geometry: exact covariant native Möbius comparison

[erinepshovel-code]

Completed result

Exact derived native-frame comparison, with no change to the existing native quotient or motion law:

q(p,e) = p + (1-e)/2 mod 2

D(a,b) = q(b) - q(a) mod 2

The comparison reconstructs the complete target, is invariant under common native motion and quotient-representative changes, composes with the correct seam carry, and negates under chart-orientation reversal. This is an opt-in derived geometric module, not selection of a full UCNS carrier or a protocol/security claim.

Exact delivery and file plan

Base: 5a042416ef62abd9b674523284c2a31193918d66.
Head: ee3df862112811711b43afed4174592a241c373d.
Tree: cdeaa2621fa06f2eff652f6703fe99ba385db949.

Four additive files only:

• src/ucns/mobius_comparison.py: immutable exact comparison, transport, inverse, composition; no IO.
• tests/test_mobius_comparison.py: six accountable witnesses and negative controls.
• docs/native-mobius-comparison.md: general algebraic proof, covariance scope, domain restraint, usage and rollback.
• docs/work-graphs/native-mobius-comparison-inputs.json: exact baseline authority graph.

Native source, CANON, root facade and existing artifacts remain unchanged.

Completed validation

• Full geometry-ci run 36214001174: successful on Python 3.10 / 3.11 / 3.12, including locked dependencies, skill contract closure, source suite, clean package build, distribution verification, and separate clean wheel/sdist replay.
• CodeQL Python and Actions analysis, run 36213999358: successful; PR CodeQL reports no new alerts.
• Downloaded Python 3.11 artifact 10896448912 independently checked against archive SHA-256 eee2e54a2ba37341e84c6bb2737d1fcc58dd445890096a1632ee45a5ef2695da.
• Its exact commit/tree, distribution hashes and archived source bytes match this head. Both clean wheel and sdist XML reports contain 233 tests, zero failures/errors/skips.
• Complete eighth-turn grid: 8,448 common-motion checks, 0 corrected changes, 2,688 naive-product changes; all 4,096 triples satisfy composition/carry.
• Exact non-grid rationals, equivalent representatives, reflection, input rejection, 360-degree frame reversal and 720-degree complete return tested.
• Prepublication contract audit found and corrected a missing deferred-annotation declaration before the delivered commit. Seven targeted local native/comparison tests also passed.

No independent second-agent review is claimed. The automatic Codex reviewer reported its exhausted quota; verification used direct exact-source inspection, full hosted gates and downloaded artifact inspection.

Consumer evidence

Stack #57 consumes this exact implementation without updating its global UCNS pin or changing URPCS v1. All 2,737 sealed observations were admitted. Separate hosted and local enumeration agree on 24,112 native-motion checks with zero corrected changes, while the original 120-check / 44-naive-change falsification remains intact. Consumer receipt SHA-256: c758837cc8e9fdd7e618f6bc008a46cab77ff644408db57fe8cd35daa1b92fab.

Authority and usage

Input graph SHA-256: bd1059322ef8223d2a6531f84591e76970db67aa3982f1cbfd174d5039b5ecd5. UCNS owns geometry, Stack its codec experiment, METAPAT domain restraint, and skill-lib build/evidence doctrine. The baseline graph and actual producer output pin are explicitly distinguished; no proof, semantic, measurement, certification or empirical standing transfers.

Use from ucns.mobius_comparison import compare_native_mobius only on already aligned native states; see the new documentation. Rollback removes these four additive files and any explicit downstream pin.

hmmm

Independent-origin alignment, arbitrary nonlinear/gauge changes, total winding/path holonomy, protocol synchronization, higher geometry and downstream utility remain outside the completed comparison law.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.