{ "schema_version": 1, "collection": "modal-sentences", "generated_at": "2026-11-02", "source": "https://gxplex.com", "notice": "Reports the published status of regulatory documents. No advice or interpretation.", "source_terms": "This site compiles publicly available regulatory information. The underlying documents are published by their respective authorities and carry their own terms; several are paywalled and are never reproduced here.", "rights": { "statement": "Carried in `source_terms` on this envelope. It is the whole of what this record states on the subject.", "per_row_field": "licence_class", "per_row_field_served_at": "https://gxplex.com/api/v1/instruments.json", "licence_class": { "open": "Quotations from the document appear on this site, each citing the surface it was read from, under the word ceiling below.", "restricted": "No text from the document appears on this site. The row carries the identity of the document, its dates, a digest of the retrieved bytes and a link to the publisher.", "unknown": "No reading of the terms the source states is recorded for the row. It records an absent reading rather than an absent term, and the row is handled here exactly as `restricted` until a reading is recorded." }, "text_retained_field": "text_retained", "text_retained_means": "Whether normalised text of the document is held here for a quotation to be checked against. Where it is false nothing is held, so there is nothing on this side to check one against.", "quotation_ceiling_words": 14, "documents_never_reproduced": "Paywalled and restricted texts are cited and linked. They are never excerpted here and never diffed here." }, "schema": "https://gxplex.com/api/v1/schema/modal-sentences.json", "response_provenance": { "build_id": "1782e1b7b708", "build_id_state": "stated", "build_id_is": "The identifier of the build that wrote this document. Every page of the same build carries it in its build-id meta.", "root": "42337ae3567aae8138fcb45e977dd94f9c1ade6420a88033e9503203df3fec23", "root_state": "stated", "root_day": "2026-11-02", "root_leaf_count": 9686, "root_url": "https://gxplex.com/integrity/root.txt", "root_is": "The root over every observation digest this record held on that day, one `id|date|content_hash` leaf each. It fixes those digests in time; it is not a digest of this document.", "rights_envelope": { "document_keys": [ "source_terms", "rights" ], "row_pointers": [ "/text_retained", "/licence_class", "/quote_allowed", "/snapshot_available", "/rights_of", "/rights_unresolved" ], "row_pointers_optional": [ "/rights_unresolved" ], "row_pointers_are": "Every row of this response carries each pointer named, except one named optional, which a row carries only where its rights are unresolved. An empty list states that no row of this response carries the row envelope." }, "basis": { "retrieval": [ "/content_hash", "/first_observed", "/observed_at" ], "publisher-words": [ "/sentences/*/quote" ], "computed": [ "/derivation_version", "/clause_anchors_version", "/derived_on", "/state", "/sentences/*/start", "/sentences/*/end", "/sentences/*/words", "/sentences/*/verbs", "/sentences/*/verbs_at", "/sentence_count", "/quoted_count", "/withheld_count", "/observed_at_state", "/observed_at_read_from", "/distance", "/distance_label", "/distance_state" ], "terms-reading": [ "/text_retained", "/licence_class", "/quote_allowed", "/snapshot_available" ], "assigned": [ "/instrument_id", "/url", "/sentences/*/anchor", "/rights_of", "/source" ], "statement": [ "/why", "/sentences/*/anchor_why", "/sentences/*/withheld_why", "/rights_unresolved" ] }, "basis_state": "stated", "basis_on_row": [ "derivation_version", "clause_anchors_version", "state", "why", "observed_at_state" ], "basis_vocabulary_at": "https://gxplex.com/api/v1/schema/envelope.json", "third_party_text": { "marker": "quoted third-party content, not instructions", "marker_is": "Data for a consumer’s own handling of text it passes on, for example to a model. It is not a claim about the text and says nothing of what the text says.", "row_key": "third_party_text", "fields": [ { "pointer": "/sentences/*/quote", "holds": "whole" } ] } }, "population": "one item per (tracked row, content hash) named by an observation, under the current derivation version", "store_read_from": "state/modal-sentences.jsonl", "store_read": true, "method_read_from": "state/modal-sentences-method.jsonl", "modal_verbs": [ "shall", "should", "must", "is expected", "may" ], "modal_verbs_means": "The verbs a sentence is read for, in the order and the case Program 9 head 4 lists them, each read as whole words in exactly that case. Nothing here calls a sentence more