Coverage

What this record holds, defined: every dimension of coverage counted over every tracked entry, each printed with the predicate that produced the count, and a grid of entries by the authority they name and the subject their own titles state.

132 of 132 tracked instruments. There are no percentages anywhere on this page. A coverage percentage needs a denominator meaning “everything there is”, and this record has never enumerated that population; every figure below is a count of entries against the population it was counted over.

Coverage, defined

A coverage figure with no published rule is this record asserting a judgement about its own completeness. So each count below carries the predicate it counts, and the predicate names the store it reads — the figure is checkable against the record rather than against this record’s word for it.

Neither figure qualifies the other, and neither is the other’s complement. 69 of the 132 entries are counted by both. The first counts what this record retrieved and retained; the second counts what an entry’s own status field holds. A retrieval is not a status, and an entry whose publisher states none can be retrieved every day.

The second figure and the lifecycle-metadata dimension below read one field between them and account for every entry: 74 and 58 over 132. Neither is the residue of the other. “Unknown” is a value this record publishes deliberately rather than a field left empty, and the several distinct readings behind it are kept apart and set out on how this record works.

The rule for the first figure. An entry is counted when prose was extracted from bytes this record retrieved for the entry, so a quotation can be checked against what is retained, and this record holds one or more retrievals of the document the entry names inside the window, each carrying a digest. The window is the 30 days ending 2026-09-18, opening on 2026-08-20 inclusive — one of the windows this record already counts in, taken through the single window rule every windowed figure here reads. Read from observations/<id>/*.jsonl · state/snapshots/<id>.<digest>.txt.gz · state/observation-class.jsonl. A retrieval counts where the snapshot the record names holds prose rather than the bytes of a document format nothing was extracted from. The projection is what says which retrievals were of the document the entry names and which were of something else.

132 of 132 entries hold at least one dated record inside the window, so the first figure is a measurement over entries that were read rather than a window in which nothing was read. A windowed count carries that third fact as well as the two every count carries, and it is stated because a measured count and an unread window look identical.

The dimensions

8 dimensions, in the order they were filed. That order is not a ranking, and nothing here weighs one dimension against another: a list read top to bottom invites the reading, so the rule is written down rather than left to the geometry. Each block carries the predicate its count was produced by, the store that predicate reads, and one sentence the count does not support.

identity 116 of 132 entries

Counted when the entry’s own source_kind holds a value other than index_only.

116 held · 16 not held · 0 unmeasured, over 132 entries. Read from src/content/instruments/<id>.md.

Not a claim that the address is the publisher’s canonical one, or that the document has only one address.

publisher 132 of 132 entries

Counted when the entry’s own authority holds a value.

132 held · 0 not held · 0 unmeasured, over 132 entries. Read from src/content/instruments/<id>.md.

Not a claim that the recorded body is the one that holds the copyright, that no other body publishes the same document, or that a count reaching every entry means anything was checked — the schema requires this field, so a value is present by construction.

lifecycle metadata 58 of 132 entries

Counted when the entry’s own status holds a value other than unknown.

58 held · 74 not held · 0 unmeasured, over 132 entries. Read from src/content/instruments/<id>.md.

Not a claim that the recorded value was read from the publisher today, which is what the verification dimension and the entry’s own page answer.

publication dates 0 of 132 entries

Counted when this record holds one or more archived versions of the entry carrying a published_date the publisher stated.

0 held · 132 not held · 0 unmeasured, over 132 entries. Read from src/content/archive/*.md. Each archive record names the entry it belongs to in its own instrument_id.

Not a claim that the publisher states no publication date, and not that this record holds no dates for the entry — nothing here searched a publisher’s surfaces for a publication date, and the dates an entry does carry are each named for what they record, none of them this one.

text retrieval 126 of 132 entries

Counted when prose was extracted from bytes this record retrieved for the entry, so a quotation can be checked against what is retained.

126 held · 6 not held · 0 unmeasured, over 132 entries. Read from observations/<id>/*.jsonl · state/snapshots/<id>.<digest>.txt.gz. A retrieval counts where the snapshot the record names holds prose rather than the bytes of a document format nothing was extracted from.

Not a claim that the document has no text, and never that its text may be reproduced — a restricted text is cited and linked, never excerpted.

historical text 77 of 132 entries

Counted when this record retains normalised text at two or more distinct digests of the document the entry names.

77 held · 55 not held · 0 unmeasured, over 132 entries. Read from observations/<id>/*.jsonl · state/observation-class.jsonl · state/snapshots/<id>.<digest>.txt.gz. A digest counts where the snapshot the record names is on disk and the projection calls that retrieval one of the document.

Not a claim that two distinct digests are two editions — a digest changes when the bytes change, and a page that oscillates between two renderings produces two digests and no new document.

verification 55 of 132 entries

Counted when the verification state this record derives for the entry is “verified” — a person checked it against the primary source.

55 held · 77 not held · 0 unmeasured, over 132 entries. Read from src/content/instruments/<id>.md. The same fields the mark on the entry’s own page is derived from.

Not a claim that an entry outside this count is awaiting a check — the observation register promises none, and the count of register-tier entries is printed beside this one.

change detection 125 of 132 entries

Counted when this record holds two or more retrievals of the document the entry names, each carrying a digest.

125 held · 7 not held · 0 unmeasured, over 132 entries. Read from observations/<id>/*.jsonl · state/observation-class.jsonl. The projection is what says which retrievals were of the document the entry names and which were of something else.

