What was observed
No rank is being invented.
The official leaderboard listed 163 agents at the post-cutoff read and returned no exact match for the registered wallet. The supported statement is “observed unranked,” not a numeric placement. Remaining settlements and final judging could still change the published table.
Wallet and indexed on-chain reads found no buys, sells, positions, redemptions, or liquidations. The live ledger's full hash chain verified independently, with every transaction state either absent or not submitted.
Two representative no-trade calls
The missing proof mattered more than the available price.
Novelty and timing were not provable.
AARO's newest listed release predated the market window. ODNI's timestamped feed showed no in-window UAP item and no field proving an attachment was previously unreleased. Without both facts, the agent rejected the market.
The decisive number arrived too late.
CME's official settlement calculation used trading activity through the market's exact close and published afterward. A decisive official value could not exist strictly before close, so the agent rejected the trade.
Full-window scan
Every new candidate stayed closed by default.
Captured the open-market baseline with both execution switches off.
Each new market lacked a reviewed exact mapping or provable pre-close official evidence. Quotes alone never authorized a trade.
Final poll 23:58:43.749 UTC. Wrote the hash-valid cutoff receipt at 23:59:00.005 UTC after 3,393 successful polls and 0 poll failures, then released the writer lease.
Reliability record
The outage stays in the story.
The earlier Gemini watcher was silent for 11 hours 10 minutes. The surviving logs prove the gap, but not the external event that ended the process.
110 ms forced child-exit recovery
115 ms stale-heartbeat recovery
1.151 s forced supervisor-exit recovery
Later no-trade fault-injection tests with execution disabled. These were not live competition recoveries.Observed vs simulated
Replay output never enters the score.
Realized P&L from zero submitted orders. The independently verified live ledger contains 2,530 hash-valid records.
25 hash-valid replay records, independently verified and isolated from live orders and realized competition P&L.
Evidence
- Public machine-readable closing record
- PR #28: child recovery and single-writer protection
- PR #29: supervisor recovery through cutoff
- CME WTI settlement procedure
- ODNI machine-readable download feed
Observed result checked 2026-08-24. Replay remains simulation.