# V2 contracts integration · 2026-09-07

The eighteen-slide guide's recorded delivery is verified. This is a bounded
integration receipt, not completion of the exhaustive presentation-center
objective, whole-guide craft acceptance, normal-speed listening or reader
review. The [state](state.json) binds the current artifacts, reports, source
inputs, audio and 157 reviewed images.

The prerequisite-linked **V2: contracts and verification** guide adds eighteen
slides and eighteen recordings. It follows a documented claim through reference
conventions, artifact ownership, floating-point build/cook policy and generated
gRPC metadata to a bounded verification result. The
[authoring brief](../../authoring/v2-contracts-guide.md) records every slide's
communication job, reading order, transition and eight-standard quality ledger.
The historical sixteen-slide draft is preserved, not overwritten or rebound to
the integrated guide.

## Source and teaching boundaries

All claims use source pin `1ceae52c7896a59ac74a4429d410f3218f0b91c6`. The new
source-execution tests run the pinned Arete Zod declaration and the actual
pinned TypeScript generator in disposable fixtures. Complete pinned proto
discovery reproduces a stored **44-service / 666-RPC** baseline versus **47 /
675** after isolated regeneration. The added services are Commentary, Director
and Esports, with two, three and four RPCs respectively. Every other generated
C++ byte is unchanged; the header changes only those counts. The first read-only
check fails without repairing files, then the regenerated fixture passes. A
separate options-body example demonstrates that a later ordinary RPC can remain
omitted while freshness passes. These are metadata observations, not Unreal
compilation, serialization, authentication or live transport tests.

Python source regressions cover actual compiler/cook tokens, the absent editor
unity override, reserved definitions without pinned C++/header consumers,
selected checker omissions, documentation-parent fallback, route-specific DDC
arguments and dated module/contract/checker inventories. A missing DDC argument
alone does not establish the effective inherited cache. No product source or
historical source snapshot is repaired to make these explanations pass.

The [five assignments](../../assignments/v2-contracts.json) cover the complete
cross-reference, documentation-enforcement, worked-determinism, tree-enforcement
and module-topology overview units. They raise reviewed documentation units to
31; **103,295 documentation units and 3,718 workflow units remain unassigned**.
The complete teaching gates must continue to reject a whole-library completion
claim. Related citations, inventory counts and this guide's passing checks are
not exhaustive V2 or whole-monorepo teaching.

## Verified delivery scope

- 187 Python and 274 Node tests pass, including source execution, authoring,
  renderer, preservation, inventory and negative-control regressions.
- The [complete bank](standalone-bank.json) passes all 12 checks for all 409
  slides: desktop/compact/mobile, accessibility, source/transcript/navigation,
  PDF sequence, actual per-track playback and HTTP failure/race handling. Its
  process returned zero. This does not claim normal-speed listening.
- The [complete center](center.json) passes all 14 checks across 39 guides, all
  435 guide-slide instances at two viewports, 78 audio boundaries and 39 PDFs.
  The 39 measured guide selections stay within the local interaction budget; the
  maximum observed duration is 244 ms, not a production latency SLA. The
  launcher observation is qualified below.
- The [standalone guide](standalone-guide.json) passes its full eighteen-slide
  non-draft suite; the [focused guide](focused.json) passes 82 states and ten
  defect injections. The [table regression](table-regression.json) passes 34
  captures and the [companion regression](companion-regression.json) passes 11
  captures and five defect injections. These focused checks do not replace the
  complete suites.
- All 409 audio files pass current fingerprints, hashes, duration, decoding and
  encoded pauses: 36,924.738 unique seconds and 39,002.060 route seconds. The
  new guide contributes 2,083.299 seconds. All 391 prior recordings, slides and
  speech plans and all 38 prior guide HTML files are preserved.
- Builder/preview freshness, repository/workflow inventory freshness, both
  teaching-report freshness checks, Ruff, scoped ESLint, Prettier and whitespace
  checks pass. The separate complete-teaching gates still reject their
  documented backlogs. The Domains audit still fails on 13 existing registry
  issues; none is silently reported as resolved by this delivery.

## Composition and actual-size review

Two new comparisons distinguish executable value validation from artifact checks
and show the reproduced freshness mismatch. Independent declaration, consumer
and evidence questions have no ordinal markers. Real investigation and
generation sequences retain their order.

Actual-size inspection shortened two prose labels and the displayed schema name
while preserving exact identifiers in notes and sources. Three source tables
explicitly opt into `sourceWrap`: native optional breaks at identifier
boundaries, a 132–146px mobile row column and 14px right cell padding.
Selectable characters, speech, body type sizes and all prior guide HTML remain
unchanged. No decorative image, mock UI or new motion is needed for these
source-reading tasks. This is how the mandatory frontend skill influenced the
integration.

The focused matrix has 82 checked states, 38 captures and ten rejected defects.
It includes all eighteen compact active-playback states, five tables at 320px
and with 20px table text at 390px, exact source/table content, word-range
geometry, row identity and playback-control bounds. Defect injections cover
hidden content/seek, changed counts, reduced type, false ordinals, reversed
order, the original enlarged prose labels and removed optional source breaks.
PDF checks preserve non-whitespace source characters and ordered cells.

