Implement generic trailing-stop adapter with volatility-aware sizing

Objective

Implement ATR and volatility-scaled trailing-stop adapters in backtest/generic_signals.py and wire them into the head-to-head experiment harness (same window, costs, one-bar lag) so the experiment detail delta table shows volatility-aware variants alongside fixed percent stops, with offline tests.

Approach

• Extended backtest/generic_signals.py with _true_range, _wilder_rma, atr_trailing_stop_signal (Wilder RMA of TR, Chandelier-style peak minus 3x ATR) and vol_trailing_stop_signal (rolling std of returns scaled by price, peak times 1 minus 2x vol), plus wrappers atr_trailing_weights, vol_trailing_weights; exposed trailing_atr and trailing_vol via all_generic_signals and all_generic_weights; causal, UTC-naive, close-only fallback documented.
• Updated research/scripts/publish_s5fi_hma_qm_kdnj1.py to compute trailing_atr and trailing_vol harness rows on the same SPY window (ensure_universe offline=True) with BacktestConfig.allocatesmartly (next_close +10bps) and one-bar lag via validate_weights, mirroring trailing_15pct/trailing_20pct wiring.
• Added tests/test_trailing_stop_volatility_harness.py (11 offline tests, no network): determinism, close-only fallback, causality prefix, warmup, ATR/vol divergence vs fixed 10pct, all_generic_signals keys, same-window harness deltas, no DB mutation, one-bar lag, payload/HTML containment.
• Rebuilt deterministic public artifacts via research.reports scaffold and research.artifacts build --mode full --offline; verified docs/experiments/qm-kdnj.1.html 9-row harness table and payload contain trailing_atr/trailing_vol.

Files changed

• backtest/generic_signals.py — new ATR/vol trailing-stop signals and weights, extended all_generic_signals/all_generic_weights
• research/scripts/publish_s5fi_hma_qm_kdnj1.py — harness wiring for trailing_atr/trailing_vol rows
• tests/test_trailing_stop_volatility_harness.py — offline harness tests
• research/experiments/qm-kdnj.1.json and docs/site-data/experiments/qm-kdnj.1.json / docs/experiments/qm-kdnj.1.html — 9-row delta table
• research/results/qm-z7yi.md — this note; docs/site-data/results.json, docs/results.html, docs/results/qm-z7yi.html rebuilt

Validation

• python3 -m pytest -q tests/test_trailing_stop_volatility_harness.py — 11 passed offline (deterministic, causal, divergence, same-window harness, no DB mutation, lag)
• python3 -m pytest -q tests/test_trailing_stop_volatility_harness.py -k 'causality or deterministic or warmup' — 3 passed
• bash -c 'grep -q trailing_atr docs/experiments/qm-kdnj.1.html && grep -q generic-exit-harness docs/experiments/qm-kdnj.1.html' — ok
• Sibling harnesses tests/test_s5fi_hma_generic_harness.py, test_vix_contango_generic_harness.py, test_adl_vol_ratio_generic_harness.py, test_fed_bal_roc_generic_harness.py — 29 passed
• python3 -m research.reports.results and python3 research/scripts/check_static_reports.py docs — static site checks pass
• python3 -m research.artifacts check --mode full --offline — deterministic

Results

The experiment detail page docs/experiments/qm-kdnj.1.html now shows a 9-row generic-exit-harness table including trailing_atr (1.1628 generic-only → 1.8785 signal+generic, delta 61.54%) and trailing_vol (0.9774 → 1.5789, delta 61.54%) alongside fixed stops, on window 2007-01-03 → 2026-07-24 with same costs/lag. Offline harness proves no live fetch and no DB mutation. Changelog entry research/results/qm-z7yi.md is searchable via git grep.