Date: 2026-10-27. Implementation author: Codex; exact underlying model identity not recorded. This is implementation verification, not independent research validation. No research experiment was executed by this assignment.
ADR-0006 precedes the implementation. The initial index contains 35 pinned source documents, three tracks and eight canonical hypotheses, with dated historical dispositions, supplementary evidence, reproduction pointers, explicit uncertainty, governance and ten open research/discrepancy/limitation entries. Documents and accepted outcomes remain in their existing locations. No protected ISA, language, VM, Axion, CanonFS, frozen fixture or historical charter/register was edited.
The Python CLI implements all six requested read operations using immutable Git objects and strict JSON requests. MCP was evaluated; the ADR chooses an offline JSON CLI equivalent to avoid a new server dependency/lifecycle before contract review. There is no MCP transport, server registration, write API, experiment execution, external API client, auto-fetch, merge operation or governance activation. Git 2.46+ is required; lazy fetching, optional locks and prompts are disabled.
Self-contained JSON Schemas cover discovery manifest, requests, response envelopes, experiment proposals, independent-review drafts, reproduction-report drafts and evidence-discrepancy drafts. Four illustrative contribution records exercise the schemas. The fresh-agent walkthrough discovers canonical H04, retrieves supplementary model negatives and scope limitations, and prepares an unexecuted, unauthorized experiment proposal with matched adaptive binary control. This walkthrough is not represented as an actual independent-agent review.
The manifest’s main baseline is
274421fd8e85c5b9bb4f165fce190715530f4e85 (#607). The original active checkout
was #605’s experiment branch at 4778e94fec498dfd431977c982dc2c1374966869;
implementation used a separate worktree based on main, preserving the original
branch and its untracked build-bench directory.
827d2accbf8ccb56afe3fe8d96dfb7251f249d7d into the
Investigation09 proposal branch, not main. Its scoped E1 negative, ledgers,
hash manifest, entry contract and reproduction README resolve at that commit.18bd6c78b1ec2c78a500a8e290cb004a062f513e. Its five
audit artifacts remain reported/disputed evidence. A literal comma-split scan
of source-like CSV paths found 46 absent candidates at the inspected main
baseline, including src/policy/axion_policy.cpp and lang/parser.cpp.
This is a pointer check, not a full audit of every capability. Reported 408-test
passes have not been independently established here. Encoding/API and ISA
version wording need reconciliation; 2-bit versus 8-bit raw packing is 4x,
whereas log2(3) describes entropy per trit.e56d1cb5..., but actual pinned
committed bytes hash to bb073baf.... Both are indexed and the discrepancy
remains open. No protected artifact was repaired to force validation.These are surfaced disagreements, not amendments to accepted research outcomes.
Environment: macOS 26.5.2, ARM64; Python 3.14.3; Git 2.53.0. Temporary venv:
/private/tmp/t81-commons-venv. Installed packages: jsonschema 4.23.0,
attrs 26.1.0, jsonschema-specifications 2025.9.1, referencing 0.37.0,
rpds-py 2026.9.1. Only the direct schema dependency is pinned in requirements;
these are the exact local resolved versions, not a cross-platform lockfile.
Executed from the implementation worktree:
/private/tmp/t81-commons-venv/bin/python -B tools/research_commons/commons.py --validate
/private/tmp/t81-commons-venv/bin/python -B -m unittest discover -s tests/research_commons -v
python3 scripts/governance/check_root_structure.py
python3 scripts/ci/check_tisc_freeze_integrity.py
git diff --check
Results: 35 source identities resolve; 16 tests pass; root-structure check passes (81 entries); freeze-integrity check passes (209 registered opcodes); whitespace check passes. The freeze checker uses its own historical v1.1.0 label; that output does not resolve the documentation’s version inconsistency.
Tests cover all published source paths at their revisions, canonical verbatim H01–H08 identities, negatives, no unsupported status promotion, strict requests, missing references/files/commits, hash/URL/path tampering, duplicate IDs/keys, proposal requirements, all contribution examples, working-tree substitution resistance and deterministic responses. All six operations run with socket construction blocked and Git calls limited to version/ls-tree/cat-file; full worktree hashes, Git HEAD/refs/status and index hash remain equal before and after. This is evidence about tested paths, not a security sandbox against arbitrary modifications to the implementation or a malicious host. Git metadata/configuration is trusted.
The demonstration receipt retains actual implementation-
author API request/response hashes for all six operations plus proposal-shape
validation. A test recomputes those hashes. Canonical manifest SHA-256:
81ea5ba5e39d47ffe69bb92e6137c3b0170c176eb8874ac8153a9d5ddd4b7b69.
No receipt claims experiment execution, genuine reviewer independence or
scientific acceptance.
Reviewed Linux, macOS, Windows and ARM64 implications: Git object reads avoid checkout newline conversion; repository paths use explicit POSIX syntax while host paths use pathlib; subprocess receives argument arrays without shell syntax; no architecture-specific native T81 dependency is imported. JSON responses omit absolute host paths and clocks. Git regular-file mode checks reject symlink, directory and submodule retrieval. Python syntax uses the project’s 3.9 minimum.
A dedicated CI workflow targets Linux x86-64, Linux ARM64, macOS ARM64 and Windows, with Python 3.9/3.13. Those platforms/versions are not locally executed here; CI results remain pending. No C++ configure/build/full CTest run, old experiment rerun, external link crawl, live security certification or benchmark replication is claimed. No native code changed, so the narrow verification slice is Python contract/schema checks plus existing governance/freeze checks.
The index is curated, not exhaustive. A valid source reference proves available bytes, not that every nested citation in a report resolves. Scientific adequacy, authentic human authorization, matched-control quality and reviewer independence need human review and independent work. Historical register dispositions remain dated. New evidence/index updates require review; missing #605 objects require explicit setup outside the read-only interface. llms.txt was omitted because it would duplicate AGENTS and this discovery guide. A future MCP wrapper can reuse the contract after its scope and dependencies receive review.
Dependency remains open. GitHub metadata checked during this update reports
PR610 open/draft, merged: false, head
d1eee548d4658df37b35d0cc3781b5a77cbae96b. Its API-generated potential merge
SHA is not treated as an actual merged revision. No correction from PR610 is
claimed as part of main, and PR609 remains draft. Completing the corrected-main
snapshot requires human review/merge of PR610, verification of its actual merge
commit, a subsequent pin/history/receipt update, and rerun validation. Commons
never performs that merge or fetches evidence on demand.
The original #609 work at 7183ae0de2a6d069fa1c422af335294909736eff is
preserved. Merged main at fabc7aa90f70ead227ec257a1ae6833fa16ced40
(PR608) was merged into the same branch without conflict. The manifest now pins
that actual main baseline and contains 39 source identities. PR607’s original
initiative at 274421fd8e85c5b9bb4f165fce190715530f4e85 remains available as
initiative; PR608 is initiative-608; PR610’s actual head objects are available
only under explicit *-610-proposed IDs with role proposal and UNMERGED
uncertainty. The PR608 reconciliation report and PR610 append-only proposal report
are also separately indexed. E1 remains pinned to its separately merged
827d2accbf8ccb56afe3fe8d96dfb7251f249d7d history.
The index retains three discrepancy-history records, including superseded PR607
ID/wording assertions with the limits of PR608’s reconciliation. PR608’s remaining
H07/H08 discrepancy stays unresolved because its correction is only proposed.
Eleven current questions remain. get_hypothesis exposes the history and pending
integration state using the existing operation. The six-operation interface,
request protocol, execution prohibition and offline Git-only reader are retained.
Canonical H01–H08 records are returned verbatim. The register SHA-256 remains
af8ca65dcef136f294eb33010b5a6621875a52db4f3e60b94a4e28db5318a1c1.
H07 retains B/C evidence, no new independent security certification, the original
confidence/next action and 2026-10-25 disposition. H08 retains C/E evidence,
Researcher intent authoritative; comparative benefit untested., unavailable
earlier prompt provenance and the prohibition on treating repeated model
summaries as independent verification. No charter/register bytes were edited.
Same macOS ARM64 / Python 3.14.3 / Git 2.53.0 environment as the initial report. Explicit operator setup outside the interface fetched main and the PR610 branch; CI setup explicitly fetches the pinned PR610 object in addition to the existing PR605 object. Installing or fetching is never a discovery operation.
git fetch origin main docs/pr608-evidence-correction
git merge --no-edit origin/main
/private/tmp/t81-commons-venv/bin/python -B tools/research_commons/commons.py --validate
/private/tmp/t81-commons-venv/bin/python -B -m unittest discover -s tests/research_commons -v
/private/tmp/t81-commons-venv/bin/python -B tools/research_commons/commons.py --validate-contribution docs/research/commons/examples/h08-proposal.json
python3 scripts/ci/check_tisc_freeze_integrity.py
python3 scripts/governance/check_root_structure.py
python3 scripts/governance/check_overclaim_guardrails.py
python3 scripts/governance/check_active_docs_link_hygiene.py
python3 scripts/governance/check_determinism_claims.py
python3 scripts/governance/check_docs_structure.py
git diff --check
a3ee2d288ac7b00176beae32f854c373194b11556d83293396bbcd7c02d688bf.demo-observation.initial.json, checked against
the original commit. The active receipt was regenerated from actual responses.not_requested human
authorization and execution: unavailable. The fresh client finds H08 via
Track M, checks its actual register evidence, retrieves the unmerged correction
as a proposal, and prepares/validates a scratch proposal. Controls include
human-only, single-assistant and adversarial-assistant arms, matched budgets,
blind independent adjudicators and a held-out confirmatory corpus. This is
fresh-process client verification, not independent-model scientific review.book/book-{en,cn,es,pt,ru} directories. This is unchanged from main
and was also documented in PR610; unrelated subsystem/checker repair is excluded.The following observations are for original commit 7183ae0d, not a claim about
this new update’s remote tests. Commons run 40558170198
reports six successful jobs: Linux x86-64, Linux ARM64 and Windows, each on Python
3.9 and 3.13. Both macOS-14 jobs are queued; their environment and results are
unverified. No Commons failure was reported at this checkpoint.
Foundation run 40558169976 has successful governance-metrics and spec/docs lint jobs, skipped optional experimental jobs, and remaining queued jobs. Determinism run 40558170231 has Linux/macOS builds in progress; CodeQL run 40558169931 is in progress. Format run 40558169951 passed. Pending or skipped jobs establish no new correctness or portability result. Results for the pushed update must be assessed separately.
External clients can inspect canonical H08 without adopting superseded summary assertions, as the fresh-process test demonstrates. Full integration of the corrected initiative on main remains blocked by unmerged PR610. Neither the proposed correction nor successful implementation tests establish H08 comparative methodology benefit or authorize experiment execution or governance activation.
PR610 is still open/draft at d1eee548d4658df37b35d0cc3781b5a77cbae96b;
merged: false. PR609’s reviewed starting head is
998de6402e0f2f48cd318ee772cd600178206a46. No source pin, canonical record,
manifest identity, active demonstration or proposed-correction status changes
in this preparation. Runtime/interface code is unchanged.
The latest Commons run 40560202300
is for 998de640. Its Windows Python 3.9 job
failed only test_recorded_demonstration_matches_pinned_responses: the checked-out
historical JSON receipt used CRLF, whereas git show returned the original LF
blob. Source validation succeeded (39 sources, unchanged manifest digest); the
other 18 tests passed. Runner: Windows Server 2025, Git 2.56.0.windows.2,
Python 3.9.13. This is a test portability regression introduced by the receipt
preservation assertion, not a demonstrated evidence-identity or runtime failure.
