1. What the system observed. Six positions — SILVER, INTC, AMD, NVDA, MU, SNDK — all closed their naked leg over the weekend of 2026-08-30. A routine check confirmed all six had actually closed on the hedge broker and that the recorded P&L was correct.
2. What the model expected. Six different instruments gapping over the same weekend should show six different, essentially random naked-leg outcomes — some up, some down.
3. What changed — the finding. Three of the six (INTC, AMD, MU) recorded
naked_leg_pnl = $0.00 exactly. A real market-open print landing on precisely the reference
price, three times out of six, is implausible enough to investigate rather than accept.
4. Root cause. The underlying price feed can lag the market's "open" status flip by a full
scan cycle (about 15 minutes) right at the open. The scanner was writing market_status="open"
in the same snapshot row as a still-stale Friday-closing price, and the exit logic fired the
instant the status flipped — locking in that stale price as the fill. The other three
instruments (SILVER, NVDA, SNDK) happened to get a fresh price on the same tick, so their
figures were already correct — pure luck, not a difference in logic.
5. The fix. Added a check that compares the latest snapshot's price against the prior "closed" snapshot's price; if they're identical, the open-transition is treated as not yet fresh and the close is deferred one more tick — typically a few minutes, not a meaningful change to the naked-leg duration.
6. Backfill. The same signature was found on three older trades, predating this fix by weeks — the bug wasn't new, just newly visible. All six affected trades were corrected using the first genuinely-fresh post-open price as a best-effort reconstruction — not a certified fill, but a large improvement over a known-stale one. Net P&L (funding minus fees plus naked-leg price movement) before and after correction:
| Symbol | Net P&L before | Net P&L after |
|---|---|---|
| INTC | +$9.49 | +$149.76 |
| AMD | +$7.94 | +$79.25 |
| MU | +$10.64 | +$68.47 |
| JP225 (older) | −$10.57 | −$142.72 |
| SNDK (older) | −$8.14 | −$397.53 |
| JP225 (older) | −$14.65 | +$86.79 |
Note the last row: correcting the stale price didn't always make the result worse — one of the older JP225 trades flipped from a recorded loss to a real gain once the actual opening print replaced the frozen one.
7. What was learned. A bug that silently mispriced exits had been running for weeks before three coincidentally-identical zeros made it visible — the kind of error that a single trade in isolation would never have surfaced. This is also the reason the naked-leg gap shown on this site's Lab Portfolio always uses the first genuinely fresh post-open price, not just the first price available after the open. See When funding profit differs from total trade profit for how this piece fits into the Net P&L vs. True P&L distinction.