Version 1.0 pilot observation ledger

[owieschon]

Purpose

This issue is the canonical product-evaluation ledger for the 30-day pilot window. It records required-check continuity, false positives, and repair time across two public repositories.

Current candidate segment

• Candidate: v1.0.0rc11
• Commit: f6ce8bd1c45810156aa627d4312c8bc7d080d553
• Common window start: 2026-07-13T15:57:37Z
• Earliest 30-day close: 2026-08-12T15:57:37Z
• Required check context: clean-docs / clean-docs
• Scheduled check: daily at 05:17 UTC, plus pull requests and pushes to main

The clock starts at the later default-branch receipt after both pilots adopted the fixed candidate and branch protection was reverified.

Pilot baseline

Repository	Adoption	Main receipt	Accepted existing findings	New or stale findings	Required check
ultra-csm	PR 160, merge 6983e98ddfefa8f5c1cb66c929979f95cafa058f	run 29264400532	32	0	configured
agent-governance-lab	PR 11, merge 616d3d9712ce6b8e209517d15321d6417078c884	run 29263925782	1	0	configured

Ultra CSM has 32 exact, reasoned exceptions for near-duplicate reference material and runtime prompt or evaluation source assets. Agent Governance Lab retains one exact exception for a protected operating-rules file. Both repositories have zero active findings and zero stale baseline entries.

Measurement rules

• A false positive is a clean-docs gate failure where source-bound documentation is current and the repository violates no documented clean-docs rule.
• A true positive is a gate failure repaired by changing a bound document, its source, its explicit baseline, or its reasoned ignore record.
• Repair time runs from the first failing required-check completion to the first passing run on the repaired commit.
• Infrastructure failures outside clean-docs are recorded separately and do not count as documentation false positives.
• Candidate or pilot-configuration changes start a new documented segment; prior observations remain part of the evaluation dataset.

Current metrics

Metric	ultra-csm	agent-governance-lab
Observed false positives	0	0
Observed true positives	0	0
Repair events	0	0
Unexplained required-check failures	0	0

Pre-window dogfood findings

These defects were found and repaired before the observation clock. They do not count toward the window metrics.

1. Mature repositories needed an exact, explicit baseline for existing documentation debt.
2. Valid directory links were rejected by one projection path.
3. Bootstrap initially moved live configuration documents instead of preserving the corpus.
4. Repository summaries were too large for a README.
5. Repository-overview evidence was hashed before canonical ordering, which made a correct bootstrap PR fail nondeterministically.
6. The reusable Action assumed the host repository had a root Python dependency file.
7. Action artifacts did not bind the installed candidate, source commit, run identity, or raw evidence digests.
8. The generated product overview duplicated the release label outside its source-bound evidence.
9. The writing standard described a specific voice, but the runtime prompt and audit path enforced only a narrow prose subset. The project README exposed the gap.
10. The prerelease build imported an optional stable-release dependency at module load, which broke the release-artifact job in a minimal build environment.
11. The stable reader-trial rubric omitted the README purpose-identification task required by the Version 1.0 E2E.
12. Python inventory evidence hashed CPython's version-specific AST serialization, so Python 3.12 and 3.14 produced different repository-overview digests for identical source.
13. Repository-wide catalog coverage was reported as direct protection, decorative lead blocks could become purpose contracts, canonical agent indexes omitted declared pages, and documented CLI examples were not parser-verified. RC10 separates direct binding from catalog coverage, skips decorative lead blocks, supports declared canonical index pages, and parses every published CLI example in tests.
14. Mature-monorepo initialization needed bounded planning, nested package and MCP registration discovery, conservative document classification, and an explicit gap instead of a manufactured purpose contract. RC11 adds those constraints and a Reter-shaped regression fixture.

Evaluation completion

The evaluation window completes after 30 observed days if both repositories retain the required check, all scheduled and change-triggered runs are accounted for, and every false positive and repair event has a receipt. This window gathers operating data and does not block the stable release. Stable release readiness is evaluated separately through the four isolated model trials and the Version 1.0 product DoD.

[owieschon]

2026-07-13 acceptance audit: main commit 84465624f0eba474210d66778df952cb9cb1c969 corrects the documented child-process isolation boundary and strengthens all six V1 E2Es against the published rubric. Public CI run 29240448412 passes. No product runtime, reusable Action, release artifact, or pilot configuration changed, so RC5 remains the candidate and the observation clock remains 2026-07-13T09:34:47Z through 2026-08-12T09:34:47Z.

[owieschon]

2026-07-13 clock correction: the original start used the later merge timestamp, but ultra-csm run 29239646422 completed at 2026-07-13T09:35:14Z. The clock now starts at 2026-07-13T09:51:53Z, the first recorded inspection after both default-branch runs passed and both branch-protection records were verified to require clean-docs / clean-docs. Earliest close is 2026-08-12T09:51:53Z. Candidate RC5 and both pilot configurations are unchanged.

