Skip to content

Prose matches evidence: ALS-T8 / T16 / S3 rewritten to their fixtures (reference-evaluator findings F7–F9) - #46

Merged
O6lvl4 merged 1 commit into
mainfrom
prose-vs-evidence-f7-f9
Aug 21, 2026
Merged

Prose matches evidence: ALS-T8 / T16 / S3 rewritten to their fixtures (reference-evaluator findings F7–F9)#46
O6lvl4 merged 1 commit into
mainfrom
prose-vs-evidence-f7-f9

Conversation

@O6lvl4

@O6lvl4 O6lvl4 commented Aug 21, 2026

Copy link
Copy Markdown
Contributor

Closes #43, closes #44, closes #45 — the three places the reference evaluator's second pass (docs/ref/PARSER-NOTES.md F7–F9) found normative prose contradicting the executable evidence it cites. The fixtures win; the text now says what they pin.

  • ALS-T8int.parse trims the ALS-T1 Unicode White_Space set first; int.from_hex has its real grammar (trim → lowercase 0x stripped repeatedly → sign → hex digits; 0X10 and -0xff are invalid digits, 0x alone is empty). C-028 retitled to match (its statement was already right).
  • ALS-T16 — a table: unsigned COUNTS saturate (take(-1) = whole, drop(-1) = empty), the repeat family is signed (negative → empty), string.slice start/end signed, list INDICES unsigned (OOB → no-op/default/append). The old 負→0 sentence is gone.
  • ALS-S3 — per-predicate table: is_alpha/is_alphanumeric full-Unicode and nonempty, is_digit ASCII-only and nonempty, is_upper/is_lower Python-style (≥1 cased, all cased hold), is_whitespace vacuously true on "". The one-size vacuous-truth sentence is gone. C-100's "ASCII alphanumeric" claim corrected (the probe matrix shows Unicode; its own fixture only has ASCII rows).
  • Each section has a judged example (doctest 182/192 on 0.58.0; the 10 red are the known judge-ahead rules) and a validation row in proofs/als-validation.toml — reviewed by the author against the named fixtures and a both-target probe run, independent = no — ceiling 127 → 124 (a tightening, riding along as the ratchet-separation gate allows).
  • Ratchet-separation ledger row: first real range (the other session's Reference evaluator, stdlib round 1: 217 functions from the chapters — 412/610 evaluated, 409 agree, abstains 434 to 198; exact float text; findings F7-F10 #42) passed — green direction observed.

Author / verifier record

role who independent?
authored Claude (this session), on behalf of @O6lvl4
verified the gates + the same session; the reference evaluator (other session) implements the same fixture-pinned behaviour no

🤖 Generated with Claude Code

…ains int.parse's Unicode trim and int.from_hex's real grammar, ALS-T16 distinguishes unsigned counts, signed repeat and slice, unsigned indices, ALS-S3 is a per-predicate table with the empty-string value per row; C-028 retitled, C-100's ASCII claim corrected; three validation rows (author review, not independent), ceiling 127 to 124; ratchet gate row notes its first real green range
@O6lvl4
O6lvl4 merged commit f1470d5 into main Aug 21, 2026
1 check passed
@O6lvl4
O6lvl4 deleted the prose-vs-evidence-f7-f9 branch August 21, 2026 01:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant