← Results index

Research walk-forward gate for insufficient-evidence experiment qm-y8j-5

2026-08-19T04:00:00Z Experiment completed Recorded insufficient_evidence gap for qm-y8j-5-max-cagr-at-vol in research/findings/qm-u94t-gap-audit.md with qm-ghn0-pattern walk-forward gate (expanding annual Jan, next_close, 10/20bps, vol-band 0.8-1.2x) and public datastore provenance (data/market.sqlite close_only v5 785k ohlcv 73k monthly), referencing follow-up qm-aaaz and closed siblings qm-7ad/qm-5ak/qm-u3m; no promotion until G11-G12 metrics publish.

Objective

For experiment qm-y8j-5-max-cagr-at-vol (insufficient_evidence — not promoted), record the explicit finding gap and file walk-forward follow-up tasks referencing datastore provenance before any parity promotion. The qm-ghn0 pattern requires quantitative walk-forward metrics (expanding annual January, next_close lag, 10/20bps, vol-band 0.8-1.2x, calibration vs untouched, regime/neighbor/cost gates) before promotion.

Approach

  1. Intake classified as fallback (not strategy/indicator/data_source/bug/feature): gap documentation + Beads filing referencing qm-ghn0-pattern gate, not a new hypothesis/data-source lifecycle. Split not required — single atomic finding update; walk-forward execution is a follow-up bead, not a current-run child. Filed follow-up qm-aaaz for the gate.
  2. Updated research/findings/qm-u94t-gap-audit.md: expanded qm-y8j-5 row to show full gap detail (G0-G4 passed, G5-G10 frozen expanding 60m/36m 5%/20%/10 caps, kernel research/optimizer3_meta.run_max_cagr_at_vol_walk_forward + backtest/meta_strategy.ASMetaTargetCAGR* exists via closed qm-7ad but G11-G12 CAGR/vol-band/Sharpe/Sortino/MaxDD/turnover vs SPX (^GSPC) and 60/40 on long_common_1970 + calibration 1991-1999 vs untouched 2000-2026 still pending, parity blocked), public provenance (as_monthly_returns 73k + ohlcv 785k close_only v5 + sim_monthly 87k in data/market.sqlite via backtest/datastore read-only + research/input_provenance public-input-provenance/v1, transactional import-cache explicit_import, coverage_ohlcv/coverage_macro with slack), and follow-up beads qm-ctpa (this record) + qm-aaaz (gate) plus closed siblings qm-7ad/qm-5ak/qm-u3m.
  3. Added Meta continuous source inventory row (as_monthly_returns + ohlcv panel + __GSPC_SHILLER_TR__ proxy, 1970..2026) and follow-ups footer clarifying qm-ctpa/qm-aaaz ownership.
  4. Preserved existing qm-kdnj/qm-tykd/legacy rows and invariants; no DB mutation, no private data, no hand-edited docs/site-data.

Files changed

Validation

Results

Gap for qm-y8j-5 explicitly listed as insufficient_evidence with walk-forward tasks referenced (qm-ctpa gap record + qm-aaaz expanding walk-forward gate per qm-ghn0, plus closed kernel siblings qm-7ad/qm-5ak and ablation qm-u3m) and public datastore provenance coverage (data/market.sqlite close_only v5 via backtest/datastore + research/input_provenance public-input-provenance/v1, import-cache offline). No parity promotion or catalog publication occurs until qm-aaaz publishes G11-G12 CAGR-at-vol metrics with vol-band and benchmark deltas. Finding is grep-searchable via git grep qm-y8j-5 research/findings/qm-u94t-gap-audit.md.

Source artifact: research/results/qm-ctpa.md