Skip to content

builder: cap a branch at ten rule tests; factor number tables by how a pair stands - #21

Merged
afflom merged 2 commits into
mainfrom
claude/chain-length
Nov 15, 2026
Merged

afflom merged 2 commits into
mainfrom
claude/chain-length

Conversation

@afflom

@afflom afflom commented Nov 14, 2026 •

Copy link
Copy Markdown
Owner

What

  • MAX_CHAIN = 10. A branch, or the whole table, is factored once it holds more than ten rule tests. The overall nesting cap, MAX_DEPTH = 12, still counts the dispatch tests.
  • Factoring by a pair of number fields. Some tables are too deep and have no choice field to factor on. Those are now factored by the pair of number fields that shortens them most, with one branch per standing: x =? y (one test), then x <? y, then the rest (two tests each). Inside a standing, comparisons of that pair that hold there are dropped, and rules that cannot hold there are left out.

Why

Checked

  • All 14 rows of node builder/test.mjs pass.
  • DR-17 also builds Allen's relations and checks two things:
    • judge is factored by how astart and bstart stand;
    • it gives the same answer as the flat first-match chain for every request with values 0..4 (625 requests).
  • afflom/me regenerated with this builder: lexlean verify in the locked SDK reports "verified 68 modules". Chronology built on its own also verifies (verified 14 modules).

🤖 Generated with Claude Code

https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd

A rule tested by its misses is a disjunction, which costs split more than a
dispatch atom: google-link's 12-rule chain of compound misses stopped proving
once factoring counted only nesting depth (#19). A chain now also factors past
MAX_CHAIN = 10 rule tests, as before #19, while dispatch still counts toward
MAX_DEPTH. afflom/me's 68 modules verify with it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd
…tands

A deep table no choice field can factor (afflom/me's chronology, Allen's
interval relations over four numbers) is now factored by the pair of number
fields that shortens it most: equal, below, above. Inside each standing the
comparisons of that pair that hold are dropped and rules that cannot hold are
left out. DR-17 checks Allen's relations against the flat first-match chain on
every request; chronology's proofs verify in the locked SDK.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd
@afflom afflom changed the title builder: a branch keeps at most ten rule tests builder: cap a branch at ten rule tests; factor number tables by how a pair stands Nov 15, 2026
@afflom
afflom merged commit f8c0300 into main Nov 15, 2026
4 checks passed
afflom pushed a commit to afflom/me that referenced this pull request Nov 15, 2026
The re-vendored builder (afflom/prism-devrel#21, merge f8c0300) factors a deep
table of number fields by how a pair of them stands (equal, below, above), so
chronology's 14 Allen-relation rules prove. It joins the prompt set; only kappa
still waits (BS-01). 68 modules verify in the locked SDK; 263 Rust tests pass.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd
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