← Updates feed  |  /audit dashboard

Asset-Class Action Items — Graphify Cross-Reference (2026-05-15)

Top action items per asset class, derived by querying the 200K-node Graphify knowledge graph (whole-repo code + docs) and cross-referencing against the May-15-2026 master plans: reports/SUPREME_PLAN_90days.md, FOOLPROOF_ACTION_PLAN.md, and the eight reports/asset_class_90day_plan_*_2026-05-15.md linked from findtorontoevents.ca/updates/index.html.

Live performance source: audit_dashboard/data/dashboard_data.json::performance.asset_class_health (verdict-grade, post-noise-filter).

✅ VERIFIED 2026-05-15. Every claim below was independently re-checked against primary code + JSON by 3 adversarial subagents (forbidden from citing plan docs) plus a main-thread reproduction. Corrections applied inline. Full method + evidence: reports/asset_class_verification_2026-05-15.md (incl. a deterministic-reproduction proof, tools/_verify_n_reproducible.py).

Scoreboard vs the plans

Class Live PF / WR / n Plan headline Drift
COMMODITY 2.37 / 60.7% / 326 PF 2.49 (plan) · 4.03 (FOOLPROOF) plans understate maturity — WR already clears Tier-1 floor
EQUITY 1.56 / 51.8% / 423 PF 1.57 / n=420 accurate — no material drift
ETF 1.33 / 57.4% / 108 PF 1.48 / n=106 PF fell 0.15, further from T2
CRYPTO 1.29 / 46.1% / 8108 "quan_engine 18% @ PF 0.70" stale villain — quan_engine already capped to 5%
FOREX 0.79 / 51.6% / 347 PF 0.27 / n=1169 massively stale — live is 3× better but still sub-floor
BOND 0.66 / 54.5% / 11 FOOLPROOF: PF 1.72 "meets T2" falsified — legacy mis-tagged futures rows
FUTURES none / 0 / 0 FOOLPROOF: "no strategies" factually wrong — 4 strategies coded + wired
PENNY/MEME (folded into EQUITY/CRYPTO) "class-wide block exists" wrong — only leaky strategy-pair blocks

Cross-cutting findings (apply to multiple classes)

  1. kill_gate.py (M-055) is wired into the kill switches, NOT the active gate. Callers: commodity_kill_switch.py, fx_kill_switch.py, policy_backtest.py. Gap: quality_gates.passes_active_gate does not consult it. Wire evaluate_kill() there so no strategy block fires below the per-class min-n floor.
  2. n is reproducible — but two metrics get conflated (corrected 2026-05-15). The verdict n (asset_class_health.n = wins+losses, post-_is_valid_resolved_pick filter) is a deterministic pure-function output — proven: 3 identical SHA256 runs (tools/_verify_n_reproducible.py). What drifts is reports citing raw closed (e.g. COMMODITY closed=513) as if it were the verdict n (326). Fix is naming discipline: cite resolved_n everywhere; stop quoting raw closed. Not a reproducibility crisis.
  3. FOOLPROOF_ACTION_PLAN.md is out of sync with the eight 90-day plans — wrong on FUTURES ("no strategies"), BOND ("meets T2"), CRYPTO ("cap quan_engine" already done), ETF ("scanner wired" but emits 0). Reconcile or supersede it.
  4. FRED_API_KEY unset (SKIP_FRED=1) blocks curve/macro data for BOND + EQUITY + COMMODITY at once — shared-infra P0, not a per-class task.

COMMODITY — PF 2.37 / WR 60.7% / n=326 (pushing Tier-1)

WR 60.7% already clears the Tier-1 55% floor. The plans frame COMMODITY as "not production-grade" — the real remaining gap is MDD verification + de-concentration, not edge.

⚠ Verified caveat (2026-05-15): the headline PF 2.37 / n=326 is itself suspect. The cot_positioning dedup ledger only landed PR #994 (2026-05-14) — n=326 carries pre-dedup COT over-emission. The COT pilot's own code comments record WR 90.1%→40% and n=101→5 once de-duplicated. Re-derive COMMODITY PF/WR on post-PR-#994 picks before any Tier-1 / real-money claim.

