Ship Nothing You Haven't Re-Read: Checksum-Gated Publishing for an Agent-Run Store
Description and use
The pipeline now verifies artifacts at the source rather than trusting agent summaries. Files are gated by SHA-256 hash comparison; listings are confirmed by independent public page reads; experiments use pre-written thresholds. This caught a real incident where a chunked file transfer flipped exactly one hex digit, producing a corrupt file that would have been sold undetected.
Problem
AI agents operating a music label and merch store produce fluent status reports, but these reports are outputs under test, not evidence. A listing can read correctly while the attached file is corrupt; a listing can show new photos while the production file behind it is still the old version; a price can be right in the dashboard and wrong on the public page. Reading agent output catches none of these artifact failures.
Outcome
The pipeline now verifies artifacts at the source rather than trusting agent summaries. Files are gated by SHA-256 hash comparison; listings are confirmed by independent public page reads; experiments use pre-written thresholds. This caught a real incident where a chunked file transfer flipped exactly one hex digit, producing a corrupt file that would have been sold undetected.
Workflow
- Generate a manifest for each product file containing name, byte count, and SHA-256 hash at the point of creation.
- The publishing machine recomputes the hash on arrival and compares it to the manifest hash. Mismatch means the file does not attach to any listing.
- For listings, an independent pass fetches the public page fresh and confirms what a buyer sees: title, price, photos, attached files.
- Any test an agent runs gets its pass/kill threshold written down before execution in a read-only file.
- Deliverables only attach to listings once they pass all gates: hash verification, byte count pre-check, public re-read confirmation, and fixed threshold rules.
Evidence
A description can read beautifully while the attached file is corrupt. A listing can show new photos while the production file behind it is still the old version. A price can be right in the dashboard and wrong on the public page. Reading catches none of these, because none of them are reading problems. They are artifact problems.
Every product file ships with a manifest — name, byte count, SHA-256 — generated where the file is made. The publishing machine recomputes the hash on arrival and compares. Mismatch means the file does not get attached to a listing, full stop.
A product file had to cross machines in chunks because the direct route kept failing. Every chunk reported success. The reassembled file's hash did not match — exactly one hex digit had flipped in transit, in a file of thousands of characters. As text it read fine. As a product it was corrupt, and it would have been sold that way.
Two listings received new artwork — re-rendered, re-cropped, visibly better — and the photos went up cleanly. What did not change was the print file bound to those products on the production side, because binding lives in a different system than photo management. Every surface check passed. A customer would have received the old artwork from a listing showing the new.
Any test an agent runs gets its pass/kill threshold written down before it starts, in a file the running agent treats as read-only. This is the same discipline as a pre-registered trial, and for the same reason: the system doing the work cannot be trusted to grade it afterwards.
Make the checker independent. A different pass, a different route, ideally a different fetch path than the producer used. Shared assumptions are invisible to a single reader.
- Record date: 2026-10-23T00:26:27.346Z