[owieschon]

2026-07-13 post-start heartbeat: manual default-branch runs passed on the unchanged RC5 pin. ultra-csm run 29240734694 reports 0 new findings, 201 accepted findings, 0 stale baseline entries, and 2 of 2 bindings/projections current. agent-governance-lab run 29240734734 reports 0 new findings, 9 accepted findings, 0 stale baseline entries, and 2 of 2 bindings/projections current. Both evidence artifacts were downloaded and inspected. Metrics remain 0 false positives, 0 true positives, 0 repairs, and 0 unexplained failures.

[owieschon]

2026-07-13 candidate change: inspection of the first RC5 pilot artifacts found that raw audit and binding reports did not identify the installed clean-docs candidate or checked repository commit. The RC5 observation window is superseded because its evidence is not self-contained. RC6 at 7f204ad38ac4c0668786c7cf7989b49fe435bb2e adds clean-docs.action-run.v1, hashes every accompanying report, rejects mutable package refs, and removes a duplicated release label that self-dogfood proved could become stale. CI run 29241466531 and release run 29241625116 pass. Wheel SHA-256 is 8d74fe28ab5c24d13899c7535c24286ec7795cb7c4b962ffab79bad53d8968c8; SPDX SHA-256 is 43e2edd32ac86aea43a1390db60c06ec4a50670c290de0c4435a3585be86f0cf; both attestations verify. The next 30-day window starts only after both pilot default branches pass RC6 and branch protection is reverified.

[owieschon]

2026-07-13 RC6 window start: both replacement pilot PRs are merged and both protected default branches pass the exact RC6 pin. ultra-csm PR 155 merged at b93cf2302a1d1859511b6f765a3f1aa27014cbff; run 29242146709 reports 0 new, 201 accepted, 0 stale, and 2 of 2 current. agent-governance-lab PR 6 merged at de075fcc4bb1b07003c5fb5df6ceeef3d5376d7a; run 29241838506 reports 0 new, 9 accepted, 0 stale, and 2 of 2 current. Each clean-docs.action-run.v1 receipt identifies RC6 and its source commit; every embedded raw-evidence digest was independently verified. Branch protection still requires clean-docs / clean-docs in both repositories and retains eval + CSM scorecard in ultra-csm. Common window: 2026-07-13T10:16:40Z through earliest close 2026-08-12T10:16:40Z.

[owieschon]

2026-07-13 RC7 window start: both pilot PRs are merged and both protected default branches pass the exact RC7 pin. ultra-csm PR 156 merged at 06b923742c5bdcbbe7716aa6a187137081b2f9b9; run 29246005695 reports 0 new, 310 accepted, 0 stale, and 2 of 2 current. agent-governance-lab PR 7 merged at 7d5811b63e2ee5c5278439a33303da24d1952630; run 29245662959 reports 0 new, 10 accepted, 0 stale, and 2 of 2 current. Each clean-docs.action-run.v1 receipt identifies RC7 and its source commit; every raw-evidence digest was independently verified. Branch protection still requires clean-docs / clean-docs in both repositories and retains eval + CSM scorecard in ultra-csm. Common window: 2026-07-13T11:22:32Z through earliest close 2026-08-12T11:22:32Z.

[owieschon]

2026-07-13 RC9 window start: RC8 was superseded before its window because the first pilot PR runs exposed Python-version-dependent repository-overview hashes. The same Git tree produced different digests under Python 3.12 and 3.14 because inventory evidence used ast.dump; RC9 hashes the exact source spans selected by the static AST walk and direct 3.12/3.14 replay now matches. RC9 also retains the fifth independent-reader task for the README purpose contract. ultra-csm PR 157 merged at 79211a1b0a633ae9963ed055b4c3911fe8b421a3; run 29247968687 reports 0 new, 310 accepted, 0 stale, and 2 of 2 current. agent-governance-lab PR 8 merged at 42a70221fe1a918498a7b99480dc8a7d19b8dcb4; run 29247481712 reports 0 new, 10 accepted, 0 stale, and 2 of 2 current. Both receipts identify exact RC9 commit 80440f944a913814cb67a1b1d8f613bd56e0ad0e; every raw-evidence digest verifies. Branch protection still requires clean-docs / clean-docs in both repositories and retains eval + CSM scorecard in ultra-csm. Common window: 2026-07-13T11:56:23Z through earliest close 2026-08-12T11:56:23Z.

[owieschon]

Independent-reader coordination now lives in issue #2: #2. It fixes RC9, its wheel digest, rubric, and four allowed context documents; separates human and fresh-agent independence boundaries; and defines the evidence files required for the stable receipt. No participant is assigned yet. The completed internal rehearsal is explicitly excluded from both required rows.

