{"methodology":"Backtested on Kraken's full available 1h OHLCV history per symbol -- v1 (original composite scorer) vs v3 (revised scorer) run head-to-head on IDENTICAL fetched data (same symbol, same window, same walk-forward harness: quant_trader.backtest.runner.run_backtest). Symbol is routed to whichever version backtests better for THAT symbol specifically -- we do not blanket-ship one scorer version across all symbols. See per-symbol notes for cases where v3 would have been a regression. 'Hit rate' / 'win_rate' = % of CLOSED trades (a position fully exited, not an open mark-to-market) that were net profitable after a 5bps fee/slippage deduction on both entry and exit -- the same definition used by the walk-forward harness's own BacktestResult.win_rate_pct, not a separately-invented metric. 'Profit factor' (PF) = gross wins / gross losses in dollar terms; PF can differ sharply from hit rate because a scorer can win more often but lose bigger on its losers (or vice versa) -- we publish both, every time, specifically so neither number alone can be used to oversell the other. SAMPLE SIZE CAVEAT (load-bearing, not a footnote): Kraken's public 1h OHLCV endpoint retains ~720 candles (~30 days) of history regardless of what is requested -- confirmed 2026-09-28, re-confirmed 2026-10-07. Every number on this page reflects that rolling ~30-day window, not a longer independently-chosen sample. Because the window rolls forward every day, re-running the identical backtest 10 days later produces a MEASURABLY DIFFERENT result, sometimes a large one -- see BTC/USD's validation_history below, where profit factor moved from 0.86 to 0.24 in 10 days on the same methodology, same scorer, same symbol. Treat every PF/hit-rate number here as a directional read for a roughly month-long window, re-validated periodically, NOT a long-horizon guarantee -- until a coarser-timeframe (4h/1d, which retain more real history) or externally-sourced longer backtest is published alongside it.","last_updated":"2026-10-07: BTC/USD and DOGE/USD re-validated against today's Kraken window (kanban t_0c8e6e9f). ETH/TAO/QNT/ZEC/DOG/VVV numbers below are still the 2026-09-27 validation run (not in scope for this pass; flagged for the next re-validation cycle).","per_symbol":{"BTC":{"live_version":"v3","v1_profit_factor":0.594,"v3_profit_factor":0.242,"v1_win_rate":0.431,"v3_win_rate":0.493,"sample_window":"2026-09-10 08:00 UTC -> 2026-10-07 20:00 UTC, 1h bars, 661 scored bars, Kraken 30-day retention cap","note":"v3 still shows a meaningfully higher hit rate than v1 (49.3% vs 43.1%), but profit factor REGRESSED hard vs the 2026-09-27 validation (0.86 -> 0.24) on this freshly-rolled window -- v3 is winning more often but losing bigger on this sample. This is exactly the kind of swing the sample-size caveat above warns about, not noise we're hiding: v1's PF also dropped in the same window (0.73 -> 0.59), so part of this is a rougher market period for BOTH scorers, not v3-specific decay. Routing is LEFT ON v3 for now (hit rate, the thing the Sept 28 pass originally flagged BTC for, still holds up) but this is now flagged for a tighter re-validation cadence rather than the prior one-and-done check. Was also briefly, unintentionally rolled back to v1 during the Oct 5-7 rebuild (stale base file, not a deliberate decision or Jev incompatibility) -- restored 2026-10-07, see kanban t_89c6d833.","validation_history":[{"as_of":"2026-09-27","v1_profit_factor":0.73,"v3_profit_factor":0.86,"v1_win_rate":0.485,"v3_win_rate":0.541},{"as_of":"2026-10-07","v1_profit_factor":0.594,"v3_profit_factor":0.242,"v1_win_rate":0.431,"v3_win_rate":0.493}]},"DOGE":{"live_version":"v3","v1_profit_factor":1.524,"v3_profit_factor":2.745,"v1_win_rate":0.419,"v3_win_rate":0.607,"sample_window":"2026-09-10 08:00 UTC -> 2026-10-07 20:00 UTC, 1h bars, 661 scored bars, Kraken 30-day retention cap","note":"Real improvement, confirmed again on a fresh window: v3 beats v1 on BOTH profit factor (2.75 vs 1.52) and hit rate (60.7% vs 41.9%) today, an even larger gap than the 2026-09-27 run showed on profit factor (1.63 vs 1.10) -- notably, the 2026-09-27 run had actually shown v3's hit rate slightly BELOW v1's (40.7% vs 45.8%) despite the better profit factor; that inversion is gone in today's window. DOGE remains the strongest, most consistent v3 case of the two symbols re-validated this pass. Was also briefly, unintentionally rolled back to v1 during the Oct 5-7 rebuild (stale base file, not a deliberate decision or Jev incompatibility) -- restored 2026-10-07, see kanban t_89c6d833.","validation_history":[{"as_of":"2026-09-27","v1_profit_factor":1.1,"v3_profit_factor":1.63,"v1_win_rate":0.458,"v3_win_rate":0.407},{"as_of":"2026-10-07","v1_profit_factor":1.524,"v3_profit_factor":2.745,"v1_win_rate":0.419,"v3_win_rate":0.607}]},"ETH":{"live_version":"v1","v1_profit_factor":0.23,"v3_profit_factor":0.4,"v1_win_rate":0.453,"v3_win_rate":0.49,"note":"v3 improves win-rate/PF but total return still worsens and PF stays sub-1.0 either way; kept on v1 pending further work. (2026-09-27 validation, not re-run this pass.)"},"TAO":{"live_version":"v1","v1_profit_factor":2.37,"v3_profit_factor":1.88,"v1_win_rate":0.577,"v3_win_rate":0.522,"note":"v1 clearly stronger, v3 not shipped for this symbol. (2026-09-27 validation, not re-run this pass.)"},"QNT":{"live_version":"v1","v1_profit_factor":2.76,"v3_profit_factor":0.56,"v1_win_rate":0.672,"v3_win_rate":0.619,"note":"v3 would have been a severe regression on one of our best-performing pairs (PF 2.76 -> 0.56). Deliberately not shipped. (2026-09-27 validation, not re-run this pass.)"},"ZEC":{"live_version":"v1","v1_profit_factor":0.9,"v3_profit_factor":0.55,"v1_win_rate":0.552,"v3_win_rate":0.503,"note":"v1 stronger, v3 not shipped for this symbol. (2026-09-27 validation, not re-run this pass.)"},"DOG":{"live_version":"v1","v1_profit_factor":1.84,"v3_profit_factor":2.02,"v1_win_rate":0.529,"v3_win_rate":0.504,"note":"v3 marginally better but v1 already solid; left on v1 rather than chase a small delta. (2026-09-27 validation, not re-run this pass.)"},"VVV":{"live_version":"v1","v1_profit_factor":1.78,"v3_profit_factor":1.55,"v1_win_rate":0.587,"v3_win_rate":0.558,"note":"v1 clearly stronger, v3 not shipped for this symbol. (2026-09-27 validation, not re-run this pass.)"}},"honesty_note":"Profit factor (PF) below 1.0 means the backtest lost money overall even with a positive win rate -- we report this plainly for ETH and any other sub-1.0 symbol rather than only publishing flattering numbers. A signal service that only shows you its winners isn't giving you a track record, it's giving you an ad. The BTC entry above is the sharpest example yet: hit rate held up on a fresh window but profit factor did not, and we're publishing that contradiction instead of only re-running until the number we like shows up."}