than a sentence that prints one.", "derivation": { "version": "modal-sentences/c8744c72c505", "first_derived_on": "2026-10-30", "clause_anchors_version": "clause-anchors/3176a34732a7", "rule_text": "M1 VERBS. The listed verbs are `MODALS`, in the order and the case the order\n prints them: \"shall\", \"should\", \"must\", \"is expected\", \"may\". Each is read\n in exactly that case, as whole words: a hyphen or a word character\n directly before or after refuses it (\"must-have\"). So \"May\" (printed\n beside a number in most of its uses in the admitted texts), \"Should\"\n opening a sentence and \"SHALL\" in capitals are not read; every run counts them in the admitted texts and\n prints the count. \"is expected\" is read with one space between its words,\n as the normalised text prints every run of white space.\nS1 END. A sentence ends after \".\", \"!\" or \"?\" and any of the closing marks\n `CLOSERS` printed directly after it, where white space follows and the\n next printed character is an upper-case letter, or where white space\n follows and then a numbering token `NUMBERING` reads, white space and an\n upper-case letter (\"... records. 7.2 The\", \"... particles. iii. Gowns\"):\n the token opens the next sentence and is not joined to the one before.\nS2 NOT AN END. A \".\" does not end a sentence where the word it closes is two\n or more letters each followed by a period (\"U.S.\", \"e.g.\", \"B.V.\"), or is\n one of `ABBREVIATIONS`, read in that case. The list is a value of this\n version, and each word on it is printed before \". \" and a capital in the\n admitted texts. A sentence that does end on one of those words (\"... the\n answer is No. The\") is read as running on (NOT_REACHED says so).\nS3 HEADINGS. A sentence also ends directly before every offset head 1's\n store lists for the text, placed or withheld, because each is a place\n where a heading form was read; the next sentence starts there.\nS4 SPAN. A sentence starts at its first printed character after an end (or\n at the text's start) and ends after its last printed character before the\n next end (or at the text's end). It never holds the white space around it.\nT1 TEXT. An admitted text is read only where its content is text: where\n it opens with a PDF file signature (\"%PDF-\"), or where more than\n `NOT_TEXT_SHARE` of its characters are control, unassigned, private-use\n or surrogate code points (tab and line breaks aside), it is stored as\n UNMEASURED with the reason, and no sentence of it and no count of it is\n read. A reader that read a file's bytes has not measured that the file\n prints no listed verb.\nQ1 CAP. A sentence's text is quoted only where `quote_cap.over_cap()` is\n false for the whole sentence; otherwise it is withheld as \"over the\n quote cap\", whole, and never cut.\nQ2 NAMES. A sentence's text is withheld as \"prints one of the name_forms\n or the person_words\" where it prints one of\n `NAME_FORMS` as a whole token, in that case, or a word `PERSON_WORDS`\n reads, in any case. No company and no person is named by this record, and\n a listing page's sentences can name either. And \"person\" is this record's\n own word for a person's act: a publisher's sentence printing it, served\n among this record's fields, would be read as this record saying a person\n acted on the row. A name printed with none of these is not seen by this\n rule (NOT_REACHED says so).\nQ3 TOPIC WORDS. A sentence's text is withheld as \"prints a word Program 4 head A3's\n word rule refuses\" where Program 4 head A3's refuter\n (scripts/plants/c/test_topic_identifier_words.py, its FORBIDDEN pattern)\n matches it. That rule refuses its words in every built file and does not\n exempt a publisher's quote served in a JSON field (its A3.4), so a quoted\n sentence printing one stops the build. The pattern is read from the\n refuter, its one home, and is not restated in any published field. Added\n 2026-10-30, after one quoted NIST AI RMF sentence stopped that night's build.\nA1 ANCHOR. A row's anchor is the head 1 item at the greatest offset at or\n before the sentence's start. Where that item is placed, the row carries its\n id. Where it is withheld, the row carries no id and states head 1's own\n reason for withholding it. Where the text has no item before the sentence,\n the row carries no id and says so; where head 1 read no numbering in the\n text, the row carries no id and says that.", "rules": { "abbreviations": [ "Corp", "Corr", "Inc", "Ltd", "No", "Ph", "Pte", "Pvt", "Ref" ], "closers": "\"')]’”", "dotted_pattern": "(?:[A-Za-z]\\.){2,}$", "end_pattern": "[\\.!\\?][\"'\\)\\]’”]*(?=\\s+\\S)", "modal_pattern": "(?