[owieschon]

2026-07-13 pilot-configuration anomaly and repair: the documentation-corpus repair added the optional projections.llms_txt.include field, but both pilot workflows still installed RC9 commit 80440f944a913814cb67a1b1d8f613bd56e0ad0e. Ultra run 29254779088 and Agent Governance Lab run 29254349753 rejected the new manifest before audit with projections.llms_txt has unknown key(s): include; their empty JSON follow-on errors were consequences of that schema rejection. Classification: pilot configuration failure, not a documentation false positive. The repair first pinned verified implementation commit 9d6749501b762454069d61b9d8f47b288b8fcd40; Ultra run 29255547746 passed 10m57s after the failed run completed, and Agent Governance Lab run 29255546966 passed 15m36s after its failed run completed. The product repair is now published as attested prerelease v1.0.0rc10 at 67a5d0bdc281a305166aa186f6847d622cd60000; release run 29256216381 passed, wheel SHA-256 is 8b9739f3e5350f66825d68b64080ea8d10fc2fac7c3fb20a80f2573f42851b11, SPDX and wheel checksums pass, and both attestations verify. Both pilots are being repinned to the published commit. The common 30-day clock restarts only after both protected default branches pass that exact pin.

[owieschon]

2026-07-13 RC10 window start: both published-candidate repins are merged and both protected default branches pass the exact RC10 commit. Ultra CSM PR 159 merged at a3294753c81bdf0947c1ff25c95d1a29e4d273bb; run 29256896291 completed at 2026-07-13T14:11:59Z with audit and bindings true. Agent Governance Lab PR 10 merged at 6a7043d7274cad7ea34459034e143838211a5ab4; run 29256459802 completed earlier with audit and bindings true. Both clean-docs.action-run.v1 receipts identify version 1.0.0rc10 and package commit 67a5d0bdc281a305166aa186f6847d622cd60000; every listed raw-evidence byte count and SHA-256 was independently verified. Branch protection still requires clean-docs / clean-docs in both repositories and also retains eval + CSM scorecard in Ultra. Common window: 2026-07-13T14:11:59Z through earliest close 2026-08-12T14:11:59Z.

[owieschon]

2026-07-13 RC11 candidate change and window start: dogfood against owieschon/reter exposed that deterministic initialization on a mature monorepo could omit nested package manifests, miss registerTool MCP registrations, overproduce content-plan facts, classify ambiguous operational documents as process artifacts, and manufacture weak purpose contracts. Classification: pre-window product coverage gap found by expanded dogfood, not a pilot false positive. Clean-docs PR 5 repairs those gaps and adds a Reter-shaped regression fixture; public CI run 29259325450 and the complete Version 1.0 DoD pass. Attested prerelease v1.0.0rc11 is pinned at f6ce8bd1c45810156aa627d4312c8bc7d080d553; release run 29263583527 passed and wheel SHA-256 is 984d9eb859bab62dc749de9c22f95b2ce7946fb69fd8037ebe12b50fa27a7fb3. Ultra CSM PR 160 merged at 6983e98ddfefa8f5c1cb66c929979f95cafa058f; default-branch run 29264400532 passed at 2026-07-13T15:57:37Z with 0 findings, 32 baselined findings, 0 stale entries, and 6 of 6 bindings/projections current. The wider inventory added its nested UI package, five package scripts, and test runner to docs/SURFACE.md. Agent Governance Lab PR 11 merged at 616d3d9712ce6b8e209517d15321d6417078c884; default-branch run 29263925782 passed earlier with 0 findings, 1 baselined finding, 0 stale entries, and 5 of 5 bindings/projections current. Both receipts identify rc11 and the exact package commit; every evidence byte count and SHA-256 verifies. Branch protection still requires clean-docs / clean-docs in both repositories and retains eval + CSM scorecard in Ultra. Common window: 2026-07-13T15:57:37Z through earliest close 2026-08-12T15:57:37Z.

[owieschon]

Observation ledger reopened

This issue tracks the non-blocking 30-day pilot evaluation window. It was closed prematurely.

The 1.0.0rc19 candidate does not repin either pilot and does not reset the current observation segment. Both pilot repositories remain bound to their existing candidate until a pilot-specific repair requires a coordinated repin. The release decision is independent of this data-gathering window.

[owieschon]

Stable 1.0.0 published

Stable 1.0.0 is published from commit b6c6f88885272f4e1a17feca02beed82a2ddeaf4 after 20/20 independent reader tasks passed.

This does not repin either pilot and does not reset the current 30-day observation segment. The pilot window remains a non-blocking data-gathering activity.