Status

This is Hunch Book's incident log, shown on the app's status page. Anything that stops a market, delays a settlement or a payout, or puts funds at risk is logged here, newest first. Each entry has a ## date: title heading, then what happened, who was affected, what was done (with links to the transactions), and what changes so it does not happen again.

2026-10-05: keeper and maker stopped for 22 hours, market #1 settled late

What happened. On Monad testnet, Hunch Book's keeper and maker bot run as background processes on one machine. Both received SIGTERM at 20:59 UTC on 2026-10-04, then a second SIGTERM 11 seconds later, the signature of the processes being stopped together with the shell session that started them. The keeper finished its cycle and exited; the maker exited before it could cancel its two resting orders. Neither was restarted until 19:43 UTC on 2026-10-05, and that first restart came up as a dry run, because the setting that turns sending on lived in the shell that had started them before, not in a file. They ran live from 19:45 UTC.

Who was affected.

• Market #1 0x2A44…3982 closed at 00:56 UTC on 2026-10-05 and was not settled for 18 hours 49 minutes. Its settlement deadline is 7 days after close, so it never risked a void. Every token in it is held by our own wallets.
• A touch market 0xD45e…1000 could have been proved YES about 22 hours earlier.
• Two maker orders stayed on market #1's Kuru book after close. No one traded against them.
• Nothing was lost: settlement and redemption never depend on the keeper (anyone can call settle), and no funds moved without their owner.

What was done.

• The keeper settled market #1 NO at 19:45 UTC (tx) and proved the touch market YES (tx).
• The maker redeemed its NO tokens (tx).
• The liveness watchdog had opened a warning issue at 11:44 UTC ("closed 10 h 47 min ago and not settled yet"). It was closed after the settlement.

What changes.

• scripts/run-local-services.sh now starts each service in a session of its own, so the shell or tool session that ran start can no longer take it down when it ends.
• Live settings (KEEPER_ENABLED, MAKER_ENABLED and the rest that are not secrets) live in .run/services.env; start warns loudly when a service would run as a dry run, and config shows what a start would use. launchd runs the same script, so both paths take the same settings.
• On macOS, start keeps the machine awake while the services run.
• The services still run on one machine. Moving them to a host that restarts them on its own (ops/README.md) is the lasting fix.