread-safe module-level counter - `_detect_10b5_1_on_transaction` modified: detector returns True → check `_has_negation(resolved_text)` → on match return False and bump the counter; detector returns False or None → pass through unchanged. Guard never fabricates a positive signal (only downgrades True → False). **Schema** `0.10.10 → 0.10.11-phase4.6` (PATCH — additive Metadata field). Triple touched: Pydantic + TypeScript + snapshot regenerated; verified clean via `python -m compute.output.schema_check`. **Rule 18 observability**: new `Metadata.form4_negation_guard_downgrade_count: int | None`. Counter is module-level + thread-safe (`threading.Lock` around an int) for the `EDGAR_MAX_WORKERS=8` parallel Form-4 fetch loop. `compute/main.py` resets before the `ThreadPoolExecutor` block and reads on the success path. `None` semantics mirrors `form4_wall_clock_seconds`: None when `FORM4_FETCH_SKIP=1` OR outer try/except fired. **Methodology** — pre-approved (no fresh consultation needed). Expected delta firing-rate per Cohen 2008 §III + Jagolinzer 2009 §3.2: `insider_sell_cluster` +5% to +10% relative on a universe-baseline cron (absolute << 1%; most 10b5-1 disclosures are affirmative). Bias direction reversed: was conservative (over-excluded legit opportunistic trades); now closer to ground truth (terminated + former plans no longer fire). **Tests** — 33 new tests in `tests/test_scoring/test_form4_negation_guard.py`: 13 pattern coverage (parametrized) + 5 negative paths + 2 ±5-token window boundary + 2 spelling variants + 5 `_detect_10b5_1_on_transaction` integration with mocked footnotes_dict + 2 thread-safety (incl. 100 concurrent bumps / 8 workers) + 2 Hypothesis properties (idempotence + monotonicity) + 2 manifest pins. Suite 1366 → 1399; zero regressions; ruff clean. `tests/test_config.py` schema pin bumped with PR 6 rationale. **Defense layer unchanged** at 33 declared boolean flags (PR hardens an existing input filter; no new flag). Cluster + C-suite weights UNCHANGED (5.0 / 3.0; promotion to 7.0 still gated on Q3 2026-08-19 audit per PR 4-eq verdict). **Closes**: residual of footgun #1 (Form-4 10b5-1 contamination) from `compute/scoring/form4_signals.py` module docstring. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * fix(scoring): PR 6 review WARN-2 + WARN-3 cleanup quantrank-reviewer (opus) flagged 3 non-blocking WARNs on the initial PR 6 commit. 2 addressed inline; 1 deferred to follow-up. - WARN-2 FIXED: `_NEGATION_REGEX` BEFORE branch gains an inline NOTE block explaining `no` is intentionally BEFORE-only (post- mention "10b5-1 plan, no shares sold" reads as affirmative disclosure + non-negation use of `no` → FP risk too high). AFTER branch carries a cross-reference comment so the asymmetry is self-documenting at the regex. - WARN-3 FIXED: new `test_M3_negation_patterns_each_appear_in_compiled_regex` drift-detector — every token in `_NEGATION_PATTERNS` must appear somewhere in `_NEGATION_REGEX.pattern` source string. Catches drift in both directions (add to frozenset without updating regex; remove from regex without updating frozenset). Carve-outs handle the `cancelled` ↔ `canceled` collapse to `cancell?ed` and the multi-word `not in effect` `\s+` split. Tests 33 → 34; full suite 1399 → 1400. - WARN-1 DEFERRED in `PHASE_STATUS_INFLIGHT.md` PR #303 entry: `_NEGATION_REGEX` anchor matches only 2 of upstream `detect_10b5_1_plan`'s 6 substrings. Bias direction remains safe (over-includes legit trades in cohort, never under-excludes). Fix path noted; gated on cron Run #72+ empirical data showing whether the gap is material. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com> * docs(phase-status): bump for PR #303 — schema 0.10.10 → 0.10.11 + Recently merged refresh (PR #300/#301/#302) * docs(workflow): bump session-start pointer to schema 0.10.10 on main + PR #303 in-flight (Phase 4.5e PR 6 negation guard 0.10.11) * test(form4): PR 6 cross-reference smoke tests in test_form4_insider.py — quality-gate Section D * ci: re-trigger simulate on 0a1fc86 (prior run cancelled by concurrency supersede) --------- Co-authored-by: Claude <noreply@anthropic.com>

vercel Bot deployed to Preview May 14, 2026 18:11 View deployment

dackclup marked this pull request as ready for review May 15, 2026 01:03

dackclup merged commit e47fa37 into main May 15, 2026
4 checks passed

dackclup deleted the feat/phase-4e-loss-chance branch May 15, 2026 01:04

dackclup mentioned this pull request May 15, 2026

Hero card 3-column row: vertical baseline misalignment + Loss Chance chip overflow (mobile) #77

Closed

4 tasks

dackclup mentioned this pull request May 28, 2026

fix(ci): Issue #288 follow-up — bump workflow cache key v4 → v5 #298

Merged

6 tasks

dackclup mentioned this pull request May 28, 2026

chore(docs): end-of-day housekeeping 2026-05-28 — drain 3 INFLIGHT markers + bump pointers #299

Merged

4 tasks

dackclup mentioned this pull request May 28, 2026

feat(scoring): Issue #67 follow-up — per-sector value_trap_risk delta instrumentation #300

Merged

7 tasks

dackclup mentioned this pull request May 28, 2026

feat(scoring): Phase 4.5e PR 6 — Form-4 10b5-1 negation guard (residual footgun #1) #303

Merged

9 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(ui): Loss Chance % heuristic chip (PR 4e)#72

feat(ui): Loss Chance % heuristic chip (PR 4e)#72
dackclup merged 1 commit into
mainfrom
feat/phase-4e-loss-chance

dackclup commented May 14, 2026 •

edited

Loading

Uh oh!

vercel Bot commented May 14, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

dackclup commented May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Pre-implementation audit + calibration

Predicted distribution

Files (+674 LOC)

Test plan

Notes

Uh oh!

vercel Bot commented May 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

dackclup commented May 14, 2026 •

edited

Loading

vercel Bot commented May 14, 2026 •

edited

Loading