September 7, 2026 UTC. Source pin: 1ceae52c7896a59ac74a4429d410f3218f0b91c6;
implementation base: 2be447c605bdeae0b01a38225fed092204a71e3d. This is a
composition-pilot increment, not an accepted benchmark or a completed
presentation library. Full acceptance remains in TODOS.md.
Meaning and implementation#
The new architecture-read-handoffs detail follows the retained conceptual
overview. It distinguishes the caller, shared Tara adapter and configured API
read dependency. These are code responsibilities, not invented deployment units.
The actual getHomeCards method starts three reads in Promise.all, waits for
all to fulfill, passes their values to composition and returns its result. One
rejection rejects the call without a partial envelope or automatic cancellation
of the other reads.
The factory can use an injected API adapter or the package HTTP client. The concrete BFF bridge demonstrates injection, not a mandatory deployed BFF hop. The client uses absolute request paths, configurable headers/fetcher and a JSON type assertion rather than runtime schema parsing. The explanation preserves the boundary between supplied user context and authenticated identity.
The frontend skill informed the separate overview/detail composition, open responsibility lanes and editorial subtraction. Wide arrows join owner centers; nonvisual endpoint names become visible on narrow screens. Optional supporting detail is used only for the three source-read names. The first design repeated both endpoint names and redundant transfer prose, overflowing desktop/compact height. Removing that repetition and using the existing 8px compact-playing section gap retained readable labels and complete controls.
The authoring brief and reader protocol document the communication job, scope and neutral tasks. No willing-reader sessions or normal-speed listening outcomes are claimed.
Evidence scope#
The full presentation Python and Node suites pass 109 and 144 tests, respectively. Seven new source tests execute the pinned orchestration, HTTP client and BFF forwarding bridge with explicit controlled dependencies. They verify pending/out-of-order reads, rejection, supplied identity, injected versus built-in dependencies, paths/headers and JSON/error boundaries. Composition spies prove handoff arguments/timing, not downstream card-model correctness, live authentication, rendering, storage or deployment acceptance.
The semantic/PDF tests cover owner identity/order, repeated exchanges, variable owner counts, invalid references/roles/kinds, orphan owners, escaping and missing scope/payload/order in exported text. Browser checks additionally verify actual route attachments and arrowhead direction in screen and print layouts.
The final focused composition run passes 64 captures, including 15 ownership views/stress cases and four rejected/restored negative controls. The complete twelve-slide architecture standalone suite passes all twelve check groups, including actual audio controls/recovery/races and PDF export.
The complete center passes all fourteen check groups: 38 guides, 388 unique slides, 414 instances, 828 embedded captures and 38 complete PDFs. All 91 table instances pass desktop/mobile readability checks (182 reports); the new ownership detail also passes desktop/mobile/print geometry. Eighteen ready guides pass 36 actual desktop/mobile playback boundaries. This remains draft/ready-audio mode, not the final fully narrated release gate.
Fresh current-audio verification decodes and pause-checks all 177 current tracks: 16,007.673 seconds and 2,465 sentence segments. The new 173.323-second detail has 26 segments. All preceding 176 current manifest entries and MP3 bytes remain unchanged. The global backlog is still 211 tracks: 191 stale and 20 missing. Automated pause/decode evidence is not normal-speed listening or pronunciation acceptance by a listener.
Static checks cover builder/preview and teaching-report freshness, the final full presentation unit suites, current composition audit, Ruff, scoped ESLint, normal and separately authored-evidence formatting, and whitespace. The complete teaching gate correctly rejects an incomplete library: 26 reviewed, 10,018 structural and 103,300 unassigned units, with no invalid assignments. No product TypeScript, API, persistence or deployed runtime is modified; the pinned TypeScript fixtures use transpilation diagnostics and controlled execution, not a whole-project typecheck or deployment gate.
Checks ran serially with resource supervision on a 15 GiB host with 31 GiB swap; the center's lowest reported available memory was 12.4 GiB. No broad Nx or product build sweep was run. State binds the source, authoring, renderer, tests and published artifacts; integrity binds the raw receipt files. Empty-output static checks retain their exit status without an empty log file. Raw reports/logs were copied without formatting.
Review limits#
The visual ledger records forty review items, separately from 41 exact-byte inherited regression observations. The new ownership treatment and the complete architecture desktop sequence have been model-inspected at original capture detail. The eight changed existing pilot images were also inspected; 41 other regression images retain exact prior bytes. New diagram acceptance is separate from those inherited observations. The final published PDF's overview/detail pair is inspected at 1280px page-fit width, with the new detail also inspected at 2000px. The complete guide's contact sheet records repeated legacy silhouettes, not small-text readability. Two tentative optical impressions were checked against actual text bounds: overview descriptions share a left edge and the rename-flow paragraphs fit. Neither impression is reported as a confirmed defect or a spacing fix. Tall mobile/enlarged images establish scrollable reading, not whole-slide fit in a phone's initial viewport. An extreme synthetic unbroken owner name can leave a short line fragment despite wrapping without clipping; concise display names remain the authoring guidance, not a claim of complete typography polish.
The architecture opening still has legacy numbered columns. Its conceptual graphs retain unlabeled arrows, and several five-step flows share a repeated silhouette. The qualified conceptual overview is retained, not promoted as the final ownership benchmark. Whole-guide migration, full typography, reader comprehension and the rest of the pilot remain open.
All 64 focused PNGs and the complete standalone desktop/compact/mobile captures are retained. All 828 center PNG hashes were checked during evidence collection; only the 24 affected architecture initial-viewport PNGs are copied into this receipt. The other 804 remain run-local automated outputs, not durable optical evidence. Re-run the complete center command to regenerate them. The failed first and second focused runs are diagnostic history, not passing results.
Reproduce#
Use installed presentation dependencies, fresh output directories and the repository resource checks. Run these commands separately, not concurrently:
python -m unittest discover -s tools/presentations/tests -p 'test_*.py'
node --test tools/presentations/tests/*.test.mjs
python tools/presentations/build-eve-oshun.py --check --preview
python tools/presentations/check-center-coverage.py --check
CHROME_CHANNEL=chromium OSHUN_COMPOSITION_QA_DIR=/tmp/oshun-ownership-new python tools/presentations/run-supervised.py node tools/presentations/tests/composition-pilot.mjs
CHROME_CHANNEL=chromium OSHUN_DECK_QA_DIR=/tmp/oshun-architecture-new python tools/presentations/run-supervised.py node tools/presentations/tests/eve-oshun-deck.mjs --guide architecture
CHROME_CHANNEL=chromium OSHUN_CENTER_QA_DIR=/tmp/oshun-center-new python tools/presentations/run-supervised.py node tools/presentations/tests/presentation-center.mjs --draft --ready-audio
node docs/presentations/presentation-center/verification/ownership-handoffs-2026-09-07/verify-evidence.mjs
For a fresh decode check of this receipt's current subset, pass its
comma-separated audio-current.json → current[].id list to
render-eve-oshun-narration.py --check --ids. The unrestricted narrator check
remains the final all-track gate after the backlog is rebuilt. The receipt
verifier checks saved evidence and current byte identity; it does not run
browser automation, decode audio or replace any human review.