Each bar is the logged misses of one month, split by who caught each one.
Counted at build time from every chained entry of the record, by the date on each entry. A logged miss is one somebody caught; what nobody caught is not here. The date is not always the day of the catch: 77 of the 740 were written into the ledger as a batch (ten or more in one second), and carry the batch's date.
Hours worked are the active hours on the home page: one operator's, counted since 2026-04-17, read on 2026-10-15 (UTC). The first miss on this record is dated 2026-06-13, so the overall rate divides by hours from before any miss was logged, which lowers it. Jun 2026 has no rate: the first miss is dated part-way through it, so its count covers less of the month than its hours do. 77 of Jul 2026's 277 were written as a batch and carry its date; without them its rate is 0.9. Hours through 2026-05-05 come from a table kept at the time. After that day they are pooled from exported chat conversations (counted through 2026-09-03), coding-session transcripts (counted from 2026-07-03) and commit times. On a day one of the first two has no records for, its hours are missing, so a rate over such days divides by too few hours and reads high. A month marked "on partial hours" is one where one of the two has no records for more than a tenth of its days.
Every miss this team logs lands in a tamper-evident record. This page counts all 740 of them and shows the newest 35, each with the opening of its note. Each one carries who its record says caught it: the human operator, the machine itself, a peer instance or an external instrument, or that nobody is named. The five gate rows are examples of the gate's rules.
The misses are the point. Every team has near-misses. Most bury them, and the chain of command never hears it. We sign ours and put them on the internet. A system you can trust is not one that never makes a mistake. It is one that catches its mistakes, records them, and shows you the record.
This shows that a miss was logged, the date on its record, and who the record says caught it. It does not prove any decision was correct. The gate rows are fixed examples, not a log of decisions. That is the honest claim, and the only one worth making.
The misses are real entries, anonymized. The gate rows are five fixed examples of the gate's rules, not dated decisions. In the record, every row carries a hash: gate rows are Ed25519-signed by the register key, misses hash-chained.
- /a miss was logged, with the date on its record and who the record says caught it (the human operator / the machine itself / a peer instance / an external instrument / nobody named)
- /the record is consistent with itself: when this page was built, the content of every entry hashed to the value stored for it, and every entry linked to the one before it. An edit to an entry's content shows as a broken link unless that entry's hash and every hash after it are rewritten too
- /that any decision was correct
- /that a human signed a gate decision: the gate rows are five fixed examples of the gate's rules, undated, signed by this register's key and by no person
- /that a signature is valid: the build recomputes the hashes and checks no signature, and an entry's signature and its integrity label are outside its hash
- /anything about the mission, location, target, or people involved
Recomputed at every build.
The gate rows are five fixed examples of the gate's rules: undated, and Ed25519-signed by the register attestation key (public half below), not by a person. No governance run's signed decisions are in this feed. Miss entries are hash-chained, and the head of the chain is signed.
The chain is recomputed from every entry each time this page is built, and the build counts a break as a broken link above. The record file itself is not on this site, so the recomputing cannot be repeated from outside: what a reader has is the chain head and the public key printed here. The site's other keys and its provenance surface are at /verify.