Repository navigation
builder: factor a deep decision by nesting depth; lengthen a prefix an earlier app of the set holds - #19
Merged
Conversation
…lengthen a prefix an earlier app of the set holds The property proofs' repeat' split gives up once ifs nest past a dozen (simp's step limit, measured on the locked SDK's Lean: 13 split, 14 do not). Factoring now counts nesting depth, not rule count: a branch drops the conditions its field fixes, a rule left with none answers the rest of its chain, and a branch still too deep is factored again. DR-17 checks a tall table against the flat first-match chain on every request. Two apps of one set with the same initials no longer refuse the set: the later one is lengthened from its last word, as for a prefix another suite holds (DR-22). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd
afflom
pushed a commit
to afflom/me
that referenced
this pull request
Nov 14, 2026
…h their full model Every keyword facet now goes through the prompt set: tools/prompts/gen.py keeps only a tree path's is-tests (first-match makes the is-not tests redundant), and the re-vendored builder (afflom/prism-devrel#19, #21) factors a deep table until no path nests past twelve ifs or ten rule tests and takes sixteen fields. The old generators skip the Lean of every listed facet; consumers use the builder's names. Only chronology and kappa still wait (BS-01). 68 modules verify in the locked SDK; 255 Rust tests pass. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013Asa3QU8Mz6i7JNPDbNRBd
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Factoring by depth.
builder/machine.mjsnow factors a decision by how deep itsifs nest, not by how many rules it has.MAX_DEPTH = 12replacesFACTOR_AT = 11.factorFieldis kept.machineDeclarationsalso returnsstepNames, and every theorem unfolds those names.Id prefixes in a set. In
builder/set.mjs, when two apps of one set share initials (afflom/me'sprism-holonandprism-host), the build used to refuse the whole set. Now the first app keeps the prefix and the later one is lengthened from its last word. This is the same rule already used for a prefix another suite holds.Why
The
linear_arithmeticproof unfolds the whole decision into one goal and runsrepeat' split. Lean'ssplitstops onceifs nest past about a dozen, because it hits simp's step limit.I measured this on the locked SDK's Lean: 13 nested
ifs split and 14 do not. A table factored by a 5-valued field still nests 5 dispatch tests plus the chain under them. That is why afflom/me's resonance decision failed with LLV7002 on all 16 theorems.With this change:
verified 14 modules);Checked
node builder/test.mjspass.MAX_DEPTH, and that the factored judge answers the same as the flat first-match chain for every request.PCL-01andPCLI-01).Generated by Claude Code