The short version: our pipeline's AI provider account ran out of funds on August 3rd, and the system that writes these briefs was effectively offline until August 12th. Monday's issue (Aug 10) went out on schedule — but it was built from August 1st data. The "as of" dates were accurate, as always; the freshness wasn't. That's on me.
What kept working: claim scoring. Resolutions run against market data independently of the AI, so every claim that matured during the outage was scored correctly and is in the ledger — 374 claims were scored in that window, none of them graded by anything other than realized returns.
What didn't: no new analysis was produced for nine days, and the weekly file you received reflected week-old reads. Nothing wrong was published — but stale isn't the standard either.
What changed: the pipeline is caught up as of today — fresh briefs on all 31 names, current scorecard 56% on 268 decided calls (95% CI 50–62%, market-adjusted) — and the monitoring that watches this system now alarms within minutes if the AI layer ever fails on billing, and within 48 hours if it goes silent for any other reason. The outage was invisible precisely because every component reported success while doing nothing. If you've followed this project, you'll recognize that failure mode: it's the same one we hunt in our own scoring. Now it's fixed in our operations too.
Next Monday's file will be normal. The full record, as always: receiptsfirst.com/receipts
— Receipts First