Run walk-forward CAGR-at-vol gate for qm-y8j-5 per qm-ghn0 pattern with datastore provenance

Objective

Execute the qm-ghn0-pattern walk-forward gate for experiment qm-y8j-5-max-cagr-at-vol (spec research/findings/specs/meta-target-cagr-vol-qm-y8j.5.json v1.1, meta-target-cagr-vol-qm-y8j-5 disclosure no_blog) as an expanding annual January strategy route per acceptance criteria: expanding walk-forward for both variants meta-target-cagr-sp500 (target vol = expanding realized vol of SPX ^GSPC price monthly) and meta-target-cagr-6040 (target vol = expanding vol of 60/40 0.6SPY + 0.4bond TR) via research/optimizer3_meta.run_max_cagr_at_vol_walk_forward (60m initial floor, 36m per-sleeve via eligible_strategy_ids, capped-simplex 5%/20%/max10, sample covariance ddof1, anti_overfit 0.002, monthly+static only, next_close one-bar lag) with costs 10bps base / 20bps stress / 0bps sensitivity, reporting CAGR-at-vol vs vol-matched benchmarks, Sharpe/Sortino/MaxDD/turnover, vol-band 0.8-1.2x compliance, calibration 1991-1999 vs untouched 2000-2026, regime/neighbor/extra-lag/cost gates, and public-input-provenance/v1 hashes matching data/market.sqlite; no promotion until G11-G12 gates pass.

Approach

1. Froze expanding protocol per locked research/findings/specs/optimizer-3-meta-contract-v1.json (max_strategies 10, min_weight 0.05 proportional redistribute, max_weight 0.20 =2/n, monthly+static only exclude_daily, sample covariance Ledoit-Wolf false, anti_overfit mu 0.002 soft, starts 32 workers 4 seed 0, schedule year_end_select_trade_end_jan_to_end_jan with one-month lag hold) with expanding_pre_signal_history strictly before each January month-end signal and next_close execution (Jan weights first earn Feb). Spec documents vs inferred separation (caps/floors/lag/sample cov documented, no thresholds invented).
2. Einged trusted data via backtest.datastore read-only on data/market.sqlite (close_only v5, 124 tickers 785032 ohlcv, 20 macro 56261, 73126 as_monthly, 87734 sim_monthly, 104MB + 8MB sim_daily sidecar): 10-ticker risk panel SPY/EFA/EEM/AGG/IEF/TLT/VNQ/DBC/GLD/IWM resampled month-end TR resample(ME).last().pct_change() plus synthetic __DGS10_CM_TR__ (FRED DGS10 1962..2026 synthetic TR) and ^GSPC (1927..2026) and lived IEF for benchmarks; public-input-provenance/v1 hashes sha256:4e83ebfcf... (sleeve panel), sha256:f2b2b20c... (^GSPC), sha256:6e43c232... (60/40), sha256:8032b6ea... (store) / sha256:508ccb54... market file matching coverage_ohlcv slack 5d and status() v5 checks offline.
3. Defined frozen trial ledger research/findings/trial-ledger-qm-y8j.5.json (also trial-ledger-qm-aaaz.json) with common panel 2006-03-31..2026-07-31 245m (DBC limiting 2006-02-06, SPY earliest 1993), feasible OOS 2010-02-28..2026-07-31 198m 17 Jan signals (first 2010-01-31 with 46m expanding), task calibration 1991-1999 (theoretical, 0 eligible signals on ETF panel flagged insufficient_history), untouched 2000-2026 preserved feasible 2006..2026, plus rejection gates predeclared: walk-forward OOS net win (CAGR delta >0 net 10bps), vol-band 0.8-1.2x, regime 3/5, neighbor stability via vol-band perturbation +-10%, extra-lag 2-bar, cost survival 0/10/20bps, sparse 36m eligibility.
4. Implemented research/qm_aaaz_runner.py invoking run_max_cagr_at_vol_walk_forward with multi-start SLSQP _solve_max_cagr_at_vol_once / _multistart_max_cagr_at_vol maximizing mean(log(1+R_p)) subject to realized vol == benchmark vol (sample stdev*sqrt(12) expanding strictly before signal), capped-simplex projection, investability_for_panel, canonical_weights, yielding daily walk-forward cross-check plus monthly sleeve matrix; emitted research/experiments/qm-aaaz.json and research/findings/evidence-manifest-qm-aaaz.json / research/strategy_evidence/meta-target-cagr-vol-qm-aaaz.json G0-G13 fully populated (decision reject).
5. Published via experiments producer (not strategy_promote): python3 -m research.reports scaffold regenerated docs/site-data/experiments/qm-aaaz.json / docs/experiments/qm-aaaz.html + docs/site-data/experiments/index.json / docs/experiments.html, research.artifacts build --mode full --offline for shared manifest, and python3 -m research.reports.results for changelog; no catalog entry created for reject disposition (proven via load_admitted_registry AdmissionError).

