# Agentic authority source review · 2026-09-07

This receipt records source investigation and **three unapplied corrections**,
not publication or acceptance of the Agentic Studio guide. Read
[the learning review](../../drafts/agentic-authority-review.md) and
[the prepared replacements](../../drafts/agentic-authority-corrections.json).
Together with the separate recovery manuscript, six slides await correction.

The new 38-test Node suite executes three complete TypeScript helper modules
directly from the immutable source pin. Controlled records establish precise
preflight, grant-selection/renewal/cascade and handoff behavior. Numeric and
empty-value counterexamples are deliberately local; they do not establish a
production bypass or endorse missing validation. No credentials, provider,
network, persistence or model are used by these probes.

Four existing Vitest suites pass, **64 tests** covering invocation, grants,
handoff and the portfolio capability broker. The latter includes rejection of
actor, tenant, purpose, resource, action, trust-zone and risk mismatches. The
tracked package equals the source pin. One worker and disabled file parallelism
kept the run at roughly 754 MiB maximum RSS and 8.5 seconds, with no swap
activity. This is not all package, HTTP, durability, security or release
testing.

Four dry Python authoring checks pass: exact baseline slide hashes, all six
drafts' disjoint identities/joint catalog membership, existing table/column text
rendering, paragraph-separated speech plans and pinned source resolution. They
do not establish browser geometry, accessible user flows, PDF readability or
synthesized speech. The preflight comparison replaces an implied automatic
launch flow in the draft only. Existing grant priorities and visual identity are
retained.

`state.json` binds the three affected slide hashes, five unassigned source
units, source bytes, draft/test inputs and logs. The unchanged delivery inputs
are compared with `reviewBaseRevision`. The verifier explicitly requires false
publication, visual, audio and whole-objective flags. This increment changes no
live slide, runtime code, narration input, HTML or PDF. The ongoing narration
job and its uncommitted tests are separate work and are not bound by this
source-only receipt.

The preceding recovery receipt also bound shared TODO text. This new TODO entry
makes that older all-input receipt historical; it does not change its source
modules, slide baselines or correction manuscript. Do not edit an old evidence
hash to conceal a later change. The joint dry check still validates that the six
pending patches fit the current guide without overlapping or editing it.

```sh
node --test tools/presentations/tests/pinned-agentic-authority.test.mjs
python -m unittest discover -s tools/presentations/tests -p 'test_agentic_authority_draft.py'
node_modules/.bin/vitest run --config libs/oshun/agentic-studio/vitest.config.ts src/invocation/invocation.test.ts src/grants/grants.test.ts src/handoff/handoff.test.ts src/grants/portfolio-capability-broker.test.ts --maxWorkers=1 --no-file-parallelism
python docs/presentations/presentation-center/verification/agentic-authority-source-review-2026-09-07/verify_evidence.py
```

Use the prepared Python environment and resource checks. After the live
narration queue is stable, revalidate and apply both drafts deliberately,
refresh citations and affected inventories, and render all six changed tracks.
Update the phase-specific authoring tests to retain proof of the applied
changes, not to silently ignore a changed baseline. Complete the full affected
standalone/embedded, table/columns, actual-audio, PDF, optical and listening
checks. Source-unit assignments require substantive teaching, not passing source
probes. The broader guide, library, craft and real-reader requirements remain
unfinished.
