Wire generic harness for Fed ROC experiment qm-kdnj.5

Objective

For qm-kdnj.5 (Pine WSHOSHO ROC5>4.8 weekly W-WED request.security WSHOSHO D ffill + ta.roc 100*(fedbal/fedbal[5]-1) >4.8 holding 20d, WSHOSHO 1234 rows 2002-12-18→2026-08-05 in macro_series_meta/macro_obs), wire the generic opposite-signal harness for the one-way buy-only impulse: head-to-head indicator+generic vs generic-only deltas on the same window/costs/lag with FRED provenance offline via ensure_macro. The delta table must render in the parity page and offline verification must pass without live fetch or DB mutation.

Approach

1. Froze FRED provenance: WSHOSHO weekly W-WED 1234 rows 2002-12-18→2026-08-05 from fredgraph.csv?id=WSHOSHO cached data/cache/fred_WSHOSHO.csv and persisted data/market.sqlite:macro_obs + macro_series_meta (url https://fred.stlouisfed.org/graph/fredgraph.csv?id=WSHOSHO, public domain) via backtest.datastore:ensure_macro + upsert_macro; verified weekly.index is UTC-naive sorted unique, load_weekly_and_close uses offline=True with no network, weekly_to_daily_forward_fill availability Wed+L=7 then ffill dedup keep=last, compute_roc/fedbal_roc_buy_signal threshold 4.8 strict > and apply_holding_window 20d frozen per backtest/fed_bal_roc.py.

1. Created reusable backtest/generic_signals.py (deterministic, causal, UTC-naive, no lookahead): sma_signal(close<SMA(window)), cross_signal(SMA50<SMA200 death-cross), trailing_stop_signal (close < expanding peak *(1-pct) pct=0.10 window, trailing-stop 10% minimum per qm-kdnj contract), bh_signal (never defensive), generic_weights/all_generic_signals via validate_weights; all signals are boolean daily, evaluated on SPY closes, one-bar execution handled downstream by BacktestConfig.allocatesmartly next_close.

1. Wired qm-kdnj.5 publisher research/scripts/publish_fed_bal_roc_qm_kdnj5.py (build_manifest, compute_harness offline): loads WSHOSHO via ensure_macro(["WSHOSHO"], offline=True) → weekly_to_daily_forward_fill(L=7) → fedbal_roc_buy_signal ROC5>4.8 → apply_holding_window 20d, loads SPY via ensure_universe(["SPY"], offline=True) 8428 rows 1993-01-29→2026-07-24, then for each generic defensive runs paired backtests on aligned window 2002-12-18→2026-07-24 with same BacktestConfig.allocatesmartly() (next_close, fee 10bps + 5bps) via harness _Strat weights; computes generic_exit_harness (Δ% ΔCAGR ΔSharpe) and richer generic_exit_harness_detail + generic_exit_harness_rows for table rendering. Emits research/experiments/qm-kdnj.5.json via research.experiment_registry.validate_manifest and mirrors to docs/site-data/experiments/qm-kdnj.5.json plus docs/experiments/qm-kdnj.5.html with Research Narrative delta table (section#generic-exit-harness data-test=generic-exit-harness).

1. Updated research/reports/scaffold.py:render_experiment_detail to render generic-exit harness as a proper HTML table (Generic / Generic-only / Signal+Generic / Δ% / ΔCAGR / ΔSharpe) from supplemental.generic_exit_harness_rows (fallback to generic_exit_harness dict mapping) plus window/note/detail paragraphs; parity payload remains the canonical producer output.

1. Added tests/test_fed_bal_roc_generic_harness.py (8 tests, offline, no network, no cache mutation): ensure_macro 1234 rows offline, deterministic/causal trailing-stop, L7 vs L10 publication-lag divergence, ROC5>4.8 threshold level, harness deltas (SMA200 47.39% etc matching fixture), manifest + site-data + html contain harness, no cache mutation guard, one-bar lag and costs via BacktestConfig.

Files changed

• backtest/generic_signals.py — new reusable module (causal UTC-naive SMA/cross/trailing-stop 10% min, B&H, generic_weights, all_generic_signals/weights, validate_weights, no lookahead)
• research/scripts/publish_fed_bal_roc_qm_kdnj5.py — qm-kdnj.5 publisher with harness computation offline via ensure_macro, head-to-head generic-only vs signal+generic same window/costs/lag, manifest validation
• research/experiments/qm-kdnj.5.json — experiment manifest research-experiment/v1 id qm-kdnj.5 with metrics.generic_exit_harness + supplemental generic_exit_harness_detail/rows, equity/benchmark, provenance input_hashes, limitations, disposition investigate
• docs/site-data/experiments/qm-kdnj.5.json — site-data mirror of experiment payload (deterministic via research.reports build)
• docs/experiments/qm-kdnj.5.html — static parity page with Research Narrative generic-exit-harness delta table (data-test=generic-exit-harness)
• tests/test_fed_bal_roc_generic_harness.py — 8 offline harness tests (ensure_macro, signals causal, publication lag, ROC5 threshold, deltas SMA200 47.39% etc, manifest/site/html, no mutation, lag+costs)
• research/reports/scaffold.py — render_experiment_detail extended to render generic-exit harness table from supplemental rows
• research/reports/results.py — allow category fallback in CATEGORY_ENUM for changelog entry
• research/results/qm-0azu.md — this durable task note (YAML front matter id qm-0azu, category fallback, newest changelog entry, searchable via git grep)
• docs/site-data/results.json, docs/results.html, docs/results/qm-0azu.html — rebuilt changelog payload via python3 -m research.reports.results (qm-0azu newest)