Not a claim that a change was detected — this counts entries where a retrieval had an earlier retrieval to be compared against, which is the condition under which “no change” is a measurement at all.

publisher is held on every entry in the corpus. A count the size of its own population is still a count over a stated rule, and it is printed rather than left out: what such a count does and does not support is in its own block above.

The paths above are this record’s own stores in the repository that produces this site. They name where a value came from; none of them is an address on this site.

25 of 132 entries are recorded at the observation register tier, where no person-verification is promised. They sit in the denominator of every dimension above, verification included, and nothing here removes them: a denominator that moves with the rule being counted is two populations wearing one fraction.

What a count here is about

Every predicate above is a fact about what this record holds. “This record retains no prose for the entry” is not “the document has no text”, and “no archived version carries a publisher-stated date” is not “the publisher states no date”. An absence sentence may only go where a search ran, and the searches behind these counts are readings of what this record already holds.

Held, not held, unmeasured

Three counts, summing to 132 on every dimension. Unmeasured is not a miss: where a reading could not be made, the entry is counted there rather than as not held, because folding a reader’s failure into an absence of evidence turns “this record could not look” into “this record looked and found none”, and those are opposite facts.

Every reading these figures need was made on this build, so unmeasured is a measured zero on all 8 dimensions and on the figure above them. What can fail to read is the projection recording which retrievals were of the document an entry names; with it unread, the dimensions that read it — historical text, change detection — and the first figure count unmeasured rather than not held. That list is derived from the same clauses the store lines above come from rather than typed beside them: a sentence naming which figures a store governs drifts the moment a predicate moves, and a page is where such a drift is least visible.

The evidence ladder counts something else

The evidence ladder on the observatory reads the same 132 entries under 12 rungs and 4 marks, and 3 of the 8 names above are rung names there as well. They are different measurements and neither corrects the other: a rung records what is held for one entry with a sentence beside it, and a dimension here is one predicate over the whole corpus, printed with its count. Where a figure here differs from a rung of the same name there, the two rules are the reason, and both are published.

The two axes

The rest of this page is the corpus counted a second way: every entry against the authority it names and the subject its own title states, both read from the entries themselves — 31 authority values and 25 column values.

Authority — 31 values

The distinct values of each entry’s own authority field, alphabetically. It is a free-text field with no registry behind it, so a value exists exactly while an entry names it, and these headings carry no permanent link for the same reason the ones on sources do not.

Subject, as stated — 25 values

Quotation rather than taxonomy: a value exists only where a publisher’s own title contains the term verbatim, and the entry carries the span it was read from and the address of that span. Matching is by containment, so an entry stating information security is counted under security as well — English morphology doing the work a synonym list would otherwise do, and a synonym list is a judgement about which words mean the same thing. Where publishers’ words share no substring the values stay apart, which is information about the vocabulary rather than a limitation. Ordered by count, then alphabetically; each column header carries its corpus count so the order can be checked against the table.

The grid

92 of 775 cells hold an entry. A grid this sparse is sparse before any question of coverage arises: a column exists as soon as one title anywhere in the corpus states its term, so most authorities have no entry under most columns by construction.

The columns sum to 167 against 132 entries, and the two are not meant to agree. An entry is counted under every column value its own title states, so an entry whose title carries several terms appears several times along its row. The column headed “title states no subject” holds 66 of 132 entries and is not a domain: it is what a reading of the title found, recorded as the epistemic fact rather than as a bucket named for everything the other columns missed. The reading was of the title as this record holds it. An entry may well state a subject in its body; nothing here has measured that.

What a cell says

A cell under a term holds one sentence: this record holds N entries recorded under this authority whose own title states this term. A cell at zero holds the same sentence with N at zero, and it is a measurement — every entry that authority names was read and none carried the term.

One column is not a term, and the sentence above is not its sentence. The column headed “title states no subject” counts entries whose own title stated no subject at all, so a cell in it reads this record holds N entries recorded under this authority whose own title states no subject, and a zero in it says every entry that authority names stated one. Its figure over the whole corpus is above, beside the grid.

What a cell does not say

Not that the authority has published nothing on the subject. Applying the rule is a judgement, and this record does not claim to be complete against it. The daily run observes the entries it already holds and does not search for what it does not have, so no cell on this page was produced by looking for a document that is absent. An absence here is not evidence that a document does not exist, is not in force, or does not govern.

Not that the authority does not use the term. The column vocabulary is read from the titles of the whole corpus, and publishers whose words share no substring stay in separate columns. A zero under one security word can sit on the same row as entries under another; that is a fact about vocabulary, not about what a publisher writes about.

Not that anyone has looked. Nothing on this page reports a search for a document this record does not hold, because no such search has been run.

So there is no gap cell on this page, and its absence is the point. An empty cell would say three things at once — this record does not hold such a document, no such document exists, nobody has looked — and only the first is something this record can state. A gap could be stated only where something enumerated what ought to be held and the corpus was measured against that enumeration. No such enumeration is published here, and none is inferred from the grid.

A source this record does not hold

A document that satisfies the inclusion rule and is absent from this record can be named at corrections@gxplex.com, which is the address this record already publishes for corrections. Whether an entry is added is a judgement someone makes and records; the addition, when it happens, appears on additions by month.

That is an address, not a form. There is no field on this page whose value the link carries and no message composed here for it to send: the link is the bare address, so nothing reaches this record except by someone opening their own mail client and choosing to send. Nothing queues, tracks or acknowledges what arrives, and there is no other channel — what this site stores on a device, and what it does not, is stated on how this record works.

Last updated: 18 September 2026