# Remaining ledger actionability audit

Date: 2026-08-02

Audited repository tip: `139c19d99a2d96237cee1d0bc1a759ac3df7becf`

## Outcome

All 149 entries from the prior row-level audit were re-read individually against
the repository and qualified host. Four V5-001 rows were locally actionable and
are now closed by source-bound UE 5.5.4 editor and Linux-cook runs. The 145
remaining unchecked entries cannot be truthfully completed by another local-only
implementation or fixture:

- 8 entries in section 0.3 are standing evidence-pack requirements, not global
  work-package completion claims.
- 125 entries across 36 task packages require external authority or state.
- 12 entries are the portfolio completion gates and depend on the preceding
  external work.

The 125 external-state entries break down as follows:

| Family    | Entries | Why they remain open                                                                                                                          |
| --------- | ------: | --------------------------------------------------------------------------------------------------------------------------------------------- |
| EVAL      |      16 | Licensed human-gold corpora, independent ratings, preregistration, and calibrated human comparisons are absent.                               |
| OPS       |      11 | Authorized live topology, a second deployed failure domain, remote backup/restore evidence, production streams, and measured SLOs are absent. |
| REL       |      18 | The evaluator contracts exist, but current deployed R0–R8, autonomy, and recertification decisions do not.                                    |
| V1–V9     |      69 | Remaining product claims require real providers/content, consented users, human review, canaries, comparisons, or production operations.      |
| P1        |       5 | A deployed all-pipeline drill matrix and authoritative current portfolio/card catalog are absent.                                             |
| P2        |       6 | Adoption requires measured production corpora, controlled comparisons, staged experiments, or an actual distilled model.                      |
| **Total** | **125** |                                                                                                                                               |

“External-state” includes composite tasks that still need implementation in an
authorized deployment, such as a second failure domain. It does not mean that a
mock configuration would close the task; the checkbox also requires exercised
failover or equivalent production evidence.

## Requirement-by-requirement disposition

`evidence/v1-v9/remaining-actionability-manifest.json` is the machine-readable
record of the review (`oshun.remaining-actionability.v1`). Every unchecked row
has:

- its authoritative ledger line, exact multiline text, and individual SHA-256;
- its task package, disposition, and explicit blocker-profile reference;
- a blocker profile that names the still-required external evidence, the
  authority needed to obtain it, and the reason local artifacts are
  insufficient;
- SHA-256-bound supporting evidence that documents the implemented local
  controls and the remaining honest boundary.

The manifest contains 145 distinct row records and 38 used blocker profiles: one
for the standing rules, one for the portfolio gates, and one for each of the 36
externally blocked task packages. There are no implicit, default, or unused
profiles. Its published SHA-256 is
`86d7580af6cfd247293013d01da73e2d2b1cd4b57c7ba783c1c7c595708fb756`.

## Repository audit findings

The audit traced unchecked claims into their admission boundaries rather than
accepting their prose at face value. It found and closed two repository-local V8
loopholes before reaching this disposition:

- generated manifests now require the exact authoritative G1–G8 gate set in both
  TypeScript and Python;
- authenticated release reports now require that same exact set before signing
  and during offline verification.

The refreshed audit also corrected the stale V4 host observation. The local UE
5.5.4 editor and `ueagent` account were available, so source commit
`0950fe9c0df44c816feabaafebd15c9dae9de52a` added the exact
`V4.IntegratedRuntime` target. Its single authenticated editor session passed
adaptive-director bounds, procgen connectivity/placement, replay, and commentary
integration with four successes, no test warnings/errors, and exit code 0. This
closes only the two qualified-runtime rows; it is not provider, human-review,
match-media, or final-polish evidence.

The same qualified host also made the V7-003 UE row locally actionable. Source
commit `4d4ff1a6efc43f033804779fdd752329f5dc9844` passed 12/12 Mawu Automation
tests, a Linux `RunUAT BuildCookRun` as `ueagent`, and runtime placement of the
exact governed three-layer, 23-primitive flagship lock from its cooked archive.
The 125,741,681-byte archive has SHA-256
`a446bb3fc0b83b6812a25d70596f36e7cf4bdda9baef92edc96e5e1b026deb14`. This closes
only cook and placement; human play/review/sign/canary and the three-way
procedural/model/best-authored comparison remain externally blocked.

The V5-001 audit then found four more repository-local rows that could be
qualified on the same host. Source commit
`cb562888f78ab1bba71d2e39b5ec12993f0f7047` passed the exact 14 selected V5
Automation tests with no warning-bearing, failed, or not-run cases. A separate
fresh Linux `RunUAT BuildCookRun` cooked 448 packages and retained a
158,312,903-byte cook tree with SHA-256
`1c9aa45c53b17fada652afe5baced5ef709e5ec3e31f241535e7cb6abc1fe609`. The cook
includes real binary `BootMap`, AI Director policy, and safety-policy packages.
This closes only the repository-local director, cooked-binary,
named-system-runtime, and HMAC/fail-closed-seam rows. Live TTS/recap/STT/
translation and signed human director/cell/continuity/rights/release receipts
remain externally blocked. The cook-only run does not claim a packaged
executable or launch smoke.

