Integrity

TEMPORAL 1 finding

The question

A dated event about a document may not fall before the year the document's own designator carries. `Regulation (EU) 2026/1744` cannot have an event in 2024.

Population

13 calendar event(s) whose row carries a number-year in its title

Test

temporal.number-year-precedes-event

Known-satisfying row

eu-cyber-resilience-act — a 2024-numbered act with an event dated 2027-12-11 — later than its own number-year, which is the satisfying shape. This family must stay silent about it. If it does not, nothing here is published: a test wrong about a row whose answer is known is a reader whose other findings cannot be trusted.

REFERENTIAL 5 findings

The question

A row that declares its address is the document must hash the document. Where the referent audit finds the address serves a hub, a listing or another artifact, the two disagree and neither is silently preferred.

Population

132 audited row(s)

Test

referential.declared-vs-served

Known-satisfying row

astm-e2363 — declared a document, and the referent audit returned AUDITED-OK: the row and the address agree. This family must stay silent about it. If it does not, nothing here is published: a test wrong about a row whose answer is known is a reader whose other findings cannot be trusted.

TYPE no finding

The question

An attribute exists on a type only where the publisher can state it. A row whose publisher states it is not a document may not publish a document's lifecycle status.

Population

0 row(s) with a stated non-document object type

Test

type.attribute-excluded-by-object

Known-satisfying row

astm-e2363 — object type `document` with a published status — the attribute its type can carry. This family must stay silent about it. If it does not, nothing here is published: a test wrong about a row whose answer is known is a reader whose other findings cannot be trusted.

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/integritygates.py require this family to produce a finding against a seeded row before its real result is trusted, so a test that reports nothing has been shown capable of reporting something.

PROVENANCE no finding

The question

A published human verification needs a recorded act behind it. A row saying a person checked it, with no event naming the actor and the date, is a claim with no receipt.

Population

58 row(s) publishing a human verification

Test

provenance.verified-without-event

Known-satisfying row

fda-device-cybersecurity-2026 — publishes a human verification AND has the event behind it; the id ends in a year, which is what the reader once mangled. This family must stay silent about it. If it does not, nothing here is published: a test wrong about a row whose answer is known is a reader whose other findings cannot be trusted.

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/integritygates.py require this family to produce a finding against a seeded row before its real result is trusted, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · one figure, two surfaces no finding

The question

A figure a reader meets on two surfaces derives from one function. Where the store and the page carry the same named figure, they must carry the same value.

Population

7 comparison(s): figures published on both the store and the page, plus populations counted under more than one name across stores

Test

reconciliation.figure-on-two-surfaces

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

Known-satisfying row

front-door — a figure published on both the store and the page, carrying the same value on each. This family must stay silent about it. If it does not, nothing here is published: a test wrong about a row whose answer is known is a reader whose other findings cannot be trusted.

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · homepage verified = entry aggregate no finding

The question

The homepage’s verified figure equals the number of entries whose own published record is in the verified state: the per-entry records /api/v1/instruments.json publishes, one per entry, and the board’s own rows.

Population

132 per-entry record(s) and 132 board row(s), against one homepage figure

Test

reconciliation.home-verified-equals-entries

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · observatory differences = authority aggregate no finding

The question

For each window, the entries /observatory counts under the section headed “document changes” equal the sum, over every authority page, of the entries whose text changed in that window; and the corpus each side counts over is one size.

Population

3 windows over 31 authority page(s) and one /observatory section

Test

reconciliation.observatory-equals-authorities

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · findings candidates = candidate store 1 finding

The question

The candidate figures /findings and /candidates publish equal what the candidate store holds: the staging directory, the verdict ledger, and the declared set of decisions that resolve a candidate. Detected equals the distinct candidates across directory and ledger; staged equals the directory, one page per staged file; and each published count of candidates with no resolving decision equals the store’s.

