Skip to content

Reduce R438 to 429 helpers with three source snapshots - #3645

Open
RossRichardson wants to merge 1 commit into
CrocSwap:mainfrom
RossRichardson:research/bit-r438-source-code
Open

RossRichardson wants to merge 1 commit into
CrocSwap:mainfrom
RossRichardson:research/bit-r438-source-code

Conversation

@RossRichardson

@RossRichardson RossRichardson commented Oct 18, 2026 •

Copy link
Copy Markdown

Summary

This source-pinned local BIT word replaces nine rank-7 copied centres with three rank-1 source snapshots, then deletes nine physical helper families that are never read. The complete local rank mass falls from 5,121 to 4,989, ADDs from 24,248 to 13,223, and arbitrary dirty helpers from 438 to 429, with one owned private field. The complete profile change is -9 e7-24 e1-9 e5.

The 132-unit reduction is local only. Under incumbent width-8 whole-family accounting, returned idle-owner children add 72 rank units per local call at beta=0, cancelling the extra 72-unit pruning reduction. This is not a net global 132-unit saving.

The changed F2 endpoint is (x,y,h) -> (Ax,y+Ax,h) for an explicit invertible 111-by-111 code. Three repaired actual target rows are minimal for the fixed clean-centre J and invertible source-recoding class. The signed lift restores helpers exactly over the integers; target correctness is asserted modulo 2.

Scope and limits

This is a finite local construction, written proof and separate exact replay, not a new kappa or global supplier admission. Source-code composition, terminal decoding at actual caps, rounded precision, physical owners/access/tape and joint outer transfer remain open. Native evidence is the exact rational cap ADDRESS/copied-total calculation and complete local frame/rank invoice, not a full physical compiler theorem. The selected record is unchanged.

Only research/bit-r438-source-code/ is added. Its two live inputs stay at canonical 20ecf2891d80b5acf18d980b2164fb44ea7faa90. The registered replay regenerates the complete ordered word and compares decompressed semantics, checking source-path and same-run parent-file provenance separately. This avoids using zlib container identity as regenerated mathematical evidence.

Test plan

  • Complete 651-column integer/F2 forward and inverse operators, all 429 dirty helpers and exact roundtrip; literal dead-family projection, source inverse and actual R/K/J section checked by a separate interpreter without producer imports.
  • All 111 rational cap projectors, 222 frame bindings and 3,552 ADDRESS columns; every literal MOVE/ADD/COPY/ERASE frame and immutable snapshot boundary in both directions.
  • Complete profile and scalar invoice; eleven construction-ancestry and seven independent defect controls, including a wrong integer sign with unchanged parity.
  • Isolated optimized Python 3.12 replay with warnings as errors and private child bytecode namespaces; package metadata and source pins.
  • Exact commit 99b901c9b5ac70cf441b3d7e585cc7b8744b5359: committed-byte/source-history preflight for all 23 package/input files, with only ancestral 20ec source history; actual read-only Git-free optimized export replay passed in 14.12 seconds with every byte unchanged. VALIDATION.json preserves the earlier precommit measurements.
  • Automatic exact-head hosted CI 39128556864 passed for 99b901c9b5ac70cf441b3d7e585cc7b8744b5359; no manual rerun.

Sources and credit

Credit Ali Romman's R438/boundary-persistence work, Sagar Thakrar's transport/rehosting, the retained DreamingOfClouds/eumemic/PR589/PR555/PR543/PR545/PR510 lineage, Ali Romman's prior source-code/unit-transfer motif in PR1862, and Hanson Sun/bbMisk's restored mutable-source-query work in PR1522. No priority is claimed for generic source coding or source pivots.

Ross Richardson, with substantial OpenAI Codex assistance in construction, proof, separate audit and packaging; Apache-2.0. Separate AI implementations are not independent human review or formal-kernel verification. SOURCES.json, ADAPTATIONS.json, PROOF.md and the registered verifier give exact provenance, scope and reproducible evidence.

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.

1 participant