Our AI reviewer invented a request. Our producer retried 245 times.

A developer's team running ~100 unattended LLM agents on local models discovered a single document that was rewritten 245 times in 5 days, with a sibling document rewritten 225 times, totaling about 470 wasted generations. The root cause was a reviewer agent hallucinating a request that never existed, combined with a retry mechanism that counted reviews instead of contract failures, allowing the loop to run unbounded. The developer's audit of 2,038 reviews found a 0.2% hallucination rate, but the real risk was the infinite retry loop, not the rate itself.

We run ~100 LLM agents unattended on local models. Last week we found one document that had been rewritten 245 times in 5 days — every attempt rejected. A sibling document: 225 times. Combined, about 470 wasted generations, all burned on the same two files. Here is the autopsy, with the actual numbers. Our pipeline is simple: a producer agent writes a document, a reviewer agent checks it against a contract minimum length, required sections, no placeholder junk , and rejected work goes back with fix instructions. The rejected document was a key-management KMS implementation spec — 4,452 characters, perfectly on-topic. The reviewer's verdict: "The request was a 3-line email triage response LOCK / VERDICT / REASON , but the answer is a long KMS spec. Rewrite as3 lines only." One problem. We grepped the document: the words "LOCK", "VERDICT", and the name of the triage service appear zero times in it. The reviewer had invented the request. Two contracts collided: No output can satisfy both. So the producer failed the contract, got re-queued, produced again, failed again — 245 times. Our retry cap counted reviews , but a contract-failed output never reaches review. The give-up mechanism existed; it just watched the wrong counter. Our review prompt contained the artifact body first 4,000 chars and the output format. It never contained the original request. We asked a model "does this match the request?" without telling it what the request was. A model asked to judge against information it doesn't have will hallucinate that information. Ours did, confidently, 245 times' worth. Bonus failure: we truncated long documents to 4,000 characters before review without saying so, and reviewers marked them "thin — cut off mid-sentence." The cut was ours, not the producer's. We audited all 2,038 reviews on file for concrete terms product names, format tokens that appear in the review but nowhere in the reviewed document . Result: That's the uncomfortable lesson: a 0.2% hallucination rate produced 470 wasted runs, because nothing ever gave up. Low rate × infinite retries = unbounded damage. The rate is not the risk; the loop is. Each fix ships with a test we deliberately broke to confirm it fails. The checker that catches broken outputs in this story empty text, language leakage, placeholder junk, contract violations is free on npm: honto-contract https://www.npmjs.com/package/honto-contract — it passed 600 downloads last week, so somebody besides us finds this useful now. The unattended-operation checklist and three of our watchdog templates are free email-gated : Unattended-Operation Kit https://gxcafe.co.jp/harness-kit/?utm source=devto&utm medium=article&utm campaign=harness-kit The full set of 7 production templates cron registry, silent-zero watch, heartbeat, output contracts — the exact ones in this story is US$59 https://gxcafe.co.jp/harness-kit/?utm source=devto&utm medium=article&utm campaign=templates-pro . Honest note: we have no customers yet. Everything above is exactly what we run on ourselves, measured on our own failures.