# V2 connection-quality integration · 2026-09-07

This receipt records the twelve-slide guide's narrated integration, not final
craft acceptance or completion of the exhaustive presentation-center objective.
The [state](state.json) binds reports, exact inputs, delivered artifacts, audio
and [177 inspected images](visual-review.json). The guide's two concrete
typography findings, normal-speed listening and representative-reader evidence
remain open.

The canonical **V2: connection quality and fallback** guide belongs under
Products → V2 competitive fighting → Netcode. Product promise and contracts are
its prerequisites; delivery/release remains last in the catalog. Its
[brief](../../authoring/v2-network-quality-guide.md) records each slide's job,
relationship and transition, four neutral reader tasks and all eight quality
standards. No game or dashboard screenshot is invented to imply a consumer that
has not been demonstrated.

## Source and native evidence

Source pin: `1ceae52c7896a59ac74a4429d410f3218f0b91c6`. The guide teaches
mean RTT/loss versus maximum jitter, ordered handshake thresholds, consecutive
trailing pressure entries, first-matching fallback branches, separate validation,
per-term score clamps and local diagnostic metadata. Its worked case returns
Borderline/Hybrid and ranked eligibility; it does not prove a live mode change.
An offered no-contest flag is not evidence of a player decline or settled match.
A valid configuration is not necessarily aligned with the default feature
policy. A stable accepted Hybrid record can fail its separate decision predicate.
Setting an export flag and endpoint name does not send a request.

The [historical draft receipt](../v2-network-quality-draft-2026-09-07/README.md)
retains twenty passing native cases in an explicitly adapted Linux UE 5.5.4
diagnostic host. This integration checks all 129 historical evidence hashes and
seven unchanged native preparation/harness inputs; it does **not** rerun or
broaden those cases. The original unchanged-pin UHT failure, twelve substitutions
in six fixture files, missing feature-pack preview startup errors and all
Blueprint/full-game/network/release limitations remain attached to that evidence.
No V2 product source was changed.

The frozen draft remains byte-identical to commit
`42e255904744c37c7eb6f131a31e7844bfe7b855`. Canonical visual fields match it;
only the closing manuscript-publication sentence becomes a permanent reminder
to retain the source revision and environment. No teaching assignment is added:
31 documentation units are reviewed, while **103,295 documentation units and
3,718 workflow units remain unassigned**. The complete teaching gates remain
required and must reject this incomplete coverage.

## Automated delivery

The retained reports and observed command exit statuses distinguish full runs
from focused checks:

- [Full guide](standalone-guide.json): all twelve slides, desktop/compact/mobile,
  accessibility, sources/transcript/navigation, PDF sequence and real audio,
  including HTTP failure and race handling; non-draft with no subset flags.
- [Full center](center.json): 40 guides, 421 unique slides, all 447 guide-slide
  instances at desktop/mobile, 894 embedded captures, 80 actual-audio boundaries
  and all 40 PDFs. All 40 measured selections meet the local interaction budget;
  maximum observed duration is 238 ms, not a production SLA.
- [Full bank](standalone-bank.json): all 421 slides, desktop/compact/mobile,
  accessibility, sources, navigation, PDF sequence and actual per-track playback;
  non-draft with no subset flags. This is automated playback, not listening.
- [Focused integration](focused.json): 168 checks, ten rejected defects and
  108 captures. Exact authored strings, inequalities, quantities, units,
  responsibilities and source annotations remain visible. Narrow 320px and
  enlarged-text probes retain their scope; CSS enlargement is not full browser
  zoom or font-fallback coverage.
- [Draft regression](draft-regression.json): all original 60 checks and eight
  rejected defects remain usable against the frozen isolated draft. The unknown
  source-specimen control now uses a genuinely unknown ID; the integrated guide
  is no longer incorrectly assumed absent from canonical content.
- [PDF content](pdf-content.json): all twelve pages retain ordered authored
  text after NFKC and whitespace-only normalization. Source-cell case,
  punctuation, inequalities and exact characters are not discarded. This checks
  extracted content, not ink visibility or legibility.

The verification ledger also requires all 217 Python and 274 serial Node tests,
Ruff, scoped ESLint/Prettier, builder/preview freshness, repository and workflow
inventory freshness, both teaching-report freshness checks and the all-421 audio
integrity gate. The two `--require-complete` results are explicitly expected
rejections, not passing complete-coverage results. No broad Nx/app-build matrix
or unrelated application verification is substituted for these presentation
checks. No TS/TSX, mobile app, API or production renderer code changes here.

## Preservation and audio

The library now contains 40 guides, 421 unique slides, 447 guide-slide instances
and 287 cited source files. All 421 recordings are current, totaling 38,115.121
unique seconds and 40,192.443 route seconds. The new guide contributes
1,190.383 seconds. Current audio means fingerprints, hashes, duration, decoding,
encoded pauses and actual playback are checked; it does not establish natural
delivery, pronunciation by listening or listener comprehension.

[Continuity](continuity.json) checks all 409 previous complete manifest entries,
recording hashes and speech scripts; all 33 prior canonical content files; all
39 previous guide definitions and HTML files; and the frozen draft. No existing
guide content is replaced. The repository/workflow projection changes are
exactly seven newly cited V2 paths, their counts and dependent hashes; pinned
file records and workflow declaration units are unchanged.

All 40 PDFs were exported by the full center run. The
[prior-PDF comparison](prior-pdf-delta.json) examines all 39 previous PDFs against
the synchronized baseline: 37 change only creation/modification timestamps; two
also consistently rename accessibility node IDs. Timestamp normalization and
bijective first-occurrence ID renaming preserve every other byte, including tag
relationships. This is not a new every-page optical review of the other guides.

## Visual review and unfinished acceptance

The mandatory frontend skill directs a restrained evidence-led composition:
cream/ink/terracotta, the established Georgia/Arial stacks, exact rule tables,
unnumbered independent responsibilities and selectable source code. Only actual
branch precedence and the investigation sequence carry order. Existing
navigation/disclosure motion respects reduced motion; no ornamental bitmap or
new motion is added.

Model inspection covers 44 standalone images, 108 final focused images, twelve
delivered PDF pages at each of 960px and 1440px width, and one complete-guide
contact sheet. Sixty final focused captures inherit this integration's earlier
inspection only through exact image equality; the other 48 are inspected anew.
All PDF images are rendered from the delivered PDF. The contact sheet reviews
sequence and pacing, not thumbnail text legibility. Repeated tables compare
different exact rules; responsibility, source-artifact and investigation slides
change composition when their communication jobs change.

The five embedded contexts are 1440×900 and 1280×720 with the sidebar expanded
and collapsed, plus 390×844 mobile. Every new slide also plays at standalone
1280×720. In the three smaller embedded contexts, 36 additional playback checks
scroll to the seek control and verify all playback controls fit inside the
actual frame. A hidden embedded seek must fail. These are reachable controls,
not a claim that a whole slide fits the first embedded viewport. Tall mobile
reading/table-end images are labeled separately from initial phone captures.

Two defects remain explicit in the brief and visual ledger:

- Mobile score labels can separate `200`/`ms` and `40`/`ms` across lines. Preserve
  the exact values and units while repairing number/unit cohesion.
- Ordinary prose-table headers look noticeably small at PDF page fit. Resolve
  effective header sizing with typography, font/zoom and visual regression
  evidence; passing table geometry is insufficient.

Neither defect is waived by integration. The complete typography standard,
reader-backed viewing thresholds, willing-reader tasks, normal-speed listening
and final editorial acceptance remain required. No participant, unaided answer,
accommodation or comprehension outcome is simulated or claimed.

## Attempts, process observation and retained evidence

The original 132-check focused integration passed. Inspection then identified
the need to test reachable playing controls in the smaller actual frames; the
expanded 168-check suite passed with its additional negative control. Its first
lint check rejected four unqualified browser globals; explicit `window` access
corrected them. No source values or renderer geometry were changed for that fix.

The final driver's first lint command accidentally omitted the documented
presentation-specific config and received 32 undefined-global errors from the
monorepo config. The corrected command uses the existing
`tools/presentations/eslint.config.mjs`, which declares the Node/browser execution
contexts used by these tests. No lint rule is disabled and no source is changed
to suppress that attempt; its failed command, status and log remain retained.

The preservation script initially asserted two incorrect JSON field names
(`cited` and `repositoryIndexSha256`). Inspection corrected them to the actual
`citedInPresentations` and `repositoryDiscoveryIndexSha256` fields. The resulting
comparison verifies the seven precise citation changes and rejects all other
projection deltas. This was a checker-authoring error, not an inventory repair.
The PDF check's success label now correctly distinguishes integrated and draft
mode, without weakening content assertions.

Full center and focused process exits were observed as zero. The same live bank
handle was polled across the intervening sync request; an observation yield was
not treated as a stopped process or permission to restart it. Verification runs
are serial and supervised with memory/disk/process observations. The host has
15 GiB RAM and 31 GiB swap; no unapproved whole-workspace gate is used.

Retained logs omit PID-prefixed host process rows and the `ps` header to avoid
publishing unrelated workspace commands. Each raw/retained hash and omission
count is recorded; these are not described as verbatim raw logs. Test output
and resource observations otherwise remain intact. Historical receipts keep
their original outcomes; this receipt never relabels a previous launcher status.

## Reproduction and next work

Use prepared repository dependencies, a fresh output directory for each suite,
`CHROME_CHANNEL=chromium` on Linux and the resource supervisor. Run expensive
commands one at a time after the required resource preflight:

```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/eve-oshun-deck.mjs --guide v2-network-quality
node tools/presentations/tests/v2-network-quality.mjs --integrated
node tools/presentations/tests/v2-network-quality.mjs
node tools/presentations/tests/presentation-center.mjs
node tools/presentations/tests/eve-oshun-deck.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
node docs/presentations/presentation-center/verification/v2-network-quality-integration-2026-09-07/scripts/check-continuity.mjs
node docs/presentations/presentation-center/verification/v2-network-quality-integration-2026-09-07/scripts/check-prior-pdfs.mjs
node docs/presentations/presentation-center/verification/v2-network-quality-integration-2026-09-07/scripts/verify-retained.mjs
```

The draft regression needs the normal `--draft v2-network-quality` builder
output first. Browser report directories use `OSHUN_DECK_QA_DIR`,
`OSHUN_CENTER_QA_DIR` or `OSHUN_QUALITY_QA_DIR` as applicable. The archived local
driver and rendering script retain their actual temporary paths for provenance;
adapt those paths for reproduction. `verify-retained.mjs` checks hashes only and
must not be called a new browser/native execution or reader result.

Continue the full [TODO contract](../../TODOS.md): address these typography
findings, finish source-unit assignments and rollback/session/transport teaching,
the complete pilot and all eight craft standards, reader/listening review,
exhaustive V1–V10/Docs Center/whole-monorepo coverage, real Eve integration and
the final Domains audit. This receipt does not reduce that scope.
