Turin LT: a near-perfect replay that still cannot trade live
Completed qm-wpx5 · 2026-08-22
Can the protected Turin LT TradingView publication be reproduced honestly through QuantModel's backtest engine without inventing its hidden breadth, A/D, or volatility settings?
Yes for historical shape, no for a live strategy. The frozen 80% event replay reaches 0.999982 correlation with the published report and matches every major rise, pause, and drawdown. But the Pine source, five signal defaults, exact input formulas, and a genuinely untouched economic window are missing, so the result is fidelity-only.
Provenance
finding · artifact · code · payload · notebook analysis
Research narrative
The short version
The historical curve is real and the match is excellent. The signal is not recoverable, so this is an engine-fidelity artifact, not a tradable strategy.
What was frozen
Exact public event dates, 80% SPY / residual cash, next-open engine lag, zero source costs, fixture hashes, a correlation gate above 0.90, and every allowed stress were fixed before the confirmatory run.
What failed
An extra event lag reverses the shape. COVID has too few event anchors for a subperiod claim. Hidden parameter neighborhoods and proxy fitting were deliberately left unevaluated.
Why no catalog entry
A catalog strategy must generate future allocations from point-in-time inputs. This replay consumes the historical orders themselves and cannot do that.
Replay vs published report — annual event summaries
Aligned window, both rebased to 1.0 at start — Both curves are normalized to 1.0; the public payload contains annual derived summaries, not raw fills or trade observations. · S&P dashed, experiment solid. X = Date, Y = Equity (rebased to 1.0).
Signal driver & thresholds (x-aligned)
What the publisher revealed
| piece | documented | hidden |
|---|---|---|
| Entry | Breadth washout plus A/D thrust; realized-volatility gate | Breadth horizon, EMA, threshold, A/D formula, vol window/floor |
| Exit | Breadth repair | Exact exit threshold |
| Position | Long only, 80% equity, pyramiding 1 | SPY chart vs SPXL publication metadata conflict |
| Execution | Zero commission/slippage; process-on-close false | No hidden same-close exception observed |
Hidden means unavailable, not estimated.
Frozen checks, including the failure
| check | correlation | terminal_ratio | result |
|---|---|---|---|
| 80% · 0 bps | 1.000 | 1.013 | pass |
| 80% · 10 bps | 1.000 | 0.972 | shape only |
| 80% · 20 bps | 1.000 | 0.933 | shape only |
| 100% weight | 0.999 | 1.573 | overshoots |
| +1 event lag | -0.768 | 0.081 | fails |
No failed cell was tuned away after the protocol freeze.
Stats
| CAGR | 10.17% | Sharpe | 26.67% |
|---|---|---|---|
| Sortino | 0.00% | maxDD | -12.02% |
Disposition
Fidelity only — keep the offline replay and evidence, publish the limitations, and do not add Turin LT to the strategy catalog or present it as a forward signal.
Follow-on work
- No explicit follow-on; file new issue if needed
Errors & data gaps
The public Pine source is protected, so there is no Pine source SHA-256. The published-page archive hash is not a substitute. The exact breadth series, A/D combination, realized-volatility formula, and five locked defaults remain unknown; no live signal was invented. All public orders and outcomes were already visible before evaluation. The 2019–2026 split checks shape only and is not an untouched economic holdout. The primary replay uses fractional 80% weights while TradingView used integer shares, explaining the small terminal-level difference. Raw TradingView HTML, screenshot, fill prices, and trade observations stay out of the public page; it publishes summaries and newly generated visuals only.
Follow-up
gh issue create --title "Follow-up Turin LT a near-perfect replay that still cannot trade live" --body "Follow-up to qm-wpx5: Can the protected Turin LT TradingView publication be reproduced honestly throug" --label "priority:p2" --label "pipeline:research" --label "agent:ready"