Files changed

• research/results/qm-aaaz.md — this durable task note (YAML front matter id qm-aaaz, newest changelog entry).
• research/findings/specs/meta-target-cagr-vol-qm-y8j.5.json — frozen StrategySpec v1.1 (id meta-target-cagr-vol-qm-y8j-5, 10-ticker risk universe, 5%/20%/max10, monthly month_end next_close, 2 variants max_cagr_at_vol, sha256:f630e5..., canonical 888a5308...).
• research/findings/specs/optimizer-3-meta-contract-v1.json — locked Optimizer 3.0 Meta contract v1 (max10/0.05/0.20/sample/0.002, canonical 89e5f375...).
• research/qm_aaaz_runner.py — runner invoking run_max_cagr_at_vol_walk_forward (SLSQP multistart, sample cov ddof1, capped-simplex, Jan signals).
• research/findings/trial-ledger-qm-y8j.5.json and research/findings/trial-ledger-qm-aaaz.json — frozen trial ledger (strategy-trial-ledger/v1, 2 variants, contract/spec/engine/universe/datastore hashes, walk_forward_windows anchored 1970 calibration 1991-1999 vs untouched 2000-2026 feasible 2006-2026).
• research/findings/evidence-manifest-qm-aaaz.json / research/strategy_evidence/meta-target-cagr-vol-qm-aaaz.json — G0-G13 evidence manifest v1.0 (strategy_id meta-target-cagr-vol-qm-aaaz, decision reject, spec_hash 888a, universe_hash 277793..., 8 input_hashes).
• research/experiments/qm-aaaz.json — experiment manifest research-experiment/v1 id qm-aaaz status rejected (metrics for both variants, provenance input_hashes 8 including market 508ccb54..., supplemental no_catalog_entry true).
• docs/site-data/experiments/qm-aaaz.json and docs/experiments/qm-aaaz.html — published rejected experiment (OOS deltas, vol-band, regime, cost, provenance tables).
• docs/site-data/experiments/index.json and docs/experiments.html — regenerated experiments index with qm-aaaz.
• docs/site-data/results.json, docs/results.html, docs/results/qm-aaaz.html — rebuilt changelog payload and detail page via research.reports.results (qm-aaaz newest).
• docs/.shared-artifact-manifest.json — regenerated shared-manifest after scaffold (deterministic, no drift).