The correction checks the preserved receipt’s committed Git bytes against the original committed bytes, and separately compares the checkout’s decoded text with universal newline handling. Exact artifact identity is retained. A scratch CRLF fixture now passes, while changed committed bytes and changed checkout content are both rejected. No JSON receipt or hash was normalized or rewritten; no runtime semantics or platform-wide Git configuration was changed.
At inspection, Windows Python 3.13 was in progress; all six Linux x86-64, Linux ARM64 and macOS jobs were queued. No result is inferred for those jobs. Other current-head workflows (Foundation CI, determinism, CodeQL and format) were queued, with optional experimental jobs skipped. These observations supersede neither earlier-head successes nor the need to validate the next head. A corrected-head Windows pass has not yet been established locally.
Commands rerun from the Commons worktree, using the previously recorded local environment (macOS ARM64, Python 3.14.3, Git 2.53.0):
/private/tmp/t81-commons-venv/bin/python -B tools/research_commons/commons.py --validate
/private/tmp/t81-commons-venv/bin/python -B -m unittest discover -s tests/research_commons -v
python3 scripts/ci/check_tisc_freeze_integrity.py
python3 scripts/governance/check_root_structure.py
python3 scripts/governance/check_overclaim_guardrails.py
python3 scripts/governance/check_active_docs_link_hygiene.py
python3 scripts/governance/check_determinism_claims.py
python3 scripts/governance/check_docs_structure.py
git diff --check
Results: 20/20 Commons tests pass after the correction, including the CRLF
fixture, unchanged deterministic response hashes, canonical H01–H08 fidelity,
negative-result discovery, pinned identities, fresh-process H08 onboarding and
read-only/no-network checks. All 39 sources validate with manifest digest
a3ee2d288ac7b00176beae32f854c373194b11556d83293396bbcd7c02d688bf.
Freeze registry, root structure, overclaim, link hygiene, determinism-boundary
and whitespace checks pass. Docs structure still fails on the five known absent
book/book-* paths; this pre-existing failure is unchanged and not repaired here.
No Windows execution is claimed from the local CRLF fixture. Passing these
implementation checks is not independent scientific replication or evidence of
H08 methodology benefit.
| Development record | Immutable commit | Sources | Canonical manifest SHA-256 |
|---|---|---|---|
| Initial discovery baseline | 7183ae0de2a6d069fa1c422af335294909736eff |
35 | 81ea5ba5e39d47ffe69bb92e6137c3b0170c176eb8874ac8153a9d5ddd4b7b69 |
| PR608 / proposed PR610 expansion | 998de6402e0f2f48cd318ee772cd600178206a46 |
39 | a3ee2d288ac7b00176beae32f854c373194b11556d83293396bbcd7c02d688bf |
Both manifest versions were read directly from their Git objects and their source
counts/digests checked during this preparation. demo-observation.initial.json
retains the initial receipt, and the 39-source receipt remains unchanged at
998de640. Future snapshots must preserve these commits and explicitly identify
which snapshot produced each receipt. Main evidence is still pinned to PR608
fabc7aa90f70ead227ec257a1ae6833fa16ced40, not the unmerged PR610 head.
awaiting_pr610_merge and proposed-source roles. A reviewed
integrated state must describe verified merged evidence without promoting the
canonical hypotheses or relabeling the old proposal snapshot as accepted fact.
Do not weaken the missing-reference, exact-identity or read-only checks.Release readiness remains conditional on PR610’s actual merge, the reviewed snapshot refresh, corrected-head required validation and human acceptance. This preparation adds no MCP transport, website, execution capability or write path; it does not modify PR610 or the canonical register, or authorize an automatic merge.
Follow-up CI inspection before publishing this correction: the Windows Python
3.13 job (122386839630) also failed the same CRLF-versus-LF receipt assertion,
with the other 18 tests passing. Linux x86-64 Python 3.13 was in progress; the
remaining five matrix jobs were queued. Both Windows failures therefore have
one directly evidenced cause addressed by this test-only fix; remote validation
of the fixed head remains outstanding.