Top action items 1. Re-aggregate COT over-emission on historical reads — cot_positioning::CT=F still fires one weekly CFTC signal ~20×/cycle; headline PF carries inflated pre-2026-05-13 data. dashboard_generator.py lacks 1-per-(symbol, COT_date, direction) dedup on historical reads. 2. Tighten the COMMODITY concentration cap — concentration_cap.py is wired (quality_gates.py:4851) but CT=F still holds 73% PnL mass. Add hard per-symbol ≤30% / per-strategy ≤25%. 3. Wire commodity_carry_momo_double_sort — audit_dashboard/data/commodity_carry_momo.json (18 symbols, Miffre/Fuertes) has zero code callers. It is the natural diversifier off CT=F. New caller in alpha_engine/scanner.py. 4. Add a per-symbol futures liquidity/MDD gate in quality_gates.py — none exists; CT=F has no micro-contract. Prerequisite for real-money.

Further enhancements (not in any plan): MDD measured on independent COT cycles (not hourly bars) inside kill_gate.py; DBMF/KMLM external-replication CI benchmark (tools/commodity_external_replication.py).

EQUITY — PF 1.56 / WR 51.8% / n=423 (locking Tier-2)

Live numbers confirm the plan baseline — plans correctly diagnose the levers.

Top action items 1. Merge the VIX-regime hard filter — branch feat/equity-vix-regime-gate-sidecar-2026-05-13 is unmerged; non_crypto_quality_gate.py only does soft VIX confidence scaling. Research (equity_vix_regime_breakthrough_20260513.md): VIX<22 lifts PF to 4.55+. Single biggest consolidation lever. 2. Split the universe — config.py:587 EQUITY_SYMBOLS mixes 10 large-caps with 8 pennies/memes (NIO/LCID/RIVN/GME/AMC). Add LARGE_CAP_EQUITY_SYMBOLS + is_liquid_equity(). Penny gap-risk is the WR drag. 3. Wire a PEAD strategy (M-009) — equity_factor_model.py has an earnings boost but no standalone post-earnings-drift strategy. New alpha_engine/strategies/pead_equity.py. 4. Wire an execution-cost/slippage model — PF 1.56 is gross; Tier-2 must hold net of 5-20bp friction (M-017/018 scaffolds unwired).

Further enhancements: reconcile the two VIX thresholds (hard-block VIX>40 vs momentum-skip VIX>22) in one place; FRED macro overlay via fred_macro_context.py; per-class ml_score≥55 hard gate (gatekeeper is 83%+ accurate but unused as a gate).

ETF — PF 1.33 / WR 57.4% / n=108 (slipping from Tier-2)

PF fell 0.15 below the plan's 1.48 baseline — the "clear path to T2" framing is now over-optimistic.

Top action items 1. Debug the ETF sector emitter's zero-pick output — corrected 2026-05-15: tools/etf_sector_emitter.py:98 defaults ETF_SECTOR_EMITTER_ENABLED="1" (already ON, not opt-in). It ran 2026-05-15 and wrote etf_sector_picks.json with picks:[]. This is a silent failure — the fix is diagnosing why it emits nothing, not "enabling" it. 2. Enforce a VIX<25 gate on the ETF emit path — vix_regime_gate.py supports ETF but isn't default-on; backtest PF 2.05→3.22 with it. 3. Add a per-symbol concentration cap (<25%) — XLE is 20.55% share / 53.74% PnL. 4. Cap/prune intermarket-flow-scout if its ETF-slice PF < 1.4 — it dilutes the cleaner rotation edge.

Further enhancements: per-source ETF entry in per_source_volume_cap.py; positive source bonus for etf_sector_momentum/etf_dual_momentum in quality_gates.py; backtest a quarterly (vs monthly) rebalance to cut friction; dashboard PF-reconciliation tile (rotation vs mixed).

CRYPTO — PF 1.29 / WR 46.1% / n=8108 (volume-weighted dilution)

The plans chase a stale quan_engine villain — it is already capped to 5% and fully blocked. The live drag has shifted.