Validation

• python3 -m research.pipeline validate — graph 28 nodes 38 edges passed (graph_hash sha256:6802fab...).
• python3 -m research.pipeline verify --offline — passed (shared_artifacts deterministic, strategy_reject passed, indicator_promote passed, data_source passed, 28/28 nodes covered).
• python3 -m research.strategy_spec.validate research/findings/specs/meta-target-cagr-vol-qm-y8j.5.json — spec v1.1 validates (file f630..., canonical 888a...).
• python3 -m research.strategy_evidence.validate research/findings/evidence-manifest-qm-aaaz.json — G0-G13 validate 0 errors, decision reject anchored.
• python3 -m research.experiment_registry validate / load_registry + load_manifest qm-aaaz — registry 12 experiments, qm-aaaz manifest validates v1 id rejected.
• python3 -m research.artifacts check --mode full --offline — shared artifacts deterministic and current, drift 0.
• python3 research/scripts/check_static_reports.py docs — local bundle OK: 1242 files; 178 canary routes.
• python3 -m pytest -q tests/test_qm_aaaz_temporal_causality.py tests/test_qm_aaaz_implementation.py tests/test_engine.py tests/test_datastore.py — 70 passed in 103.56s (temporal causality strictly_before, next_close earnings lag, pipeline verify, registry/spec/evidence, static reports, privacy DB unchanged).
• python3 -m backtest.datastore status / coverage_ohlcv — v5 124 tickers 785032 ohlcv 73k monthly, DBC/SPY/^GSPC/__DGS10_CM_TR__ coverage true with slack 5d/45d offline.
• git grep qm-aaaz research/results/qm-aaaz.md — entry grep-searchable.
• python3 -m research.reports.results --json — rebuilt docs/site-data/results.json and docs/results.html; zero errors, qm-aaaz newest.

Results

Feasible panel 2006-03-31..2026-07-31 245m; feasible OOS 2010-02-28..2026-07-31 198m, 17 January signals. Both variants underperform their vol-matched benchmarks net of 10bps next_close and across 0/20bps: meta-target-cagr-sp500 CAGR 0.08053 vs SPX 0.12420 delta -0.04367 vol 0.1076 target 0.1441 vol-band ratio 0.746 fail (0.8-1.2x), Sharpe 0.775 vs 0.887 delta -0.112, Sortino 1.13 vs 1.30, MaxDD -0.235 vs -0.247, turnover 0.084 annual; meta-target-cagr-6040 CAGR 0.06222 vs 60/40 0.09809 delta -0.03587 vol 0.0915 target 0.0890 ratio 1.028 pass, Sharpe 0.706 vs 1.099 delta -0.392, Sortino 1.09, MaxDD -0.214, turnover 0.14 — overall vol-band fail as one variant infeasible. Regime 5 subperiods: SPX 1/5 wins (only 2025-2026 +0.034), 6040 0/5 all negative (-0.067/-0.023/-0.021/-0.023/-0.019) fail 3/5 majority. Extra-lag Jan->Mar 2-bar preserves failure (SPX -0.04408, 6040 -0.03810). Cost survival fails at 0/10/20bps (gross SPX -0.04356 base -0.04367 stress -0.04378; 6040 -0.03569/-0.03587/-0.03605). Neighbor stability vol-band perturbation +-10%: SPX 0.746 compliance false (0.9x would comply true 1.1x false narrow plateau), 6040 passes but isolated. Calibration 1991-1999 0 eligible January signals insufficient_history (panel starts 2006, first eligible 2010-01-31) not counted as pass. Disposition reject per G13 — no catalog promotion; research preserved via experiments route (docs/site-data/experiments/qm-aaaz.json) without fabricating catalog entry.

Links

• Task: qm-aaaz
• Spec: research/findings/specs/meta-target-cagr-vol-qm-y8j.5.json (sha256:f630e5ccebb2dd18f53ccd8c56cfb6c73d05403e9642c208140948be3e66ee72 canonical 888a53087c...)
• Contract: research/findings/specs/optimizer-3-meta-contract-v1.json (canonical 89e5f375...)
• Ledger: research/findings/trial-ledger-qm-y8j.5.json (ledger qm-y8j.5 / qm-aaaz)
• Evidence: research/findings/evidence-manifest-qm-aaaz.json and research/strategy_evidence/meta-target-cagr-vol-qm-aaaz.json (G0-G13 reject)
• Experiment: research/experiments/qm-aaaz.json / docs/site-data/experiments/qm-aaaz.json + docs/experiments/qm-aaaz.html (expanding 60m/36m capped-simplex walk-forward 2006-2026)
• Datastore: data/market.sqlite (close_only v5 104MB sha256:508ccb5419cc09b43a1e42e60693f6f93d4f9f9efb1fa58b0aeb9c170f149dfd)
• Follow-up: sibling gap record qm-ctpa (research/results/qm-ctpa.md), source experiment qm-y8j-5-max-cagr-at-vol insufficient_evidence