The current audit also replaces V2's stale “Unreal is absent” observation with
machine evidence. On 2026-08-02 the qualified host has UE 5.5.4,
`UnrealEditor-Cmd`, `RunUAT.sh`, `ueagent`, and the V2-enabled EOS editor
binary. The retained V2 editor modules are not current, however: project build
ID `43c40849-16a9-4317-b4b9-c7a9a7e8b72d` differs from engine build ID
`9729efc9-d09a-46a2-bf07-036739f64d68`. V2 therefore needs a resource-safe
editor rebuild before qualified automation/cook. The preflight claims no Isis
asset, cook, playback, provider, or human evidence, so no V2 checkbox closes.
The exact host record is `evidence/v1-v9/V2-unreal-host-preflight.json`.

The same audit corrected V3's stale missing-host prose. UE 5.5.4,
`UnrealEditor-Cmd`, `RunUAT.sh`, `ueagent`, the exact V3 project descriptor, and
the enabled EOS editor binary are present. Unlike V2's stale build ID, V3 has no
retained `V3/ue/Binaries/Linux/UnrealEditor.modules`; the shared preflight
therefore reports `rebuild-required` and retains the exact `V3Editor` build
command. The required resource gate found another checkout's Vitest worker using
about 7 GiB RSS on this 15 GiB host, so the build was deferred. No provider,
cook, playback, or flagship checkbox is claimed. The machine record is
`evidence/v1-v9/V3-unreal-host-preflight.json`.

The V2–V8 Metis bridges are also present as governed, authority-minimized
repository seams. Their composite V9-004 checkbox remains open because none of
those adapters is cooked/runtime placement evidence, and the same checkbox still
requires production mobile/offline, localization-scale, embodiment, and payout
proof where launch scope requires them.

Composite requirements were specifically checked for unfinished repository
prerequisites. The local topology, restore, metric, release, autonomy,
recertification, provider-admission, runtime-admission, study-admission,
flagship-readiness, and adoption gates named by their supporting packs are
implemented and fail closed. Their unchecked rows remain composite because the
same rows also require an authorized deployment, real provider or engine output,
licensed or consented data, human judgment, production observations, or release
authority. In particular:

- the V5 authored-case manifest honestly contains zero admitted cases;
- V6 and V7 local resilience/backend paths do not claim deployed Themis,
  vendors, transport, promotion, or full-engine parity;
- V8 gates do not stand in for licensed human gold, real-player outcomes, or UE
  placement;
- V9 mobile and Metis seams do not stand in for deployed delivery, cooked
  placement, localization review, embodiment, or payout;
- P1/P2 evaluators do not stand in for deployed drills, current organizational
  decisions, production outcomes, controlled comparisons, or an actual distilled
  model.

No checkbox was converted from unchecked to checked based only on fixtures,
contracts, deterministic baselines, or local simulations. The two V4 rows, the
V7 cook/place row, and the four V5 repository-local rows were closed only after
execution on the exact qualified engine, project, user, target, and source
commit with authenticated log/report digests.

## Current-tip delta review

The previous row-level audit was based on
`cb562888f78ab1bba71d2e39b5ec12993f0f7047`. The ledger, evidence, qualified
runtime, remote integration, capability-truth recertification, V2/V3
host-preflight changes, historical AUD-004 prose qualification, and V5/V8
canonical audit reconciliation through
`139c19d99a2d96237cee1d0bc1a759ac3df7becf` were reviewed before issuing this
updated record:

- the exact unchecked-text digest is now
  `ad5e5dd05391de577451cb825d05c60478e026de26f1c45035a9db23403ce6b6`;
- the intervening changes add repository implementation, tests, generated
  controls, completed-package evidence, and source-bound V2/V3 host
  dispositions;
- no remaining blocker pack acquired authoritative live-provider, deployed
  infrastructure, production-runtime, licensed-human, consented-participant,
  release-decision, or financial evidence.

The disposition is now 8 standing rules, 125 external-state requirements, and 12
portfolio-dependent gates, with an auditable record for every row.

## Drift control

`scripts/audit/remaining-actionability-manifest.mjs` fails closed on:

- changed unchecked wording or row count;
- any unchecked task without an explicit blocker profile;
- any stale or unused profile;
- classification or family-count drift;
- missing or changed supporting evidence;
- any difference between generated and checked-in manifest bytes.

`scripts/verify-remaining-actionability-audit.mjs` independently enforces the
audited digest and 8/125/12 summary while delegating exact row/profile/evidence
validation to the manifest checker. Any drift requires a fresh manual
actionability review and regenerated evidence.

## Verification

- `node --test scripts/audit/remaining-actionability-manifest.test.mjs`
- `node scripts/audit/remaining-actionability-manifest.mjs`
- `node scripts/verify-remaining-actionability-audit.mjs`
- `node scripts/verify-v8-docs-drift.mjs`
- focused ESLint, Prettier, syntax checks, and `git diff --check`

## Boundary

This audit is a repository disposition, not external evidence. It does not
authorize provider spend, infrastructure deployment, data acquisition, human
recruitment, production access, release decisions, or completion of any
remaining checkbox.
