Same strategy name. Different experiment.
Reconstructing decommissioned OU, Gold and portfolio systems—and establishing exactly what their saved results can tell us.
Identity comes before performance.
“The OU bot” is not a sufficient specification. An earlier parameter set, a subsequently deployed configuration and an extended research engine can share a family name while producing different positions.
The archive contains January–August 2026 simulations using pre-fetched OANDA practice-feed, mid-price candles. The review distinguished source-reported summaries from saved return ledgers, checked the measurement windows and reconciled accounting where raw series were available.
No retired strategy was redeployed for this review. No broker connection or new strategy backtest was needed to check the arithmetic in the saved ledgers.
Reconciling a curve proves the accounting agrees. It does not prove the simulation describes executable trades.
Keep the label attached
to the number.
| Configuration / window | Annualised return / cumulative | Replay drawdown | Evidence / limitations |
|---|---|---|---|
| OU Top-5 · deployed configuration 2026-01-20 — 2026-08-21 · 214 calendar days | +20.90% Cumulative: +11.77% | -24.90% | Recomputed from saved ledger Gross mid-price replay. Actual execution, spread, slippage and financing are not represented. Source: ou_equity.csv · Costs excluded |
| Gold H4 · v2 configuration 2026-01-01 — 2026-08-21 · 233 calendar days | -13.53% Cumulative: -8.87% | -21.48% | Recomputed from saved ledger Gross mid-price replay. Actual execution, spread, slippage and financing are not represented. Source: gold_equity.csv · Costs excluded |
| Strategy 17 · ensemble 2026-02-20 — 2026-08-21 · 183 calendar days | +1.70% Cumulative: +0.85% | -2.79% | Recomputed from saved ledger Gross mid-price replay. Actual execution, spread, slippage and financing are not represented. Source: s17_equity.csv · 1bp turnover approximation; other costs excluded |
| OU Top-5 · pre-recalibration 2026-01-01 — 2026-08-21 · 233 calendar days | +39.15% Cumulative: +23.48% | -25.74% | Source-reported JSON · unvalidated Separate configuration / paper proxy. No standalone raw returns ledger available for this archived summary. Source: 05_optimised_top5_prerecal.json · Costs excluded |
| Gold H4 · v1 configuration 2026-01-01 — 2026-08-21 · 233 calendar days | +29.86% Cumulative: +18.15% | -16.74% | Source-reported JSON · unvalidated Only six recorded trades. Missing volume changes the leverage behaviour; fills approximate the signal-bar close. Source: 02_gold_v1.json · Costs excluded |
| CWMR · paper replication 2026-01-01 — 2026-08-21 · 233 calendar days | +10.87% Cumulative: +6.81% | -3.14% | Source-reported JSON · unvalidated Separate configuration / paper proxy. No standalone raw returns ledger available for this archived summary. Source: 09_cwmr_2011.json · Costs excluded |
| Lane 46 · time-series mean reversion 2026-01-01 — 2026-08-21 · 233 calendar days | +235.92% Cumulative: +116.73% | -38.02% | Source-reported JSON · selection-biased / partial Selection bias: the configuration and instrument subset were selected within the reported window; the headline includes the selection/training period, not a fully out-of-sample track record. Source coverage is partial (M30 inputs missing). A separate de-selected configuration reports OOS Sharpe 0.28; that is not this headline configuration. The source uses position-sign flips as activity counts, not independent completed trades. Source: 46_zoo_ts_meanrev.json · Source-model turnover costs deducted; not verified execution costs |
Three questions.
Three different tests.
-
Does the arithmetic reconcile?
Compound the saved OU return series and compare every observation with saved equity. In this review they matched with zero numerical difference. Gold and S17 summaries were recomputed from their archived NAV paths.
-
Does the replay match the intended system?
Check parameter provenance, missing inputs, fill conventions and the exact replay window. Gold v1’s missing-volume behaviour and Gold v2’s fill approximation remain material limitations.
-
Is there valid evidence of a repeatable edge?
That requires a separate point-in-time, selection, cost and out-of-sample validation. This archive review does not establish it, and it does not override previous validation retractions.
The private source bundle is the January–August replay archive: the master lane sweep, configuration-specific JSON summaries and the saved OU, Gold and Strategy 17 ledgers. A local review receipt records input hashes and the accounting calculation.
Underlying trading parameters and datasets are not distributed through this website. The anomalous extended Gold v2 output is excluded from the comparison because of unresolved internal inconsistencies. The earlier OU “gate-pass” narrative remains withdrawn.
Discuss the technical work ↗