How this is kept

What is tracked

132 regulations, guidance documents and standards from 31 publishers across 10 jurisdictions. Each one is collected on a daily schedule, and not every day yields a reading: its observations page lists each record of it with its day, and with a digest where one was recorded. The series begins 2026-07-28 and has run for 98 days.

For each document this record holds what the publisher's own page said, the day it was read, the address it answered at, and — where the text is public — a digest of that text so a later change can be proved rather than asserted.

8 of the 132 tracked entries declare an address on eur-lex.europa.eu. Their documents were digest-bearing on 30% of the calendar days expected: 225 of 757, summed over the 8, each counted from its own first document day to 2026-11-02. The publisher serves an automated-reader challenge (read 2026-09-28 at the declared address of one of them: HTTP 202, 2,035 bytes, a response header whose value is “challenge”), which this record does not run. The nightly re-read of that host’s robots file and of the terms page its filing classified measured both on 3 of 30 days from 2026-10-04 to 2026-11-02, most recently 2026-10-18, and no read differed from the one before it; on 27 others both were unmeasured (robots address: HTTP 202). Retrieval limit on the entry whose address was read.

What is not tracked, and what is not yet determined

Some documents cannot be read in full, and the reason is recorded on each one.

• 23 are published under a licence that does not permit taking the text. The publisher's own catalogue entry is followed instead, so a stage change is still visible; the words are not reproduced.
• 9 exist only as an entry in a list, with no document behind them yet.
• 6 have no public document at all.
• 3 come from sites that refuse automated requests. That refusal is honoured.
• 3 are registers, which record entries rather than carry a status of their own.

A further 44 are not re-checked, and the reason each one records is: 23 paywalled — progression tracked; 9 listing only; 6 not a document; 3 blocks automated requests; 3 register states no status. Ordered by count, then alphabetically.

A document that cannot be read is still listed, with the reason, rather than quietly dropped. An absence you can see is worth more than a tidy list.

Why a status is not yet determined

74 of 132 entries carry no determined status. That is this record’s largest single fact about itself, and every one of them records why:

Every reason below is read from a row’s own recorded field or from an outcome its retrieval ladder recorded — never from this archive’s own prose about the row. No status is inferred: a row with nothing recorded is counted as having no recorded reason.

not yet requested — 0

the ladder made no request — this run's budget or a robots policy

register entry — no confirmation promised — 26

record_tier: register, or frozen_reason: register_no_status

text behind a paywall — 13

frozen_reason: paywalled_stage_tracked

published only within a listing — 9

frozen_reason: listing_only

no public document — 6

frozen_reason: no_public_document

access blocked to an identified client — 3

frozen_reason: access_blocked

read, and it states no status — 17

a rung at a publisher-published address fetched the document; no status passage was found

the address returns not-found — 0

a rung at a publisher-published address received HTTP 404 or 410

the request was refused — 0

a rung at a publisher-published address received HTTP 401, 402 or 403

no reason recorded — 0

nothing held names a reason — this is an absence of a reason, not a kind

What the verification marks mean

The board carries a verification column. Each mark is defined here — and it is defined SOMEWHERE, on purpose: the same labels are drawn onto the social cards, which are images and cannot carry a correction. A label that appears only on an uncorrectable image is a claim with no place to be fixed.

Nothing here is interpreted

This record reports what a document says and when it changed. It does not say what a document requires of you, whether it applies to your product, or what action it calls for. There is no advice here, no assessment, and no ranking.

That is a deliberate limit, not an unfinished feature. A reported status is checkable against the publisher's own page; an interpretation is not, and a record that published both would make the checkable half harder to trust.

Beneath that limit sits a narrower list: readings available from what this record already holds, and not taken. What this record refuses to infer names each one with the gate that stops a build where it appears and the dated correction filed when this record drew it.

Who keeps it

One person, working alone, with the collection automated and every confirmation stating the act that made it. No organisation stands behind it and no employer is involved.

What survives that does not depend on them: the underlying record is plain text, one object per line, with the digests and the dates in the open. It can be read, checked and rebuilt without this site and without its code.

How the record is produced — who does the work, under what orders, what no model writes, and what only a person does — is set out on How this record is made.

The rules by which the record reads the texts it retains — how a clause is anchored, which defining forms and listed verbs are read, and each threshold — are set out with every version and its first day on Derivation methods.

What this site stores on a device. A value is saved in the browser’s local storage, under a key beginning gxplex-, only when one of these controls is used — the compact-rows control on the board (gxplex-density), the orientation’s dismiss control on the board (gxplex-orientation), the keep control on Developers (gxplex-kept, with the day it was used), and the control that keeps the figures on Developers (gxplex-figures, with the day it was used) — and nothing is stored on a visit that uses none of them; there are no cookies and no accounts, nothing saved is ever sent anywhere or read by this site’s servers — every visitor is served identical bytes — and Forget this device on the board clears all of it in one tap.

When something here is wrong

The corrections policy. A published claim about a document — its status, a date, a quoted span, a recorded value — is never silently edited: the correction is added with its date and the wording it replaced, and it stays on corrections permanently; corrections to this record’s own machinery, such as a counting rule or a page template, are kept in the repository history instead, because they change nothing this site has said about any document. The exception is a page whose published figures changed what they count, which corrections lists in a table of its own.

Factual claims are never edited silently. A correction is added, dated, and kept beside what it corrects — 301 correction(s) recorded so far, counted over every changelog entry whose note opens with the word “correction”. Document claims: 143, each a correction to a claim about a tracked document or about an axis over all of them, every one listed on corrections. Site machinery: 158, every correction not listed as a document claim; of those, 87 are scoped to a surface this record publishes and are listed on corrections in their own table, and 71 are kept in the repository history, 26 of them scoped to a row this record no longer tracks, whose subject is a document, and the rest scoped to a collection. Figures: 7 of the 113 corrections that state a class name the class figure, a published number that was wrong or that named a population other than its label; the other 188 state no class, so whether they repair a figure is not recorded. This subset crosses the other two. A record whose past statements can change without trace cannot be cited, because a reader who quoted it last month has no way to know whether the quotation still stands.

What gets corrected: any published factual claim — a status, a date, a digest, a count, or a sentence that overstated what the record establishes. How: the original stays, the correction is added beside it with its date and what it changed. Where: corrections lists the document claims and the surface-scoped table; the whole log is kept in the repository history, and Developers prints the command that reads it.

An error can be sent to corrections@gxplex.com. There is no mechanism that would record whether a correction originated outside this record, so no count of that kind is published.

Standing statements

Their changelog entry is dated 2026-10-31. The order behind them names a third, on removal and correction, which this page does not carry. As in the lists on Independence, each closes by saying whether code holds it or no gate measures it.

Independence

No financer, sponsor, investor or partner stands behind this record, and no employer is involved in it. The individual who keeps it, described under Who keeps it, is not named on this site.

Policy — no gate measures it.

What silence means

A signed release is one closed month of the record, cut into a directory of its own and signed with the record’s key. No signed release is in this build, and the build serves no readable signing key.

The order this record is kept under, Program 14 item 4.1 of 2026-10-29, which is not published on this site, names 90 days without a new signed release as the point at which the last signed release stands as the frozen record, with its final root and count published. No .py file under scripts/ or ops/ counts those days, outside the plants and the regression corpus.

The code that acts on silence is the dead-man in scripts/succession.py. It counts as a completed night any run in the record’s daily log, attended or not, that reached its end with its deploy reported and no hard stop. Invoked once 14 whole record days have passed with none, it takes the newest signed release that verifies against the record’s published key and writes a separate mark over it, signed with the record’s key and reading “final unless resumed”; the release itself is not changed, and a silence that continues gets no second mark. With no such release it writes no mark, and where the log cannot be read or holds no completed night to count from, it does not act. A later completed night resumes the record, and the mark is kept. The mark is kept in the record’s own state, and no route on this site serves it. The nightly does not invoke it.

The verification kit the same order names for every release, a VERIFY.md holding each command and its expected output, is not built. Verify states that no signing key exists yet.

Held in part. The dead-man threshold in scripts/succession.py holds its 14 days; whether the nightly invokes it and whether the release cutter names the kit are read from scripts/daily.py, the Makefile and scripts/record_release.py when this page is built; the releases and the key are read as Verify reads them. The order’s interval is the order’s own; scripts/plants/b/test_standing_statements.py refuses this page on the day a .py file under scripts/ or ops/, outside the plants and the regression corpus, counts it in a form its census reads while the sentence above still says none does, or names a module its census did not find.