Compose integer labels with checked XOR scatter export

[hexapode]

Change and mathematical scope

PR3273's XOR boundary still accepts PR3100's fractional-label witness: changing lines[0][1] from 0 to 0.5 passes the symbolic checker but changes exported geometry. PR3100's integer-label guard is composed only with the older simple-scatter contract. This adds integer-xor-scatter-v1: both existing contracts, full symbolic checking before export, and the same private input snapshot throughout. Selected baseline and cancelling-pair exports remain byte-identical to the original exporter.

This is an opt-in correctness repair, not a new bug discovery, supplier admission or kappa improvement. Frozen sources are unchanged.

Task and overlap

Refined supplier-boundary-audit; base 6f100d4, priority revision 6. Exact input blobs and SHA256 pins are in research/integer-xor-scatter-boundary/SOURCES.json; all are available in base history. TASK.json records the scope and limited live overlap search. Nearest work: #2977, #2992, #3100, #3273 and #3326. This composes the repairs instead of repeating their individual audits.

Validation

• python3.12 -B scripts/research_package.py --package research/integer-xor-scatter-boundary
• python3.12 -I -O -Werror -B research/integer-xor-scatter-boundary/verify.py
• python3.12 -I -O -Werror -B research/integer-xor-scatter-boundary/test_boundary.py -v
• python3.12 -B scripts/check_research_commit.py --package research/integer-xor-scatter-boundary --check-source-history

All pass locally on macOS, CPython 3.12.1. The deterministic audit also agrees under 3.11.6. Five behavioral groups cover malformed encodings, corruption of each authenticated delegate, output preservation, optimized CLI behavior and replacing the original input after validation. A Git-free export of exact committed bytes passed the audit and all five groups under -I -O -Werror -B; all 15 export files remained unchanged.

The audit runs the actual old XOR boundary to reproduce both fractional-only and mixed fractional/duplicate acceptance, then verifies rejection by the composed boundary. Wrong integer geometry and a wrong scalar endpoint are rejected by full checking. No fresh native/Boost, physical hosting, full selected ancestry or all-size theorem replay is claimed. Hosted CI is separate.

Credit

Nick Petalas (#2977/#3273), josecarlosgonzalezroque-design (#3100), Arham Tahir (#2992), and the scalar/native/selected-source lineage detailed in README.md retain credit. Substantial OpenAI Codex assistance across implementation, separate automated adversarial review and writing. Apache-2.0; no independent human review claimed.

Open next questions

README.md and TASK.json name the next step: bind a concrete producer/consumer to the combined contract and replay its complete native/physical invoice. This wrapper does not adopt itself into a production supplier.