# Eve supplemental-revision assignment · 2026-09-08

This receipt records a fail-closed bridge between the presentation center's
primary source inventory and unchanged Markdown cited by a later, committed
slide snapshot. It also records the first section-level assignment enabled by
that bridge: the complete **“Two planes”** section of
`libs/oshun/product-graph/README.md` to `eve-cap-graph-planes` in the
`eve-product-graph` guide.

## Revision boundary

The primary inventory remains pinned to
`1ceae52c7896a59ac74a4429d410f3218f0b91c6`. The twelve Eve capability guides are
intentionally pinned to `8c27a8d3350fe7484c18585b570f72a561090540`. A newer
slide cannot close a primary-inventory unit merely because it cites the same
path.

Assignment files now retain their declared full source SHA during validation.
For a non-primary revision, the coverage gate reads the exact committed Git blob
and compares the hash of the complete file with the primary inventory's pinned
file hash. The assignment also has to match the unit path and heading, the slide
has to declare the assignment revision, and the existing unit, slide, citation,
guide-membership, rationale and explanation checks still apply. A changed or
missing file, working-tree substitute, wrong revision, wrong path, wrong heading
or stale slide remains rejected. Workflow declaration units use their existing
pointer/role contract and do not acquire a Markdown-heading requirement.

The [review record](review.json) shows that the product-graph README has SHA-256
`9553bf8a8ad300335fc19d33d5cd6486c8b4c3cd3e0dbfea96e248f584a89212` at both
revisions. No equivalence is inferred for the other supplemental sources.

## Editorial review

The complete 61-word source unit says that repository knowledge is compiled and
read-only at runtime, while conversation-born intent lives in Postgres and
reaches graph analysis only as an intent projection. Promotion occurs through a
source change and artifact rebuild.

The dedicated slide teaches those same boundaries through a three-part visual,
three notes and three questions. It further qualifies the assistant's serving
runtime as a narrower manifest-checked composition than the complete compiled
graph. This assignment does not cover projection setup, cross-plane query packs,
implementation files, live database state, deployment behavior, comprehension,
listening or product acceptance.

The generated teaching report moves from 41 reviewed / 103,285 unassigned to
**42 reviewed / 103,284 unassigned**, with 10,018 structural units unchanged and
no invalid or stale assignments. The global completion gate still exits 1 and
the backfill checkbox remains open.

## Verification

- A complete pinned Docs Center recollection matches all 105 inventory shards:
  9,262 authored sources, 319 generated pages and 3,505 entities.
- Repository and workflow inventory checks pass. Their indexes changed only to
  retain the new shared collector hash and, in the workflow index, the resulting
  upstream repository-index hash; all 140,123 repository entries and all 3,718
  workflow declaration units remain unchanged.
- All 20 inventory tests and 20 workflow inventory tests pass, including
  byte-identical admission and changed/unattested/revision/path/heading
  rejection. All 255 presentation Python tests pass.
- Builder freshness remains 578 slides from 347 cited sources. No presentation
  content, generated HTML/PDF, narration, catalog or coverage artifact changed.
- Coverage freshness passes at 42 reviewed units; its intentionally strict
  global completion mode returns 1.
- Task-scoped Ruff, Python compilation, Prettier, local links, diff checks and
  this receipt verifier pass.

The implementation baseline is `67b3ec573070d67153981a205b00b6644fe53045`. No UI
or delivered presentation artifact changed, so browser and visual review from
the bound Eve delivery remain valid and were not rerun for this
assignment/tooling package.
