Evidence from a fresh clone of the public repository, running exactly the commands an IEEE artifact reviewer is given. Every exit code and timing below was observed; nothing is estimated.
| Date (UTC) | 2026-09-09T05:09:08Z |
| Commit | b4d7c67c1c17825889fba436fd31c81fade5b21d |
| Source | https://github.com/Sukhmangill977/gc-ir-reference.git (public clone, no local state) |
| Tags present after clone | preregister-tier0-v2, preregister-tier0-v2.1, preregister-tier0-v2.2, v1.0.0 |
| OS | Darwin 25.6.0 (macOS 26.6.2) |
| Architecture | arm64 |
| Python | CPython 3.11.15 |
| git | 2.50.1 (Apple Git-155) |
| Docker | 29.6.2, build dfc4efb |
The clone was made into an empty temporary directory. No file was copied from the development checkout.
| # | Command | Exit | Time | Outcome |
|---|---|---|---|---|
| 1 | git clone … |
0 | 59 s | 4 tags fetched, HEAD b4d7c67 |
| 2 | make install PYTHON=python3.11 |
0 | 15 s | .venv created, pinned dependencies installed |
| 3 | make test |
0 | 13 s | 301 passed |
| 4 | make verify-hashes |
0 | <1 s | Both reference hashes MATCH |
| 5 | python tools/freeze_check.py --final-v2 |
0 | 4 s | FREEZE CHECK PASSED — 24 checks, 0 failures |
| 6 | python tools/verify_reported_results.py |
0 | <1 s | 48/48 paper-facing claims verified, 22 cross-checks passed |
| 7 | make ieee-check |
0 | 16 s | ALL 8 STAGES PASSED |
Total from clone to full verification: under two minutes.
make verify-hashes recompiled both cases from the committed governance
artifacts and reproduced:
case_a f5cbc3a8016159d2074005fa021bf76ca04fb7aed1408f5ec845418460a43536 MATCH
case_b 2850155a2ee7d4c01db4748a891891bae27c48f4c9c2c7eccd30beb7d1aa33ce MATCH
freeze_check.py --final-v2 verified the Case B → L-DREA mapping against the
external artifact of record: 9 rows, {exact: 4, family: 3, not_established: 2},
0 failures — including the two gaps, which are declared rather than hidden.
verify_reported_results.py confirmed, among the 22 cross-checks, that
SNR and DF remain status=DEFERRED, value=null for both cases.
| # | Command | Exit | Time |
|---|---|---|---|
| 8 | docker build -t gcir . |
0 | 20 s |
| 9 | docker run --rm gcir make reproduce |
0 | 47 s |
Container environment, reported by the container itself:
Linux-6.12.76-linuxkit-aarch64-with-glibc2.36
aarch64
CPython 3.11.11
All eleven reproduction steps completed:
OK 1. Regenerate case artifacts and verify the committed tree 0.35 s
OK 2. Compile Case A and Case B 0.26 s
OK 3. Test suites (unit, property-based, adversarial, integration) 12.59 s
OK 4. Adversarial corpus, structural checks, validation seeds 4.74 s
OK 5. Traceability audit queries and negative controls 0.46 s
OK 6. Primary metrics 0.10 s
OK 7. Gate divergence and Proposition 1/2 premises 0.69 s
OK 8. Determinism experiment (31 runs per case) 5.37 s
OK 9. Monte Carlo rating robustness 22.53 s
OK 10. Hash manifest 0.01 s
OK 11. SUMMARY.md 0.01 s
Values produced inside the container, from a different OS, libc and architecture than the host:
case_a f5cbc3a8016159d2074005fa021bf76ca04fb7aed1408f5ec845418460a43536
case_b 2850155a2ee7d4c01db4748a891891bae27c48f4c9c2c7eccd30beb7d1aa33ce
TD 1.0 62/62
Byte-identical to the committed reference hashes and to the host run. This is a
second architecture (aarch64 Linux, glibc 2.36) reproducing the same canonical
payload hashes as macOS/arm64 — consistent with, and independent of, the eight
environments recorded in results/final_v2/CI_STATUS.md.
The first clean-clone attempt used the machine’s default python3, which was
3.9.6. make install failed with:
ERROR: Could not find a version that satisfies the requirement cffi==2.1.1
ERROR: No matching distribution found for cffi==2.1.1
make: *** [install] Error 1
That message never mentions the Python version, so a reviewer would have had no
way to diagnose it. A check-python guard was added: make install now stops
immediately with the interpreter it found and how to override it. Verified in
both directions — exit 2 with a clear message on 3.9.6, exit 0 on 3.11.15. The
run recorded above is from the clone after that fix.
This is the reason for running the clean-clone test rather than assuming it would work.
Establishes. The artifact installs, tests, compiles, verifies its frozen evidence and reproduces its reference hashes from a fresh public clone, on both a macOS/arm64 host and an aarch64 Linux container, with no hidden local state and no manual intervention.
Does not establish. Platform independence — this is two environments out of
the eight tested, and a finite matrix is not the set of all environments. Nor does
it establish anything about the content of the results; it establishes that they
regenerate. What the numbers mean, and what they do not, is in
PAPER_TO_ARTIFACT_RESULTS.md.