Population

108 candidate(s) in the store, 108 staged, 108 candidate page(s), 2 surfaces

Test

reconciliation.findings-equal-candidate-store

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

RECONCILIATION · calendar future + past = total no finding

The question

On /calendar, forthcoming plus past equals the stated total; each part equals the rows rendered under it; and the total equals the dated events the calendar store holds.

Population

17 dated event(s) in the store, 17 rendered row(s)

Test

reconciliation.calendar-partition

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · sources = board no finding

The question

The entries /sources lists are the entries the board carries, id for id, and the count /sources states is the board’s row count.

Population

132 entry link(s) on /sources, 132 board row(s)

Test

reconciliation.sources-equal-board

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · status totals = board population no finding

The question

Every published status total sums to the board’s population and equals, value for value, the rows the board marks with that status: the board’s own status facet, and the status distribution /findings publishes.

Population

132 board row(s), 6 facet value(s), 6 distribution value(s)

Test

reconciliation.status-totals-equal-board

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · every verified state has a receipt no finding

The question

Every per-entry record that publishes a verification carries the act behind it. A state saying a person read the status needs a person’s status event in the verification log; a published verification date needs a person’s act on that day; a state read by tooling needs a status extraction event on its check date. An event is a person’s act only where its actor is a person, and an event the log later voids or disowns is no receipt.

Population

56 entry record(s) in a person’s state, 58 publishing a verification date, 1 read by tooling, over 132 per-entry records

Test

reconciliation.verified-state-has-receipt

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · every publisher event has a citation 86 findings

The question

Every event this record dates on a publisher’s word cites the publisher: each dated event in the calendar store carries a source address, each row /calendar renders carries a source link, the two cite the same sources, each status event /api/v1/events.json publishes carries its source address, and each record the signal store holds carries its source address.

Population

17 dated calendar event(s), 17 rendered row(s), 55 published status event(s); 86 signal record(s)

Test

reconciliation.publisher-event-has-citation

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

RECONCILIATION · every detected difference has two observations no finding

The question

Every difference this record detects names two observations the store holds, with different digests: each staged candidate’s before and after, each on the day the candidate names, and each difference /api/v1/events.json publishes, the later on its own day and the earlier on or before it.

Population

108 staged candidate(s) and 312 published difference(s)

Test

reconciliation.difference-has-two-observations

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

RECONCILIATION · every quotation ≤ ceiling 2 findings

The question

Every quotation this record publishes is at most 14 words, the one ceiling its quote-cap module holds. A quotation is text the built site marks as borrowed: any element carrying the class this record’s templates give a quotation, every other blockquote whatever its class, an inline quotation element, and a quote-named value in the API. Typographic quotation marks in running prose are not such a mark, since they enclose this record’s own wording as often as a publisher’s, and this test does not read them.

Population

39136 span(s) on built pages, by form: blockquote 40, blockquote.pubpassage 5, q 1, the quotation class 39090; and 1997 API value(s). Set apart and not measured as borrowed text: 172 a correction’s from/to wording, quoted from this record’s own changelog; 1 blockquote(s) on /papers/, declared by the quote-fidelity gate as hand-written paper prose; 2 this record’s own generated status sentence(s)

Test

reconciliation.quotation-within-ceiling

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

RECONCILIATION · every correction references the prior claim 32 findings

The question

Every correction names the claim it corrects: each changelog entry whose note opens with the word correction carries its prior claim in the entry’s own field, and each such entry /corrections renders shows that prior claim on the from side of its from/to wording. The note beside it is not read for the claim.