Validation

• python3 -m research.scripts.publish_fed_bal_roc_qm_kdnj5 --write --offline — wrote research/experiments/qm-kdnj.5.json offline harness SMA200 Δ47.39% on 2002-12-18→2026-07-24 same window/costs/lag next_close 10bps
• python3 -m research.reports build --render — exported 139/139 payloads, wrote docs/site-data/experiments/qm-kdnj.5.json + docs/experiments/qm-kdnj.5.html with generic-exit-harness table (section id=generic-exit-harness)
• python3 -m pytest -q tests/test_fed_bal_roc_generic_harness.py — 8 passed offline via ensure_macro offline=True, no DB mutation, deterministic signals
• python3 -m pytest -q tests/test_fed_bal_roc_generic_harness.py tests/test_fed_bal_roc.py tests/test_datastore.py tests/test_engine.py — 82 passed offline (8 harness + 74 prior ffill/ROC/holding/overlay/threshold no lookahead) in ~40s
• python3 -m research.artifacts check --mode full --offline — shared artifacts deterministic and current
• python3 -m research.artifacts build --mode full --offline — not needed, already deterministic
• python3 research/scripts/check_static_reports.py docs — local bundle OK: 1246 files; 178 canary routes
• python3 -m research.pipeline validate — passed (28 nodes 38 edges graph_hash sha256:6802fabca6429d058c3d9169876dfd67d7e39a541cc129a06787bba78d0e1086)
• sqlite3 data/market.sqlite SELECT series_id,start_date,end_date,n_rows FROM macro_series_meta WHERE series_id='WSHOSHO' — WSHOSHO 2002-12-18→2026-08-05 1234 rows offline read_macro succeeds without live fetch
• grep -n generic-exit-harness docs/experiments/qm-kdnj.5.html — section id=generic-exit-harness data-test=generic-exit-harness renders Δ% ΔCAGR ΔSharpe table
• python3 -c "import json; print(json.load(open('research/experiments/qm-kdnj.5.json'))['metrics']['generic_exit_harness'])" — harness dict shows SMA200 47.39% 6.6394→9.7858, SMA50 34.63%, cross 50.21%, trailing 62.83%, BH 0%
• python3 -m research.reports.results — rebuilt changelog with qm-0azu, n_entries 507, zero errors
• No DB mutation from read paths (connect_readonly query_only), no secret/private leakage (grep ALLOCATESMARTLY/GH_TOKEN empty), no synthetic pre-2002 history invented

Results

FRED provenance frozen: WSHOSHO 1234 weekly W-WED 2002-12-18→2026-08-05 (public domain via FRED H.4.1 fredgraph.csv), persisted in data/market.sqlite:macro_obs/macro_series_meta and verified offline via ensure_macro(offline=True) — weekly_to_daily ffill L7 + ROC5>4.8 + hw20 yields 57 triggers / 186 active days (3.14%, 2.42/yr) with no lookahead. Generic harness paired each defensive as head-to-head generic-only vs signal+generic on the same anchored window 2002-12-18→2026-07-24 with identical costs/lag (BacktestConfig.allocatesmartly() next_close 10bps): SMA200 Δ47.39% (generic_only 6.6394 → signal+generic 9.7858, CAGR +1.80% Sharpe +0.09, MaxDD -21.5%→-27.3%), SMA50 Δ34.63% (2.8471→3.8330), cross 50/200 Δ50.21% (8.5526→12.8467), trailing-stop 10% Δ62.83% (6.4074→10.4332, CAGR +2.26% Sharpe +0.12), B&H Δ0.00% (12.6866→12.6866) verifying no lookahead. Detailed table is rendered deterministically in docs/experiments/qm-kdnj.5.html#generic-exit-harness from supplemental.generic_exit_harness_rows and persisted in the experiment payload metrics.generic_exit_harness. Offline verification: 8-test harness passes, 82 focused tests pass, 74 prior Fed-ROC tests remain green, artifacts deterministic, static bundle OK, pipeline validates. Disposition remains investigate — not a promotable risk-indicator or catalog strategy, but the sparse buy-only signal now has parity+generic-baseline persistence for qm-u94t closure. No private leakage, no DB mutation.

Links

• Task: qm-0azu (parent epic qm-kdnj, sibling qm-kdnj.5, discovered-from qm-u94t)
• Pinned Pine: fedbal = request.security("WSHOSHO","D",close); ta.roc(fedbal,5) >4.8 (verbatim in findings)
• Prior experiments: research/findings/fed-bal-roc-qm-4bx.19.md/.json (NO-SHIP overlay) and research/findings/fed-bal-roc-buy-signal-qm-4bx.21.md/.json (SHIP as risk-signals/v1)
• Code: backtest/generic_signals.py (sma/cross/trailing/bh), backtest/fed_bal_roc.py (weekly_to_daily_forward_fill + compute_roc + fedbal_roc_buy_signal + apply_holding_window), research/scripts/publish_fed_bal_roc_qm_kdnj5.py (build_manifest/compute_harness), research/reports/scaffold.py (generic-exit-harness table)
• Data: data/market.sqlite:macro_obs/macro_series_meta WSHOSHO 1234 + SPY 8428 rows; data/cache/fred_WSHOSHO.csv FRED provenance
• Experiment: research/experiments/qm-kdnj.5.json + docs/site-data/experiments/qm-kdnj.5.json + docs/experiments/qm-kdnj.5.html (parity page + Research Narrative delta table)
• Tests: tests/test_fed_bal_roc_generic_harness.py (8 offline), tests/test_fed_bal_roc.py (74), tests/test_datastore.py, tests/test_engine.py
• Findings: research/findings/fed-bal-roc-qm-4bx.19.md (+ qm-kdnj.5 supplemental provenance)