Draft: INFOS BAS — held source review and offline inventory
Held INFOS source proposal
Adds one unreleased infos_bas registry entry, a hash-pinned offline archive
inventory preparer, an exhaustive candidate review-queue option, source contracts
and rights/provenance evidence. No canonical shard or corpus admission is proposed.
Acquired evidence
54 original publisher PDFs for consecutive issues 288-341 were acquired
through ordinary browser publication-page PDF links: 216 pages / 8,038,738
bytes. All parse as four-page PDFs and their first-page issue numbers match.
Literal article CC BY 3.0 PL notices were detected in all 54 pinned PDFs.infos_bas.pdf-evidence.json records each source/PDF URL, SHA256, notice location
and local download timestamp. Raw PDFs and exploratory extraction remain private
under ignored logs/bas-input/infos/; full text is not published by this PR.
Earlier exploratory URLs were used as leads only; the quarantined prototype's
text and licence labels were not reused. 2026 items were found through official
publication/news pages. Direct HTTP still returned Human Verification and the
archive year AJAX failed, but normal browser detail pages and PDF downloads
worked. No CAPTCHA solving, cookie export or impersonation was used.
Consecutive PDF coverage is not a certified complete archive/year inventory.
Rights decision: held
The source is published by Kancelaria Sejmu through the official Sejm BAS/BEOS
archive, with individual authors credited. Article notices name CC BY 3.0 PL,
but literal notice detection is not a completed manual per-item/third-party review.
The complete main-content text of the SIS copyright/reuse page was read and
retained in infos_bas.publisher-terms.txt, with file and normalized-text hashes.
This is a browser visible-DOM text transcription, not raw HTTP bytes.
The page contains a restriction on materials forming part of a commercial offer.
Its scope/priority against explicit article CC grants is unresolved and requires
maintainer/legal review. Other applicable publisher terms have not been
exhaustively reviewed. Registry remains NOASSERTION, decision
held_for_review, release: None.
Review preparation and validation
The offline archive preparer verifies locally saved HTML hashes/provenance,
rejects malformed/empty input and duplicate publisher IDs, and retains pre-cutoff
records as withheld. Post-cutoff selection does not assign licences.--review-plan queues all inventoried candidates chronologically, pins its
inventory/manifest and leaves reviewer labels, third-party spans and counts null.
A separate private review queue now contains all 54 acquired PDFs, with no labels.
Focused source contracts: 25 passed, 1 skipped (canonical shard unbuilt).
All 54 acquired PDFs separately passed signature, parsing and issue-ID checks.
The latest full-suite run, before evidence-only updates, was 778 passed,
184 failed, 34 errors, 27 skipped under Python 3.14 with a writable basetemp.
The full suite is not green and this is not a main-baseline regression claim.
Before ready review or admission
Reconcile complete archive/year coverage; resolve publisher-term scope and
manually verify per-item rights/quoted material. Review all candidate briefs,
recording classifications, spans, Wilson results and token shares. Validate
two-column extraction, normalization and boilerplate; complete pinned
exact/near/containment overlap against target shards. Apply pinned external
NERGAL, then use build_dynaword.py, reconcile canonical stats/attribution and
include three real canonical inline samples via render_sample.py.
The provisional sibling-paper decision schema is still not validated.
This draft contains evidence and metadata only; required canonical samples do
not yet exist. No source is ready, merged or released by these changes.