Model review covers all eighteen desktop, compact, mobile, embedded desktop and
embedded mobile states, all eleven mobile table-end captures and all eighteen
PDF pages at 1440px page fit. Compact playing and narrow/enlarged evidence are
reviewed separately. Unchanged prior integration captures are inherited only
through exact image equality; non-overflowing start/end pairs are identical
images, not independent observations. All eighteen delivered PDF page rasters
exactly equal the inspected rasters. Re-exporting the other PDFs does not imply
new whole-library optical acceptance or byte preservation of their PDF metadata.
The [prior-PDF delta](prior-pdf-delta.json) checks all 38 against the baseline:
35 change only creation/modification timestamps; three also consistently rename
accessibility node IDs. After timestamp normalization and bijective ID renaming,
every other byte is identical, including tag-reference relationships. Reproduce
that comparison from the repository root with
`node docs/presentations/presentation-center/verification/v2-contracts-integration-2026-09-07/scripts/check-prior-pdfs.mjs`.

The mobile frame scrolls vertically. Initial embedded captures do not prove
whole-slide first-screen fit; full-height reading and table-end captures cover
the remaining content separately. CSS text enlargement is not a complete browser
zoom/font-fallback test. These are model observations and automated geometry
checks, not representative-reader comprehension or normal-speed listening.

## Attempts and evidence integrity

Earlier label and source-wrap attempts are not final passes. The first enlarged
source treatment still fragmented `Definitions`; a later 180px minimum value
column did not fit beside the sticky row at 320px. Both were rejected before the
passing 132–146px/padding treatment. A new PDF assertion initially expected
mixed-case headers despite the existing uppercase print style; the check was
corrected without relaxing source-cell case, punctuation or ordering.

Final freshness checks also rejected stale repository, workflow and workflow
teaching index hashes after the source-table metadata changed. The standard
generators refreshed the dependent projections; both inventory checks and the
workflow teaching freshness check then passed. The pinned file identities and
workflow units are unchanged. The repository's 27 newly cited paths and their
dependent citation counts/hashes are the intended discovery delta. Initial
failure and regeneration logs are retained alongside the final checks; complete
workflow teaching still correctly fails with 3,718 unassigned units.

The complete center launcher reported status 143 while its child was still
observably running. That same child subsequently wrote the completed report,
passed every assertion and exited; all tested HTML and delivered PDF hashes were
checked afterward. Its termination status was not observed as zero. The cause of
the launcher discrepancy is not established. The receipt separates that
observation from the actual completed report and does not restart a job merely
because a launcher observation ends. An earlier standalone attempt also reported
143; the later completed standalone report is retained separately.

Retained logs omit only host `ps` rows/header that expose unrelated workspace
commands. The state records each original log hash, retained log hash and
omission count. Test output and memory/disk observations are otherwise
unchanged; the filtered logs are not described as verbatim originals. Historical
receipts and canonical earlier full-bank/center reports keep their original
hashes.

## Remaining objective

Normal-speed listening, willing-reader sessions, the accepted benchmark,
complete typography and composition adoption, all eight craft standards, every
Domains route, real Eve integration and exhaustive V1–V10/Docs Center/monorepo
coverage remain open. No participants or comprehension results are invented.

The [orientation map](../../drafts/v2-orientation-learning-map.md) and
[rollback map](../../drafts/v2-rollback-learning-map.md) prepare the next
coherent source work. The rollback continuation now distinguishes inclusive
resend counts, selected decoder assertions, parity, datagram metadata,
snapshot-budget predicates versus session admission, and actual transport; those
new C++ cases are planned, not executed or counted as reviewed teaching units.
The local Unreal editor exists and must be used for applicable execution when
resource-safe.

## Reproduction

Use the repository resource preflight and prepared presentation dependencies.
Run expensive gates one at a time through
`tools/presentations/run-supervised.py`. Linux uses `CHROME_CHANNEL=chromium`;
set a fresh output directory for each browser suite's `OSHUN_*_QA_DIR`
environment variable.

```sh
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test --test-concurrency=1 tools/presentations/tests/*.test.mjs
node tools/presentations/tests/v2-contracts-guide.mjs
node tools/presentations/tests/eve-oshun-deck.mjs --guide v2-contracts
node tools/presentations/tests/eve-oshun-deck.mjs
node tools/presentations/tests/presentation-center.mjs
node tools/presentations/tests/table-pilot.mjs
node tools/presentations/tests/v2-companion-label.mjs
python tools/presentations/render-eve-oshun-narration.py --check
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/check-center-coverage.py --check
python tools/presentations/inventory-repository.py --check
python tools/presentations/inventory-workflows.py --check
python tools/presentations/check-workflow-coverage.py --check
```

The separate documentation/workflow `--require-complete` gates and final Domains
audit remain required; do not turn their current incomplete results into release
passes. See [TODOS.md](../../TODOS.md) for the full unchanged objective.
