1. What the system observed. On 2026-06-25, ARM's funding rate on Coinbase International (INTX) cleared the entry threshold with sufficient persistence and acceptable basis. A position opened: short the ARM perp, long ARM stock via Interactive Brokers as the hedge.
2. Why the trade was entered. Every entry gate passed — funding above threshold, basis inside limits, liquidity sufficient. Nothing about the entry itself was unusual.
3. What the model expected. A funding-rate arbitrage trade with price risk cancelled between the two legs, closed out within weeks once funding decayed or flipped, per the strategy's normal exit rules.
4. What changed. On 2026-07-01, the exit trigger fired (fr_decay, funding had faded below
the research floor). Under the strategy's rules, the perp leg closes immediately to stop the
funding bleed, and the hedge leg is deferred to the next market open if IB is closed — normally
a same-day or overnight gap. This time, the scanner and daemon lost connectivity (a DNS/scanner
outage) for 15 days, from 2026-07-01 to 2026-07-16. The hedge leg sat naked — unhedged,
fully exposed to ARM's price — for the entire outage instead of a normal few-hour window.
5. Funding, price movements, and costs. Funding earned before the exit trigger: +$37.80. ARM's stock price moved sharply against the naked position over those 15 days while the daemon was blind to it.
6. Final result. Naked-leg P&L: −$2,848.10. Net P&L: −$2,821.80 — by a wide margin the largest loss in this project's history, and larger than every other naked-leg outcome combined.
7. What was learned. The funding-rate thesis was not wrong — the loss came entirely from an operational failure (a monitoring/connectivity outage) turning a normally bounded risk (hours) into an unbounded one (15 days). This incident is the direct reason two alerts now exist: a 45-minute scanner-staleness alert, and a 96-hour stuck-naked-leg alert — both designed so that if this exact failure mode recurs, a human finds out in under two hours instead of two weeks. See What happens when the perp closes before the hedge for the general mechanism, and Methodology for the current alert thresholds.
This trade is included in the Lab Portfolio record like every other — its Net P&L is real and counted in the running totals. Its True P&L (all-in) and per-leg price breakdown show as unavailable rather than zero, because the source system didn't yet record entry/exit prices on both legs this far back in the project's history — see Methodology for what that distinction means. It is not excluded as an outlier — it's documented here, with its real numbers, in full.