Population

301 correction(s) in the changelog, 230 row(s) on /corrections, 225 of them rendering an entry that carries a prior claim

Test

reconciliation.correction-references-prior

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

RECONCILIATION · one word, one population on a page no finding

The question

A word a page writes beside two numbers names one population, or the page carries a sentence writing both numbers under the word that says which is which; a word followed by its agent (by a person) names its own population, and a figure confirmed by analysis or by derivation is read under the bare word, because the record writes a derived act as confirmed by analysis. The words read are: confirmed; a word outside them is not read, and a sentence writing two numbers under the word is read as their bridge whatever it says, two equal numbers being read as one. A figure is a numeral in a page's text: a number spelled out in words, or written in an attribute value, is not read, and after a colon only the first number is read. The confirmed-revisions figure is the same on the front door, /changes and /candidates, where /candidates' figure is the members of its confirmed group whose subject is the instrument, each member read, never a subtraction.

Population

9 word figure(s) over 1351 built page(s); the confirmed-revisions figure on 3 surface(s)

Test

reconciliation.one-word-one-population

A finding here

is reported, and withholds nothing until this family’s first measured zero (above)

No row in that population answers the question the wrong way. A zero here is a measured zero: the fixtures in scripts/invariants.py require each place this test writes a finding to produce a finding against a seeded row of its own before its real result is trusted, and they fail on a copy of its code in which any one comparison, or any one clause of a compound condition, that decides whether it writes a finding is reversed or removed, so a test that reports nothing has been shown capable of reporting something.

Bytes not transferred on a publisher’s validator

The collector sends a row’s request with the ETag and Last-Modified values its last observation kept, where that observation was taken at the address the row now names and vouches for a digest. A server that answers 304 at the address the validators came from is recorded as “checked, unchanged” with basis “publisher-validator” and no digest of its own; a 304 at any other address is recorded as a fetch. A row’s request carries no validator once its last full fetch is 7 days old. The figure for a record so checked is the response body bytes this collector read on the full reading whose digest the record vouches for, as that reading recorded them; that is the unit of every byte figure in this section. The collector reads a body up to 4,000,000 bytes, so a reading cut at that ceiling counts the ceiling and not the document’s length. A night is the record’s own day, as the collecting machine’s calendar dated it; a host is the host of the address that answered, as that address spells it.

Of 24,082 dated records in the store, 252 keep a validator for a later request, 198 were made with one, and 60 were checked unchanged. 31 night-and-host pairs hold them, below, newest night first.

CRAWL: this record’s own host, as named crawlers are answered

Each night that reaches the deploy step, the collecting machine asks this record’s own host for /robots.txt and for one page of the build that prints documents’ own clauses, once for each crawler that item 0.1 of the program of 2026-10-29 names, with that crawler’s name first in the user-agent string and this record’s name after it. A pause separates every two requests, and no other host is asked. The page probed stays the same from night to night while the build holds it. The store, state/crawl-probe.jsonl, keeps every request’s user-agent string, status and body digest, and the Content-Signal lines of every /robots.txt answer. A night is red when a request is answered with any status other than 200, when a body is a challenge page (read from the body, never from the status), or when the served robots file is not byte for byte the robots file of the build the host serves. A request with no answer is unmeasured, not red, and so is a not-found answer to the probed page from a build other than this one. Where a night holds more than one run, the last run’s row is the night’s. The probed page is compared with this record’s build only on a night the host serves that build; on any other night it reads “not compared”.

A red night opens an operator item, K8, on the weekly list the closure rules keep, in that rule’s own store, naming four host settings to be read on the host. A later green night closes it, and so does its clock, which the item states when it opens, whichever comes first; an acknowledgement recorded on the weekly list closes it too.

1 night probed in 1 run, newest first. No K8 item is open. What the probe cannot see: a host rule keyed on a crawler’s own network addresses, any page but the one probed, and a rule that answers this probe one way and the crawler itself another.

Files in the build’s output directory

On the build these results were measured on, which the gates ran over before this page was written, build 560a30388c8c, 4,250 files were under the build’s output directory once the bundler and the search index had written it, counted on 2026-11-02. That directory is the one the deploy step hands the host. The order this record is kept under, Program 14 item 2.8 of 2026-10-29, which is not published on this site, names 15,000 files as the figure this count is compared with, and this count is below it.