1. What happened
On 3 September 2026, for twelve minutes, the public register of this Registry published an edition line reading “9 fleets · 42 agents · 41 in class · 5 with terms lodged”, and beneath it the sentence “We are still on our own register — alongside 8 fleets we do not control.”
One fleet on the register belonged to a member. The other eight were demonstration fleets built and operated by FleetRegistry. They were not fleets we do not control. They were ours.
The record was never wrong. Rule F4.3 — demonstration entries are excluded from every published total — was in force, and public_register().totals, the function that implements it, computed the exclusion correctly for the entire period. The page did not call it. It summed the entries it had just finished rendering and printed the sum, and it arrived at “fleets we do not control” by counting the listings and subtracting one.
A correct rule, a correct function implementing it, and a second path to the same figures that consulted neither. Every automated check in place asserted that public_register() was correct, and it was. Not one asserted what the page rendered.
It was found by the founder, looking at the website.
2. What FleetRegistry would not have done
FleetRegistry would not have caught this.
That is the sentence this series writes about somebody else five times, and it is not softened here. We do not survey source code. We do not compare a rendered page against the function that is supposed to produce it. No notation suspends when a page and a query disagree, and there is no class condition that opens because a number on a website was arrived at twice. A registry entry for this fleet, held by this Registry, would have said nothing about this incident for as long as it lasted.
We also do not know how many people read the false line while it stood, and we are not able to find out.
3. What the record held at the time
The record was correct, and it was available, for the whole period.
As this page is served, public_register().totals reports 1 fleet · 5 agents · 3 in class, with 8 demonstration fleets excluded from every one of those figures under Rule F4.3. That function returned the correct answer throughout the incident. Nobody asked it.
The correction is the primary source for this report, and it is read rather than retold. What follows is the entry as it stands in the corrections register at the moment this page was served — the same block, from the same function, that /corrections renders. If it is amended, this report changes with it. If it cannot be read, this page does not render.
FR-C-0002/register — the public register, edition line and ownership statementOn 2026-09-03 · 12 minutes
What we published
On 3 September 2026 the public register published an edition line reading "9 fleets · 42 agents · 41 in class · 5 with terms lodged", and beneath it the statement "We are still on our own register — alongside 8 fleets we do not control."
The rule it was accepted under
Rules F4.3 — demonstration entries are excluded from every published total. The rule was in force and the function implementing it, public_register().totals, computed the exclusion correctly throughout. The page did not consult it: it re-derived the edition line by summing every listed entry, and counted "fleets we do not control" as the listing count minus one, with no filter for demonstrations.
What has changed
The edition line and the ownership statement are now read from a single source, public_register().totals, which honours F4.3. Where that source is unavailable the figures are omitted rather than recomputed, because a number the Registry cannot derive correctly is not published. A database trigger now refuses to list a demonstration fleet on the public register at all, and every demonstration entry carries a visible banner in addition to its DEM notation.
The correction
One member fleet was entered in the register at that time, with five agents, three of them in class. The other eight fleets were demonstration fleets built and operated by FleetRegistry itself. They were not fleets we do not control; they were ours. No member fleet was misrepresented and no member's classification was affected.
Corrections are append-only and hash-chained; each seal commits to the one before it. The entry above cannot be edited or removed, and this report cannot quietly disagree with it.
4. What the record still could not answer
Who saw it. The register is a public page. It is not instrumented to identify readers, and we would not publish that if it were. Anyone who read the register during the period and discounted it, or over-counted it, did so on our word and we cannot reach them.Whether anyone acted on it. A false figure that nobody relies on is still a false figure, and this Registry does not grade its own errors by whether they were expensive.Whether it was the only one. The footer of every page on this site, including the register itself, had asserted “design prototype, not deployed” since 2 September while the register was live and in operation. It was corrected the same night as this one (FR-C-0001). Two published falsehoods about our own standing, found within thirty hours of each other, both by a person reading a page rather than a check firing. We do not know that there is not a third.
5. What changed
One source, or none. The edition line and the ownership sentence are read from public_register().totals and from nowhere else. Where that source is unavailable the figures are omitted rather than recomputed — a number this Registry cannot derive correctly is not published.The rule moved into the database. A trigger now refuses to list a demonstration fleet on the public register at all, so the condition that made the mistake possible cannot be re-created by a page.Demonstrations are visible, not merely notated. Every demonstration entry carries a banner in plain words. DEM inside a class notation was never enough at the entry level.This index counts itself. The page you reached this report from used to state, by hand, how many reports had been published, how many failure modes had no coverage, and how many incidents in the series this instrument would have prevented. Three claims about the record, typed. They are now counted from the register of reports, which is the same table that decides which reports exist — so the index cannot describe a series other than the one it lists.
6. The mode this report adds
FM-11 · parallel derivation. An authority for a published figure existed and was correct; the figure that was published came from a second path that did not consult it.
A mode earns a place in this taxonomy only when it implies a different survey scope, and this one does. No behavioural survey finds it: the agents behaved, the function answered, the data was sound. It is found by asking, of every published figure, which single authority produces it — and then looking for a second path that can produce it without asking. That is a survey of the renderer, not of the fleet, and it is a scope this Registry did not have until it needed one. The taxonomy now defines eleven modes.
Lloyd’s rules grew this way. Every clause is a shipwreck; this clause is ours.
7. Sources
This Registry’s corrections register, entry FR-C-0002 — read live at the time this page was served (/corrections)Rules for the Classification of Agent Fleets — F4.3, G1.3, G1.4, G2.2; the correction itself is lodged under F4.3, G2.2, G1.3The public register — the page the incident occurred on, and the figures it publishes now
FleetRegistry does not prevent incidents. It makes them reconstructable and it makes the authority behind them attributable. This one was reconstructed from our own record, which is the only test of that claim that means anything.