Skip to content

Prove the symbolic agreement core of the two-target compact switch - #7738

Open
petalas wants to merge 2 commits into
CrocSwap:mainfrom
petalas:research/w4-1118c-raw-agreement-cover
Open

petalas wants to merge 2 commits into
CrocSwap:mainfrom
petalas:research/w4-1118c-raw-agreement-cover

Conversation

@petalas

@petalas petalas commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Scoped algebraic proof with finite corroboration. For the two-target compact-switch word with constant source bits (1,1), binary control radix B and window modulus N>B, the maximal invariant agreement core is

u0 < B-1,  t0 < B-1,  t1 < B-1,  W0 + t0 < N

A carry calculation proves necessity and sufficiency on both points of each ideal XOR orbit. The core excludes any four-entry independent translation menu at B>=4. Five entries are optimal when B>=8 and 5(B-2)<=N, with an invariant selector.

This contribution was produced by Nick Petalas. It answers the symbolic-core question in #7197 and credits its finite core and four-entry adversary. zhaobenny's #6812 supplies the raw word and five-phase upper construction. Advance claim

Evidence

  • The direct checker compares 984,064 packed inputs with the formula across five new radix and width pairs. It finds 534,112 invariant agreements. Three grids range over complete dirty fields and two use stated upper-field representatives.
  • All 557,056 relevant coordinate tuples in the two checked cover geometries admit a phase. Removing the last phase misses 72 tuples at (B,N)=(8,64) and 168 at (16,128). Seven controls include an agreement whose ideal mate fails and a case whose error wraps at N=B.
  • The warning-strict normal and optimized commands take 1.56 s and 1.52 s locally and produce identical hash-bound reports. Registered replay and committed-byte/source-history preflight pass for all six package files. The selected baseline passes its metadata bindings.

Run from the repository root. Local environment was CPython 3.12.15 on macOS 27.0.1 arm64 with the standard library.

python3.12 -I -W error -B research/compact-two-target-symbolic-core/verify.py
python3.12 -I -O -W error -B research/compact-two-target-symbolic-core/verify.py
python3.12 -B scripts/research_package.py --package research/compact-two-target-symbolic-core
python3.12 -B scripts/check_research_commit.py --package research/compact-two-target-symbolic-core --check-source-history

Limitations

The proof fixes the two target positions, the all-one source pair, two-digit control fields and the live-digit word. The menu bound requires independent coordinate translations and an invariant selector. Movement, physical selection and cleanup, tape, precision, hosting and the complete recursive bill remain open. The prior finite census, selected scientific cold replay, Lean and Docker were outside this run. This result supplies no new κ. Independent human and formal proof-kernel review are unclaimed.

Open next questions

  • Derive the invariant core for a constant source pair with one zero bit. Begin by calculating the net lower-window and temporary changes when the low parity stays fixed.
  • Find the minimum translation menu for B>=4 outside the five-phase packing range. Begin with the four restored coordinates at B=4, N=8, where the lower bound still applies.

Related open work

  • #7197 proves the core and five-entry minimum at B=8, N=32 by finite methods. Substituting those parameters in the symbolic formula gives the same predicate. The predecessor census was not replayed.

CI notes

This PR carries the shared CI-doc and replay-gap sync used by other open PRs.

All three hosted checks passed on 3d6b5586a43240d5d166912b4873a2009b281de7, including planning and the symbolic-core and historical replay-gap package replays. CI run 44394137014 The six status tests passed locally on CPython 3.12.15. The historical replay-gap check returned PASS_REGISTERED_RESEARCH_PACKAGE.

Before the sync, the base's test_current_ci_section_matches_selected_record failed locally because the selected record names make frontier7000-verify and the CI guide still names make frontier6030-verify. The same assertion stopped the onboarding plan job on #7734, with one failure and 24 skips in 760 tests. Hosted log

Merge danger

Two-way change.

Derive the invariant agreement set across binary radices and window
widths by tracking the temporary and source carries. Extend the
five-translation minimum to the stated packing range.
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