Skip to content

Claude/2026 sota benchmark xyp1zh - #23

Merged
jesusvilela merged 3 commits into
mainfrom
claude/2026-sota-benchmark-xyp1zh
Jul 6, 2026
Merged

jesusvilela merged 3 commits into
mainfrom
claude/2026-sota-benchmark-xyp1zh

Conversation

@jesusvilela

Copy link
Copy Markdown
Owner

No description provided.

claude added 3 commits July 6, 2026 11:45
…w ask)

The one recommendation named in three consecutive reviews: automate the tests.
.github/workflows/tests.yml runs pytest on push (main + claude/**) and every PR,
on Python 3.11 and 3.12. External solvers are installed best-effort; the
@requires_kissat integration tests skip gracefully when absent, so CI is green at
494 passed / 27 skipped with no solver and runs the full set when kissat is
present. README carries the status badge.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gkw2ccXyb4DEtqFo29ouQD
…e collapsing frame)

Self-resolve AND meta-resolve contradiction, discharged. The blocker was
diagnosed honestly: naive Tseitin gate-encoding flattens frame structure (aux
gate vars hide parity/counting patterns), so the frames never fire -- epsilon in
the flesh. The contract: a structure-preserving encoder.

- backend/lambda_sat.py: BXor is now a first-class Boolean-lambda constructor.
  When the beta-normal form is a parity system (conjunction of XOR-over-literals),
  it is encoded FRAME-NATIVE -- parity clause groups over the original variables,
  NO auxiliary gates -- so the GF(2) frame recognizes it and gf2_xor_solve decides
  it in poly time, certified. Result carries resolved_by in {parity,2sat,counting,
  direct}: the frame that collapsed the term. Non-parity terms gate-encode, try
  the three sound frames, else a re-verified brute-force decision.
- Because the parity encoding has no aux vars, its GF(2) UNSAT is exactly the
  refutation whose soundness is stated in proofs/XorSoundness.lean.
- 6 new tests incl. a 200-instance soundness stress (every parity verdict matches
  brute-force source truth); the reduction-reveals-parity case shows the frame is
  discovered AFTER beta-reduction, not assumed. 527 tests green.
- docs/ladder/LAMBDA_SAT_NOTE.md: meta-resolution recorded; counting-frame
  recognizer + the epsilon>0 living remainder named as the remaining contracts.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gkw2ccXyb4DEtqFo29ouQD
…ving remainder (ε>0)

The operator's fuller vision (a Y combinator "in a Calabi-Yau Möbius torus"): its
honest load-bearing core is the fixpoint, and full Y is undecidable (no certificate
can exist -- the halting boundary), so "maybe that's too much" is precisely true.
Its decidable spine IS the ε>0 contract.

- backend/lambda_sat.py: unroll_fix(λs.body, k) + fixpoint_sat(...). Unroll a
  Boolean fixpoint to depth k with the un-reached tail as a free variable -- the
  remainder made literal. remainder_active (ε>0) = the depth-k truth still depends
  on the tail (fixpoint not pinned down); False (ε=0) = stabilized. Bounded model
  checking in miniature; the first place the remainder carries lost truth, not just
  provenance.
- 5 tests: λs.a and λs.a∧(a∨s) (absorption) stabilize (ε=0); λs.a∨s boundary-
  dependent (ε>0); λs.¬s -- the Möbius fixpoint -- eternally ε>0 at every depth
  (self-negation never orients). 532 tests green.
- docs/ladder/LAMBDA_SAT_NOTE.md: the Y-spine recorded; the Calabi-Yau/360/
  curvature envelope kept as the lens (the geometry of the remainder space).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Gkw2ccXyb4DEtqFo29ouQD
@jesusvilela
jesusvilela merged commit fc10d75 into main Jul 6, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants