GxPlex

Compare

This record retrieves each tracked source on a repeating schedule and, where it can extract text, retains the normalised text it hashed. Where two of those readings are held for one source, the difference between them can be computed. This page indexes what can be compared, and states the conditions under which a comparison is refused.

A refusal is an answer, not a failure. Every pair that is not shown carries the name of the condition that stopped it, and the conditions are evaluated in a fixed order, so a pair refused on the first is never tested against the rest.

What the corpus holds

Each of the 132 tracked entries sits in exactly one of the four states below. The state records what this record retained, and not anything a publisher stated.

Every row in the tracker, by what its retained readings allow
StateRowsWhat the state records
two or more points 83 Two or more distinct retained digests, so at least one pair can be formed. Listed below.
one point 43 Every reading that produced a digest produced the same one. Nothing has been observed to differ from it, and a comparison needs two points.
no retained text none Digests exist and no normalised text is held for any of them, so there is nothing on this side to compare.
no digest 6 The row has been observed and no observation produced a digest. That is an absence in this record, not an absence of change at the source.

What the comparisons found

2781 pairs can be formed from the points held for the 83 rows in the first state above. Each is computed at build time and each carries one verdict.

Every pair the corpus can form, by verdict
VerdictPairsWhat the verdict records
difference 239 The words that differ are fewer than the ceiling and are shown.
withheld 1722 More words differ than the ceiling of 14 this site publishes quoted text under. The counts are shown and the words are not: a shortened quote is a fabricated measurement of what a publisher wrote, so it is withheld whole rather than cut to fit.
refused 820 A condition stopped the comparison before it was computed. The conditions are listed below.
no difference in the words none The two digests differ and the normalised words do not: what changed lies in spacing this record’s tokeniser folds together.
not measured none This record could not look. An unmeasured result is not a negative one and is not counted with the refusals.

The conditions, in the order they are evaluated

A pair names the FIRST condition that fired and makes no claim about the ones beneath it, so a count against one condition is not a count of the pairs that would have failed it. Not reached is not zero: three of the conditions belong to the form of the comparison that takes two dates, and every pair counted on this site is formed from points addressed by the key the store gives them, so those three are never evaluated here.

The conditions a comparison is evaluated against, in order, with the pairs each one refused across the whole corpus
ConditionPairs What the condition states
licence-restricted 762 No text from this document appears on this site, so no difference between two readings of it is shown either. The digests, the dates and the addresses are the record for this row.
licence-unread none No reading of the terms this source states is recorded for the row. That is an absent reading rather than an absent term, and the row is handled here exactly as a restricted one until a reading is recorded.
date-not-observed not reached No observation of this row on that day holds retained text.
date-not-unique not reached That day holds more than one distinct retained digest for this row, so a date alone does not name one observation. The addressed form — the day and the sequence within it — does.
one-point not reached Both dates resolve to the same observation.
referent-unclassified none The class of one of these observations could not be read, so neither was shown to be an observation of the document itself.
referent-class 15 One of these observations is not of the document. A listing page that names the document and the document are different subjects, and a difference between them is not a difference in the document.
referent-normalisation 39 These two digests were taken over different kinds of text — one over markup, one over prose extracted from a file — so the words they yield differ for a reason that is not the publisher’s.
referent-address 4 These two observations answered at different addresses. A difference between them is at least in part a difference between two pages.
text-not-retained none No normalised text is held here for one of these observations, so there is nothing on this side to compare.
snapshot-unreadable none Normalised text is recorded for one of these observations and could not be read from this record.
span-unverified none A span this comparison would have shown could not be found, word for word and in that order, in the retained text it was drawn from. The comparison is withheld rather than shown: a phrase assembled out of words a publisher did not write next to each other is a fabricated measurement of what the publisher wrote.
over-ceiling 1722 A faithful difference is longer than the word ceiling this site publishes under. It is withheld whole rather than cut to fit: a shortened quote is a fabricated measurement of what a publisher wrote. The figures below are counts of the difference, not the difference.

Text this record may not reproduce is not compared. 16 of the 83 rows with two or more points record their terms as something other than open, and every pair on those rows is refused before any retained text is opened. Their points, dates and digests are published; no words from them appear here.

17 of the 83 rows interleave their points. A digest was observed again after a later one had already appeared, so those points are not a sequence of versions but a set of readings, ordered by when each was first seen. Each such page says so on its face.

Rows with two or more points

Rows holding two or more distinct retained digests, with the verdict counts for their own pairs
Source Points Pairs Difference Withheld Refused Terms
ASTM E2363 3 3 none none 3 restricted
ASTM E2500 3 3 none none 3 restricted
ASTM E2537 2 1 none none 1 restricted
CISA CPG 2.0 2 1 none none 1 open
EMA AI reflection paper points interleave 2 1 none 1 none open
Annex 11 concept paper 2 1 none none 1 open
EMA CT computerised systems 2 1 none none 1 open
EMA data quality framework 2 1 none none 1 open
Annex 15 revision concept paper 2 1 none none 1 open
DI revision concept paper 2 1 none none 1 open
EMA work programme points interleave 2 1 1 none none open
EU AI Act 2 1 1 none none open
EU GMP Annex 1 2 1 none none 1 open
EU GMP Annex 10 2 1 none none 1 open
EU GMP Annex 11 points interleave 2 1 none none 1 open
EU GMP Annex 11 (draft) 2 1 none none 1 open
EU GMP Annex 12 2 1 none none 1 open
EU GMP Annex 13 2 1 none none 1 open
EU GMP Annex 14 2 1 none none 1 open
EU GMP Annex 15 2 1 none none 1 open
EU GMP Annex 16 2 1 none none 1 open
EU GMP Annex 17 2 1 none none 1 open
EU GMP Annex 19 2 1 none none 1 open
EU GMP Annex 2 2 1 none none 1 open
EU GMP Annex 21 2 1 none none 1 open
EU GMP Annex 22 3 3 none none 3 open
EU GMP Annex 3 2 1 none none 1 open
EU GMP Annex 6 2 1 none none 1 open
EU GMP Annex 7 2 1 none none 1 open
EU GMP Annex 8 2 1 none none 1 open
EU GMP Annex 9 2 1 none none 1 open
EU GMP Chapter 1 2 1 none none 1 open
EU GMP Chapter 2 2 1 none none 1 open
EU GMP Chapter 3 2 1 none none 1 open
EU GMP Chapter 4 2 1 none none 1 open
EU GMP Chapter 4 (draft) 2 1 none none 1 open
EU GMP Chapter 5 2 1 none none 1 open
EU GMP Chapter 6 2 1 none none 1 open
EU GMP Chapter 7 2 1 none none 1 open
EU GMP Chapter 8 2 1 none none 1 open
EU GMP Chapter 9 2 1 none none 1 open
EU Machinery Regulation 2 1 1 none none open
EU MDR 2 1 1 none none open
FDA AI guidance points interleave 4 6 none none 6 open
FDA AI device list 2 1 none none 1 open
FDA device cybersecurity points interleave 2 1 none none 1 open
FDA device cybersecurity (2026) 3 3 none none 3 open
FDA electronic systems (clinical) 2 1 none none 1 open
FDA/EMA AI principles 2 1 none none 1 open
FDA guidance agenda points interleave 2 1 none none 1 open
Health Canada ML 2 1 none 1 none open
IEC 62304 points interleave 2 1 none none 1 restricted
IEC 81001-5-1 3 3 none none 3 restricted
ISO 14971 points interleave 2 1 none none 1 restricted
ISO/IEC 22989 2 1 none none 1 restricted
ISO/IEC 24028 points interleave 18 153 none none 153 restricted
ISO/IEC 27017:2026 points interleave 2 1 none none 1 restricted
ISO/IEC 38507 points interleave 20 190 none none 190 restricted
ISO/IEC 42005 points interleave 21 210 none none 210 restricted
ISO/IEC 5259 points interleave 18 153 none none 153 restricted
ISO/IEC SC 42 points interleave 10 45 11 34 none open
GAMP AI Guide 7 21 none none 21 restricted
ISPE GAMP RDI 5 10 none none 10 restricted
GAMP 5 2nd Ed 5 10 none none 10 restricted
MHRA AI points interleave 8 28 28 none none open
MHRA AI Airlock 9 36 16 20 none open
MHRA data integrity points interleave 9 36 36 none none open
NIST agent standards 8 28 1 27 none open
NIST AI 600-1 2 1 none none 1 open
NIST AI RMF points interleave 3 3 none none 3 open
NIST AI RMF Playbook 2 1 none none 1 open
NIST CSF 2.0 2 1 none none 1 open
NISTIR 8183 2 1 none none 1 open
NMPA portal 3 3 1 2 none open
PIC/S GMP Guide 2 1 none none 1 restricted
PIC/S PI 041 2 1 1 none none open
PMDA AI 36 630 13 617 none open
21 CFR Part 11 12 66 55 11 none open
21 CFR Part 211 10 45 36 9 none open
21 CFR Part 820 (QMSR) 10 45 36 9 none open
WHO AI health 2 1 1 none none open
WHO data management 45 990 none 990 none open
WHO computerised systems 2 1 none 1 none open

The same thing for a machine

Every point and every verdict on this page is published at /api/v1/compare.json, with the condition that refused each pair. The observations the points are drawn from are at /api/v1/observations.json.

A difference here is a difference between two readings this record retained of one address. It is not a statement about what a document requires, about what changed in substance, or about anything that follows from it. The digest is of the normalised text rather than of the bytes served, so hashing a source as served will not reproduce it. Normalisation removes scripts, comments, navigation and page chrome, then strips tags. Where a tag is malformed the strip does not match it, so a fragment of a tag attribute can survive into the text that was hashed — a span shown here is a span of the normalised text this record hashed, and not necessarily of the document’s prose. How this record is built.

Last updated: