Skip to content

Certify sharp linear wrapper costs for complete high routes - #2982

Merged
Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
petalas:research/high-route-cost-audit
Oct 16, 2026
Merged

Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
petalas:research/high-route-cost-audit

Conversation

@petalas

@petalas petalas commented Oct 16, 2026

Copy link
Copy Markdown
Contributor

Summary

Add the isolated research/high-route-prefix-cost package, following PR2968's released route-cost handoff.

complete packet traffic = sum of actual cube invoices + wrapper invoice
wrapper invoice <= 265 V
with returned-output disposal <= 277 V

The exact source-level invoice covers canonical prefix/high/fibre counts, cube extraction/appending, pooled acquisitions, clearing and final rewinds. Both constants are sharp within this wrapper's physical read/write/unit-move currency. A single-cube tariff lifts without another prefix factor.

This changes the drawn stock-envelope scope to a released local H2 obligation. It closes only the wrapper component, not the complete all-width cube tariff or a whole-stage stock coefficient. Task, proof and source pins retain the boundary and next questions. Canonical investigation.

Evidence

  • Before, the released handoff had a complete scaling/pool invoice but no explicit full-route wrapper charge. After, the written proof gives the exact invoice and sharp linear bounds.
  • Registered CPython 3.12.15 replay passes 44 complete reduced routes, 125,689 payload bits and 1,314 actual native B1 calls. Raw physical meters agree with a separate I/O observer; affine endpoints, borrowed state and cleanup are checked.
  • Thirteen controls pass normally/optimized, including CPython 3.13.5. An endpoint-preserving C-squared cost mutant fails the invoice. A padded-prefix control establishes why canonical counts are required. Native fixtures use B1's elementary branch, not its fitted selected GL64 branch.
  • Workflow suite passes 158 tests with 11 platform skips. Ruff and committed-byte preflight pass for all 32 package/input files.
python3.12 -Werror -B scripts/research_package.py --package research/high-route-prefix-cost
python3.12 -Werror -B -m unittest discover -s tests -p 'test_research_*.py'
python3 -B scripts/check_research_commit.py --package research/high-route-prefix-cost

Unknowns remain explicit: the uniform single-cube tariff, complete machine-step/allocator/provisioning charges, physical storage and inherited caller/refinement dependencies. No selected cold replay, independent human/formal review, hosted-CI result, supplier admission or new kappa is claimed.

Credit Jose Manuel/jmbarrancoml for PR2915/2899/2809, David Arias for PR2753, Arham Tahir for PR2877's countdown, and the retained PR2157/PR589/source lineage. PR2951/2968's scaling and PR2708's separate storage result are not claimed as new.

Merge danger

Two-way; research-only. Nine new package files and an opt-in bounded replay. No production, frozen-source, selected-result or shared workflow changes. Archival publication would not discharge scientific review.

@Mister-Meeseeks
Mister-Meeseeks merged commit 5d93a57 into CrocSwap:main Oct 16, 2026
1 check passed
Mister-Meeseeks pushed a commit that referenced this pull request Oct 17, 2026
#2915's ordered high-field ADD

Answers the ordered-ADD component of PR #2982's single-cube handoff for
the unchanged PR #2915 kernels.added and its pooled stream.add_pass. For
odd Q>=3, canonical positive B,M,S and c in {-4,-2,-1,1,2,4}, with
V=BQMQS: exactly 10+[S>1] additional scratch roles suffice and are
attained in every call, sharpening PR #2809's conservative 32-role
lexical allowance; six additional numerical words suffice and one
fixture attains six (PR #2810's add_pass bound composed with the actual
coefficient callback); and the complete local physical read/write/
unit-move bill, including encoded arithmetic and pool acquisitions, is
at most 4096V, or 4102V with returned-output disposal. The exact
non-arithmetic invoice extends PR #2877's and PR #2473's countdown and
payload bills, zero case included, with pool acquisitions, the
one-bit-suffix path and the marginal lifecycles. It matches the
unchanged source on 109 complete packets (37,241 payload bits), and one
fewer scratch role or numerical slot rejects at the real allocators.
Route setup, B1 swaps, machine steps, physical backing and whole-stage
stock remain open; no supplier, selected-frontier or kappa change is
claimed.

Adds research/high-add-scratch-invoice/ only.

Submitted by @commdata2338 with substantial AI assistance (OpenAI Codex
research and implementation; packaged with Claude Code).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG
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