Repository navigation
Conversation
…SE/IT)
Before: paperkit proves that a number in the text traces back to a signed,
hashed source. It cannot tell whether that source is a statute, whether the
statute is in force on the reference date, or whether the number really
appears in the statute — and all three failures are silent. A French statute
that was amended keeps its old version online, so citing by article number
quietly yields the previous wording; statutes also spell numbers as words
("trente-cinq heures"), which a literal digit check never sees.
After: a third-party integration under integrations/ (like minifactscribe)
mechanically judges the legal layer over official open-data artifacts:
verify France — LEGI article XML; the value must really appear in the
official text, French number words included
verify-cn China — FLK official detail JSON; validity, article number,
official bytes unchanged
verify-case China — official case corpus; case number, date, amount and
the receipt chain must match the official copy
verify-it Italy — Normattiva AKN; the value must appear in the AKN text
Two-layer evidence: the ledger points at a derived evidence card, and the
gate recomputes the hash of the official bytes the card declares. Tampering
with the official XML is caught even though upstream stays green — the two
layers do not overlap.
Exit codes follow the frozen contract (0 green / 1 red / 2 config / 3 yellow).
stdlib only, no network in tests, no eval; factscribe/ is untouched and the
package is not added to `packages`.
Deliberately not attempted: automatic judgement-retrieval (each jurisdiction
publishes differently — France ships data files, China ships a reading UI
whose agreement forbids bulk pulls), and Chinese legal-text extraction (a
different class of problem, not an evidence-chain gate).
Tests: 50 offline cases, each red code with both a triggering and a
non-triggering project.
… what it does not judge Records the measured readings (France 0 red / 1 yellow, China 0/0, China case 0/0, Italy 0/0), the three disciplines the implementation is built on (dimension ownership by filename suffix; a 3.10 TOML fallback that must read byte-identically to stdlib tomllib; official artifacts identified by official metadata rather than by our own filenames), and what each gate deliberately does not judge.
The integration is a third-party package under integrations/, so the root `discover -s tests` does not pick it up. Run it in the main matrix instead of a side job on purpose: its fact ledger is read with stdlib tomllib on 3.11+ and with a bundled fallback parser on 3.10, and the matrix (3.10-3.14) is exactly the regression line for that.
Under [Unreleased] / Added, with the new codes and the boundary that the core package and its exit-code contract are unchanged.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
probe