t81-foundation

Executed verification — independent integration audit

Executed 2026-10-25 America/New_York on macOS Darwin 25.5.0 ARM64, Apple clang 21.0.0 (clang-2100.1.1.101), CMake 4.3.0, Ninja, Python 3.14.3. Working tree: /private/tmp/t81-h04-r02-integration, based on main 01c85019. Explicit target requirement: C++17. Source/output hashes: evidence/manifest.json.

Commands actually executed

cmake -S . -B /private/tmp/t81-h04-r02-clean -G Ninja -DT81_BUILD_TESTS=ON -DT81_BUILD_TUI=OFF -DT81_BUILD_EXAMPLES=OFF -DT81_BUILD_BENCHMARKS=OFF -DT81_BUILD_FUZZ_TESTS=OFF -DT81_ENABLE_TERNARYOS=OFF

First configure failed (exit 1): sandbox DNS prevented fetching Asio. No tests were counted as passes. Retry used an existing, clean Asio checkout tagged asio-1-30-2 at 12e0ce9e0500bf0f247dbd1ae894272656456079:

cmake -S . -B /private/tmp/t81-h04-r02-clean -G Ninja -DT81_BUILD_TESTS=ON -DT81_BUILD_TUI=OFF -DT81_BUILD_EXAMPLES=OFF -DT81_BUILD_BENCHMARKS=OFF -DT81_BUILD_FUZZ_TESTS=OFF -DT81_ENABLE_TERNARYOS=OFF -DFETCHCONTENT_SOURCE_DIR_ASIO=/private/tmp/t81-accountable-build/_deps/asio-src
cmake --build /private/tmp/t81-h04-r02-clean --target t81_h04_r02_adaptive
/private/tmp/t81-h04-r02-clean/experiments/h04_r02_adaptive_representation/t81_h04_r02_adaptive --json-out experiments/h04_r02_adaptive_representation/results.json
ctest --test-dir /private/tmp/t81-h04-r02-clean -R h04_r02 --output-on-failure --no-tests=error
python3 experiments/h04_r02_adaptive_representation/oracle.py
python3 experiments/h04_r02_adaptive_representation/packing.py
python3 tests/ci/h04_r02_audit_regression_test.py --bin-path /private/tmp/t81-h04-r02-clean/experiments/h04_r02_adaptive_representation/t81_h04_r02_adaptive -v
python3 scripts/ci/run_research_regressions.py --build-dir /private/tmp/t81-h04-r02-research-clean --asio-source /private/tmp/t81-accountable-build/_deps/asio-src --jobs 4

Sanitizer configure/build/tests used a second fresh build directory:

cmake -S . -B /private/tmp/t81-h04-r02-sanitized -G Ninja -DT81_BUILD_TESTS=ON -DT81_BUILD_TUI=OFF -DT81_BUILD_EXAMPLES=OFF -DT81_BUILD_BENCHMARKS=OFF -DT81_BUILD_FUZZ_TESTS=OFF -DT81_ENABLE_TERNARYOS=OFF -DT81_ENABLE_ASAN=ON -DT81_ENABLE_UBSAN=ON -DFETCHCONTENT_SOURCE_DIR_ASIO=/private/tmp/t81-accountable-build/_deps/asio-src
cmake --build /private/tmp/t81-h04-r02-sanitized --target t81_h04_r02_adaptive
ctest --test-dir /private/tmp/t81-h04-r02-sanitized -R h04_r02 --output-on-failure --no-tests=error

Recovered R01 was extracted from git archive def47c68 docs/research/continuity into /private/tmp/t81-h04-r01-audit-source; the deterministic gzip archive is preserved in evidence. Its executed commands:

python3 /private/tmp/t81-h04-r01-audit-source/docs/research/continuity/probes/h04_radix_adaptation.py --output-dir /private/tmp/t81-h04-r01-audit-run
python3 /private/tmp/t81-h04-r01-audit-source/docs/research/continuity/probes/verify_h04_radix_adaptation.py --packet /private/tmp/t81-h04-r01-audit-run
python3 -m unittest discover -s /private/tmp/t81-h04-r01-audit-source/docs/research/continuity/probes -p test_h04_radix_adaptation.py -v
python3 scripts/ci/check_tisc_freeze_integrity.py

C++ files were formatted with repository clang-format configuration, then the normal H04 build/CTest was rerun. Those final logs are retained alongside the initial passing and sanitizer runs. This formatting change did not alter output.

Outcomes and limits

Check Actually observed
Clean H04 CTest 3/3 pass; nested Python audit 8/8 pass
ASan + UBSan H04 CTest 3/3 pass, no sanitizer finding
Standalone Python audit 8/8 pass
Exhaustive checks within audit All 64 length-three traces over four boundary classes at each of 2 transition costs; exact rational optima and crossover
Block/comparator sweeps within audit 36 configurations, shared-mode feasibility, metadata invariance and actual emitted costs
Packing All 729 standalone six-trit values; all 1024 signed 10-bit values; seeded streams of length 0..150; canonicality rejection
Bounded research regression script 8/8 pass (separate suite, no bypass)
Recovered R01 producer/verifier Pass: 96 operations, 384 arms, 192 disabled equalities, 1596 sweep points
Recovered R01 tests 19/19 pass; generated files match pinned reference bytes
Frozen TISC check Pass: 209 opcodes, bitwise mappings and reserved range

No executed test failed. The initial configure failed before tests because the dependency could not be downloaded; the offline retry resolved it. Initial H04 build warned about C++17 extensions; explicit target features removed those warnings. The wider research build reports duplicate-library linker warnings for libt81_core.a; no test failure resulted and production link code was not changed. Graphviz dot was unavailable to Doxygen discovery, but not required by this build. The full repository test inventory was not built or run.

Portability review: changed C++ uses standard C++17 facilities, no platform intrinsics, ABI assumptions or nonportable wide integers. CMake passes the exact target path (including Windows .exe); Python uses subprocess argument arrays, standard-library modules, temporary paths, and explicit little-endian framing. Linux/Windows execution and x86 timings were not performed. Remote CI snapshots are evidence of their reported states, not local cross-platform verification. Main and the existing PRs were not uniformly green at inspection.

Reexecuting R01’s own independent verifier does not independently prove every R01 assumption. There are no new execution/energy/bandwidth measurements here.

Original integration output SHA-256 (head 13f8377f; correction below)

Complete logs and source/snapshot/output hashes are in evidence/. To reproduce elsewhere, replace the local Asio source path with a clean asio-1-30-2 checkout or allow CMake to fetch that dependency. Use fresh build directories.

PR #572 precision correction

Fresh clean and ASan/UBSan builds each pass all 3 H04 CTests; the standalone Python audit passes 9/9, including 49 new shared-overhead scale configurations. The first Python audit run correctly failed artifact parity after one crossover changed at its final floating-point bits. The executable regenerated that field; all final checks pass. Full commands, initial/final logs, reproducer and CI snapshot details: PRECISION_CORRECTION.md.

Current results.json SHA-256: 5ea3ed12b16d94a9cdc86fccf808091da614d1b4f4ac6edc45e2330c7987e6e9. The original evidence manifest is preserved as history/manifest-13f8377f.json; the current manifest records this correction.

This site is open source. Improve this page.