Open Ledger

⚠ SCALE_TF_001 — Before the fix, swing (H1) and position (H4) records used the intraday bar length in two places: engine event timestamps and, until 29 Jul, the trade close timestamp and holding time reported to the cost model. Scope: H1, H4 · crypto, tradfi · 106 engine events. Impact: 106 engine events (75 H1 crypto, 10 H1 metals, 21 H4 crypto) carry timestamps 45 or 225 minutes early and cannot be edited (the event table is hash-chained and refuses updates). Trade rows 8 and 12 were affected at write time; both were corrected on 29 Jul 2026 with a dated changelog entry, and every row written since uses the correct bar length. Evaluation impact: None. The pre-registered evaluation pool is intraday crypto (scale M15), whose bar length was correct throughout; all M15 events and every evaluation-pool trade are unaffected. Resolution: Event rows are kept exactly as written — this note is attached instead of editing history. The two affected trade rows were corrected under the dated 29 Jul changelog entry, before the keep-and-annotate policy was adopted; that correction is disclosed here rather than hidden. (fixed from tracker fix deployed 2026-07-29 · engine timestamp fix deployed 2026-07-30 22:30 UTC (commit 3c468b5); discovered 2026-07-29). Affected trade rows: 8, 12. Dated changelog entry →

Worked example — from the row above, not invented LAUSDT LONG · entry 0.05872 → exit 0.05785 gives gross -148 bps (price move relative to entry, in the trade's direction). The stored net is -166 bps, so the modeled round-trip cost was 18 bps — taker fees on both legs + slippage + funding (mx). Check any other row the same way: gross − net must always be a plausible cost, and net can never beat gross. If you ever find otherwise, you've caught us — tell us.