45 of 132, and what the score was worth
In September an agent-run audit scored this fleet 45 of 132 possible points on an enterprise rubric of 33 criteria worth 0 to 4 each. Several of them assume a team, such as a second reviewer and an on-call rota; one person cannot pass the rubric. The audit used 72.5 million tokens across 358 runs, and the result was mixed at best. As of 2026-09-08.
What it did
It read seven repositories and their history, every CI file, the alert rules and the published sites, then the running systems, read-only. It graded all of it against an enterprise rubric of 33 criteria scored 0 to 4.
What it cost
| Pass | Tokens | Tool calls | Agent runs |
|---|---|---|---|
| Component, lens, document, history and rules | 36,290,808 | 11,200 | 98 |
| Verification of all 109 reports | 27,460,793 | 8,106 | 218 |
| Findings register and inventory | 3,872,917 | 388 | 13 |
| Re-check of the 86 critical findings | 3,326,489 | 1,164 | 23 |
| Independent triple double-check | 1,369,950 | 471 | 5 |
| Editorial pass | 230,887 | 51 | 1 |
| Measured total | 72,551,844 | 21,380 | 358 |
Two runs reported no figure, so the total is a floor.
How much of it was right
The audit’s first pass wrote 109 reports. A second pass, separate agent runs, re-read every one of them.
2,269 items went into verification. 275 came out as not real.
Of the 2,269 items in those reports, 1,482 held unchanged, 512 needed a different severity or wording, and 275 were not real. Of the 86 critical findings re-checked against the live systems, 19 were off: 12 weaker than written, 6 at the wrong severity, 1 false. It also graded one report against a stale repository, inflated a finding through a broken search pattern, and handed checks back as “run this” instead of running them.
What the score measured
45 of 132. The rubric cannot be passed by one person: a bus factor of one, no second reviewer and no on-call rota all count as defects. It read merge requests merged by the account that opened them as missing review. There is one account here.
What came of it
As of 2026-09-08, 27 findings were closed, 61 had a change written and 2,327 had nothing. The nine open critical findings are not published.