Top action items 1. Cap luxalgo_filters CRYPTO volume share — per_source_volume_cap.py:24-25 caps only quan_engine; luxalgo_filters is uncapped. Verified live numbers (2026-05-15): PF 1.12, WR 44.3%, 1,421 resolved ≈ 17.5% of CRYPTO n — marginally profitable but dilutive (below class PF 1.29). Add {"luxalgo_filters": {"CRYPTO": 0.10}}. (Earlier "~23% vol / PF ~1.0" was overstated.) Also fix the quan_engine cap desync: per_source_volume_cap.py says 5%, quarantine_manifest.json + tests say 12% — pin one value. 2. Give enforce_cap() a second caller — it is INTAKE-only, called only from smart_picks_engine.py; picks via production_scanner.py bypass it. Wire it into production_scanner.py. 3. Pair-block remaining sub-PF-1.0 sources (alpha_engine generic PF 0.99, copy_trader_highscore 0.80, battleground 0.65, regime_terminal 0.95) in quality_gates.py:1494 BLOCKED_SOURCE_SYSTEMS. 4. Build a crypto liquidity/ADV gate — no is_liquid_crypto() exists; stop emission on ~179-symbol meme/alt noise. 5. Fix the ml_crypto_pred 95.3% resolver-exclusion in forward_validator.py (808/848 closed picks unresolved hides rot).

Further enhancements: LONG-only volume penalty for luxalgo_confluence (35% WR LONG); replace per-source whack-a-mole with a class-level PF-weighted volume budget in quality_gates.py (auto-starves any sub-1.0 source).

FOREX — PF 0.79 / WR 51.6% / n=347 (sub-floor, mutate-before-kill)

Live PF is 3× the stale plan figure (0.27) but still losing. The fix is mutation, not new volume — and it is a small PR, not a 90-day project.

Top action items 1. Wire a directional gate — autopsy (forex_mutation_autopsy_20260515.md): LONG = 29.4% WR / PF 0.80, SHORT = PF 8.11. BLOCKED_ASSET_STRATEGY_TRIPLES (quality_gates.py:2045) exists but has zero FOREX entries. Highest-leverage move; pure mutation. 2. Wire a BLOCKED_SYMBOLS_BY_CLASS["FOREX"] allowlist — autopsy KILLs NZDUSD=X/EURJPY=X/USDCHF=X, BOOSTs AUDUSD=X/AUDJPY=X; the symbol gate is unwired. 3. Replace the COT proxy — cot_positioning_forex (forex_strategies.py:536) is a price-zscore proxy, not real CFTC 6E/6B/6J data — disguised mean-reversion. Wire free CFTC legacy COT. 4. Enforce a <25% concentration cap — USDJPY=X is 36% PnL mass via cta_fx_multifactor.

Further enhancements: make n-reconciliation a P0 (FOREX n is unreproducible); use MyFXBook/ZuluTrade retail sentiment as a contrarian gate input, not just copy-replication; run a 4th mutation axis — time-of-day / session (London-NY overlap vs Asia) — never tried for FOREX.

BOND — PF 0.66 / WR 54.5% / n=11 (n is the blocker, not edge)

n=11 is the problem. FOOLPROOF's "PF 1.72, meets T2" is falsified — legacy mis-tagged ZN=F futures rows that rolled off.

Top action items 1. Lower BOND_ELITE_FLOOR to 32-35 (bond-agent.yml) — the emitter produces ~10 raw signals/day but quality=0 because a crypto-calibrated elite floor of 40 is unreachable for low-vol bond signals. This is the n=11 root cause. 2. Wire the 3 spec'd pilots (TIPS-breakeven MR, Cochrane-Piazzesi curve carry, HYG-LQD credit MR) from bond_deep_dive_round2_2026-05-13.md into bond_strategies.py — uses the unused IEF/SHY/TIP/LQD universe (~114 events/yr → reaches n≥80). 3. Confirm BR-3 merge lands picks in active_picks.json (blocked upstream by #1). 4. Do NOT run kill_gate against BOND — n=11 < MIN_N=30 → it correctly returns INSUFFICIENT_EVIDENCE. There is nothing to kill, only an emitter to unblock.

