Repository navigation
Construct a shorter G20 word: 245 fewer ADDs per call with exact endpoints - #5807
Merged
Mister-Meeseeks merged 1 commit intoOct 27, 2026
Merged
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The actual PR5708 G20 word now uses 245 fewer ADDs per local call, with the same complete endpoint and every native/owned boundary value. Its 18-call context can remove 4,410 ADDs and 8,820 primitive-count terms. The local positive majorant falls from 3,189,524 to 2,653,582. The recursive paid/target profile is unchanged; no new kappa is claimed.
The construction applies the exact PR5793 synthesis tool to 191 actual windows over three rounds. A dynamic program chooses the best disjoint window cover in each pass. Every two-, three- and four-role endpoint has a certified minimum word; a final complete pass has no strict reduction. Neutral rewrites, larger windows and changed native frames remain outside this fixed point.
The package includes the new WORD, exact EDITS, complete receipt, source pins, unchanged tool bytes, proof and controls. All 685 arbitrary input axes, 48 dirty owned axes, native frames, target boundaries and owned clears are retained. Forward and reverse full-word interpreters agree; a literal truth-table check independently validates each rewrite. Registered normal/optimized checks and read-only Git-free replay pass. Source and semantic corruption controls reject. Independent scientific acceptance is pending. Hosted status is recorded below.
The proof permits substitution in the retained 18-call context. This PR emits the local word; global index/hash descriptors and a fresh compiler receipt must be rebuilt before source admission. The source native/ROW, ownership, precision, compiler and analytic assumptions remain. No global circuit optimum is claimed.
Package:
research/g20-four-role-resynthesis. Replay:python3.12 -B scripts/research_package.py --package research/g20-four-role-resynthesis.Canonical claim. Credit Dugongue, ai-nanachi, PR5679/5708, the PR5793 exact synthesis tool and all retained source lineage. Source head
829d01fc031f7323ffcfa423fdfd1f24dcbe3190. Self-contained frozen bytes avoid any dependency on an unmerged Git object.Base:
de5ec2e7a6074068452f45b1a191b5c72832df4a. Verified head:9f80a2dfc160351e49ea82d3b3d9354d7329ead8.Prepared for Ali Romman with substantial OpenAI Codex assistance. Retained tool and mathematical predecessor credits apply. No selected-frontier change.
Hosted status at this exact head: the planning job fails in the unchanged
test_research_current_assignment.pysuite (eight failures and two errors); scientific replay is skipped. Exact logs were checked. Existing PR5761 contains the shared test repair. The complete local and sealed-export checks above are the scientific evidence supplied here; hosted scientific success is not claimed.Follow-through: PR5834 installs the released local word in all 18 physical calls and rebuilds every global query, lifetime and cost record. That integration is a separate verification scope and does not add a second local-savings claim here.