ae38412e93
The widened 100-pack tour surfaced four packs whose closings bucket
included a bare '.' entry: assertive_v1, blunt_v1, dry_v1, and
mathematical_v1. Seeded variation selects from buckets uniformly, so
roughly 1/N of turns would emit a '..'-style surface artefact
(e.g. 'Truth is what is true..') when '.' was selected and appended
to a surface CORE's deterministic realizer already terminated with
'.'.
This commit:
* Strips the bare '.' entry from the four affected packs' closings.
The remaining entries (e.g. ' — qed.', ' — that's it.', '') all
carry their own leading separator, so concatenation produces a
clean surface.
* Re-runs scripts/ratify_register_packs.py to re-seal each pack's
mastery_report.json with a fresh pack_source_sha256 and
self_seal_sha256 — these companion files are part of the diff
because pack content changed.
* Adds _validate_discourse_markers_shape() to
scripts/ratify_register_packs.py as defense in depth. The new
R4-tier gate refuses any closing entry whose first character is
bare sentence-final punctuation ('.', '!', '?', ';', ':') or an
alphanumeric — both classes would produce concatenation artefacts.
Ellipsis ('...', '…') is explicitly exempt as a legitimate
stylistic ending.
Verified:
* core eval cognition byte-identical (the eval runs unregistered)
* tests/test_cognition_eval_register_matrix.py: 801 passed in 5:23
(100 registers × 8 projection invariants + meta-test), confirming
the closing fix did not perturb ADR-0072 trace_hash invariance
or any per-case projection.
* Hand-verification across {assertive,blunt,dry,mathematical}_v1
× 4 prompts = 16 surfaces: 0 remaining double-period artefacts.
* Re-running ratify against a re-introduced bad closing produces
SystemExit with the new gate's diagnostic message.
mathematical_v1 now reads beautifully under "What is light?":
"Lemma: Light is visible medium that reveals truth. — qed."
27 lines
633 B
JSON
27 lines
633 B
JSON
{
|
|
"register_id": "blunt_v1",
|
|
"version": "1.0.0",
|
|
"description": "Blunt register — short, sharp, no qualification.",
|
|
"schema_version": "1.0.0",
|
|
"mastery_report_sha256": "0c435910a295c7cade443cccfd175c90f0302cc90f0c5a6c613ef62c190462ed",
|
|
"display_name": "Blunt",
|
|
"depth_preference": "terse",
|
|
"realizer_overrides": {
|
|
"disclosure_domain_count": 1,
|
|
"compress_gloss": true,
|
|
"drop_articles": true,
|
|
"drop_provenance_tag": true
|
|
},
|
|
"discourse_markers": {
|
|
"openings": [
|
|
"",
|
|
"Look — ",
|
|
"Plainly: ",
|
|
"Said: "
|
|
],
|
|
"transitions": [],
|
|
"closings": [
|
|
""
|
|
]
|
|
}
|
|
}
|