Outcome
The historical curve is replicated; the strategy is not.
The frozen replay reached 0.9999816653 normalized equity-level correlation, 100% aligned-event coverage, 0.027824 normalized RMSE, and 100% event-change sign agreement. The generated overlay preserves the full rise, plateau, and drawdown sequence. Those are strong engine-fidelity results.
The disposition is still fidelity_only. The TradingView publication exposes its description, compiled settings, trade list, and report metrics, but the Pine text is protected. There is no Pine source hash, no exact live breadth/A-D/volatility signal, and no outcome-unseen economic window. Turin LT therefore remains outside the public strategy catalog.
What was documented
| Area | Public evidence | Unavailable — not invented |
|---|---|---|
| Entry | Nasdaq breadth washout plus combined advance/decline thrust, gated by realized volatility | Breadth horizon, EMA length, entry threshold, A/D formula/crossover, vol window/scaling/floor |
| Exit | Breadth repair | Exact exit threshold |
| Position | Long only, 80% equity, pyramiding one | SPY chart/report versus SPXL publication-metadata conflict |
| Costs | Commission 0%, slippage 0 ticks | No hidden exception inferred |
| Timing | Daily chart; process_orders_on_close=false, calc_on_every_tick=false | Pine text unavailable; same_close rejected |
| Display only | BAMLH0A0HYM2 z-score panel | It is not used as an order input |
The public archive records script.content="", access=3, and has_access=false. TradingView exposes an internal compiled digest, but its algorithm is undisclosed and it is not presented as a Pine source hash.
Implementation
backtest/turin_lt_claudified.py loads the frozen offline event fixture, validates a strict initial/entry/exit/terminal-mark state machine, normalizes UTC instants to a sorted unique UTC-naive engine index, creates event-price MarketData, emits nonnegative 0.8/0.0 SPY targets through validate_weights, and calls backtest.engine.run_backtest with next_open and explicit costs.
This mapping makes the entry target earn only the following entry-fill-to-exit-fill interval and makes the exit target flat for the next inter-trade gap. same_close, missing dates, duplicate timestamps, reversed trades, and overlapping entries fail closed. A dedicated boundary test maps Friday 2024-08-30 to Tuesday 2024-09-03 across month-end and Labor Day.
Robustness
| Frozen check | Correlation | Terminal normalized ratio | Result |
|---|---|---|---|
| 80% target, source 0 bps | 0.9999817 | 1.01303 | Fidelity pass |
| 80% target, 10 bps | 0.9999853 | 0.97245 | Shape survives; endpoint falls |
| 80% target, 20 bps | 0.9999441 | 0.93347 | Shape survives; endpoint falls |
| 100% weight ablation | 0.9987828 | 1.57287 | Overshoots materially |
| Include final open mark | 0.9999804 | 1.01315 | Stable ablation |
| One extra event lag | -0.7679792 | 0.08121 | Fails decisively |
| Daily adjusted-close SPY path | 0.9994738 | — | Path ablation only; not raw open fills |
The label-only 2005–2018 and 2019–2026 splits correlate at 0.9999994 and 0.9999348. GFC and 2022 subperiods remain above 0.99999; COVID has only two event anchors and is reported as insufficient history. No protected parameter neighborhood or outcome-selected proxy was evaluated.
Data range and provenance
The exact public event replay spans 2005-05-03 through the final closed exit and open entry on 2026-05-28; the final 2026-08-20 publication-time mark is an ablation. The offline SPY cache supports the daily path through 2026-07-24.
The mandated extension search checked relaxed proxies, synthetic extension, point-in-time constituent rebuild, third-party assembled history, and exact event replay. Candidate breadth/A-D histories begin between 1996 and 2018, but none identifies the protected target. Current-membership constituent reconstruction was rejected for survivorship leakage. Exact public event replay is the longest defensible historical-fidelity path and cannot operate live.
Raw TradingView HTML, its screenshot, fill prices, and individual observations are retained only as research fixtures. Public pages contain hashes, rule summaries, annual normalized summaries, and newly generated visuals.
Files
research/artifacts/fixtures/turin_lt_claudified_pine.json— private frozen page/source-availability archiveresearch/artifacts/fixtures/turin_lt_equity.csv— private exact event/equity fixtureresearch/findings/specs/turin-lt-claudified.json— validated strategy specificationbacktest/turin_lt_claudified.py— offline fidelity replayresearch/findings/turin-lt-robustness-qm-wpx5.json— frozen robustness matrixresearch/artifacts/generated/turin_lt_equity_overlay.png— newly generated overlayresearch/strategy_evidence/turin-lt-claudified-qm-wpx5.json— validated G0–G13 manifest withfidelity_onlydecisionresearch/experiments/qm-wpx5.json— public experiment manifestresearch/experiments/qm-wpx5.ipynb— executed offline notebook with a 0–20 bps interactive viewerdocs/experiments/qm-wpx5-analysis.html— self-contained public analysis page
Validation
python3 -m research.strategy_spec validate research/findings/specs/turin-lt-claudified.jsonpython3 -m research.scripts.evaluate_turin_lt_robustnesspython3 -m pytest -q tests/test_turin_lt_claudified.pypython3 -m research.reports notebooks --id qm-wpx5python3 -m research.reports scaffoldpython3 -m research.reports resultspython3 -m research.artifacts check --mode full --offline
Conclusion
The target curve can be reproduced without inventing hidden parameters, and the visual agreement is excellent. That success says the event-to-engine mapping is faithful; it does not recover Turin LT's signal. Keep the replay as reproducible research, keep the failures visible, and do not trade or catalog it as a live strategy.