Further enhancements: set FRED_API_KEY (shared-infra, unblocks 5 classes); frozen-snapshot mechanism for reproducible BOND verdicts; wire Kalshi rate/macro gating (kalshi_signals.py, no-key API) for duration timing.

FUTURES — no data (alive but starved, NOT dead)

4 strategies (futures_tsmom, futures_connors_rsi2, futures_cross_asset_momentum, futures_vol_regime_breakout) are coded in alpha_engine/futures_strategies.py and wired into non_crypto_agent/main.py. FOOLPROOF's "no strategies, no scanner" is factually wrong.

Top action items 1. Decide merge-vs-retire formally — plans recommend merging into a unified tile; execute or formally retire. Do not leave a zombie tile. 2. Fix classification — dashboard_generator.py:3168-3197 routes most =F roots to COMMODITY; only ES/NQ/ZN-type go to FUTURES, starving the tile. Add a contract_type tag. 3. Lower conf_floor from 0.50 → 0.40 for the 4 unvalidated strategies (non_crypto_agent/main.py:431) so signals reach shadow emission and accrue n — currently a self-fulfilling kill loop.

Further enhancements: non-empty-results assertion on the futures backtest job (new_strategies/futures_regime_report.json is empty-bodied); rename the unified tile FUTURES_CTA for honesty; per-gate rejection-reason telemetry in curate_quality_picks.

PENNY/MEME — folded into EQUITY/CRYPTO (leaky quarantine)

Not a separate asset_class_health key. quality_gates.py:1933-1974 has a MEMECOIN quarantine but it is strategy-pair-only (exact (MEMECOIN,strategy) tuples, no wildcard) — there is NO class-wide gate, despite the 90-day plan describing one as if it exists. Verified worse (2026-05-15): the string PENNY_STOCK does not appear in quality_gates.py at all — penny stocks have zero gating, not even leaky pair-blocks.

Top action items 1. Add a class-wide gate — class in ("MEMECOIN","PENNY_STOCK") → fail in quality_gates.py + non_crypto_quality_gate.py. Replaces the leaky pair list. 2. Add is_low_quality_or_meme(symbol) to config.py (mktcap<$2B / ADV<$5M), gated in scanner.py/production_scanner.py before emit. 3. Map CATEGORY_RISK "meme"/"penny" to BLOCK (config.py:173) — currently -15%SL/+35%TP amplifies losses. 4. Deprecate penny_volume_breakout + community_penny_volume_surge (return []).

Further enhancements: n<20 → "noise" pill on the dashboard symbol-perf render (small-n meme winners display as 100% WR); remove penny/meme from ml_ranker.py:420 CATEGORY_MAP; gap-risk circuit breaker (no overnight holds for float<10M).

Priority stack (ranked, cross-class)

  1. Standardize the cited verdict metric — name it resolved_n everywhere; stop reports quoting raw closed. The metric is already deterministic; this is a citation-discipline fix, not a reproducibility fix.
  2. Merge the EQUITY VIX-regime gate — biggest single PF lever, branch ready.
  3. Cap luxalgo_filters + second enforce_cap() caller — CRYPTO's real drag.
  4. Lower BOND_ELITE_FLOOR — one-line change that makes BOND solvable.
  5. Wire the FOREX directional gate — small PR, structure already exists.
  6. Class-wide PENNY/MEME gate — closes the biggest live leak.
  7. Fix FUTURES classification + conf_floor — ends the self-fulfilling kill loop.
  8. COMMODITY COT re-aggregation + concentration cap — makes the Tier-1 push real.
  9. Wire kill_gate.evaluate_kill() into quality_gates.passes_active_gate.
  10. Set FRED_API_KEY — unblocks macro data for 3 classes at once.

Generated 2026-05-15 from the Graphify whole-repo knowledge graph + 4-agent plan cross-reference. Source plans: SUPREME_PLAN_90days.md, FOOLPROOF_ACTION_PLAN.md, asset_class_90day_plan*2026-05-15.md.