Create OPDA notebook artefacts and their presentation indexes
Status and scope
Accepted decision, followed by the owner’s explicit commission to implement and generate the finite Property Pack learning catalogue on 26 October 2026. The OPDA source, generation, delivery and local index adapters are now implemented and exercised. Runtime completion evidence is recorded below; generation completion does not imply content approval. No notebook was shared publicly and the media index is served only on loopback. ADR-0081’s historical identities remain intact.
Context and decision
ADR-0081 is Accepted and already owns purpose-specific portfolios, source manifests, preparation scripts and private notebooks. Its README label of “implementation pending” is an incomplete description of the subsequently recorded preparation; this ADR distinguishes existing preparation from the new artefact pipeline.
Reuse those owners and the user-level notebooklm-authoring skill. Transfer the
successful authoring and presentation practices from AI Strategy Business
Architecture, not H&M’s topics, audience, source bodies, notebook IDs or deployment.
Do not mirror OPDA’s website sitemap mechanically: a useful media account may span
several routes, while a technical reference route may not need media at all.
Organisation: portfolios, sections, chapters and topics
A portfolio/notebook is a source and audience boundary. A section is a coherent reader interest. A chapter answers a substantial question within it. A topic is an individually useful explanation for a medium, not simply a smaller heading. A topic may cover the complete chapter or one meaningful aspect. Stable IDs join these entities; titles and route labels remain human-readable. Keep provenance through renames and splits; never silently repurpose an ID tied to old media.
Start with the existing six purpose-specific notebooks and aggregate discovery notebook. Revalidate their current authority and audience against the latest OPDA research; historical numerical limits and August source counts are not today’s service limits or proof of completeness. The aggregate is for discovery, not the automatic source selection for every artefact.
Use this initial organisational brief, refining it from current evidence before freezing the catalogue:
| Section candidate | Primary audience and purpose | Candidate chapter questions |
|---|---|---|
| Programme and shared value | Sponsors and industry decision-makers; establish the case for participation | What changes through shared property information? What is delivered and what remains proposed? |
| Property information in practice | Property professionals and data contributors; explain useful changes in work | How does a property pack move through a transaction? How are meaning, context and provenance retained? |
| Standards and governance | Chairs, reviewers and accountable owners; explain authority and change | Who owns meaning? How are proposals reviewed, challenged, versioned and adopted? |
| Semantic modelling | Modellers and domain specialists; build shared competence | How do glossary, taxonomy, dictionary and model fit together? How are examples and boundaries investigated? |
| Participation and implementation | Working-group participants and implementers; support useful contribution | What should a participant bring? How does evidence become a reviewed contribution? How does a consumer adopt a released model? |
These are design candidates, not a completed new OPDA catalogue or assertions that all proposed standards are operative. Retain the Property Pack/SPDTF/PDTF authority and historical distinctions from their current owners. Archive/history material is selected only where the subject needs it, not presented as current endorsement.
For each section/chapter record purpose, audience, question, useful scope, parent, related subjects, source coverage and topic IDs. For each topic record title, plain-language description, medium, selected source identities and contribution. Choose video, deck, audio and infographic topics independently where their useful accounts differ; do not mechanically copy every topic across all formats. Include a complete-chapter topic when it adds a connected account instead of duplicating all narrower topics. A section overview is a genuine overview, not a concatenation.
Authoring: three inspectable inputs
| Input | What it contains | What it must not become |
|---|---|---|
| Prompt | Short subject commission and relevant audience; a specific intervention only when justified | A prescribed storyline, slide quota, palette, repeated cautions or required ending |
| Source material / direction | Attributed tailored explanation with situations, mechanisms, distinctions, alternatives and consequences; optional synthesis only for a missing connection | A disguised long producer instruction repeated across global/chapter/topic layers |
| Data sources | Exact selected notebook source IDs, complete bodies, versions, hashes, provenance and selection rationale | A list of local paths or citations assumed to have been uploaded |
Distil dispersed technical evidence into worthwhile audience-appropriate readings while retaining consequential qualifications. Select complementary originals only where they add evidence, examples, relationships or useful visual material. Selecting a bundled source exposes its complete membership, not just the cited fragment. Verify what is actually supplied. Original citations not supplied belong in provenance and must not be displayed as selected NotebookLM content.
A suitable starting prompt is “Explain how shared property information changes a transaction, for property professionals.” This is an example, not a universal recipe. Let NotebookLM decide the explanatory sequence, visual language and composition. Avoid “standalone presentation”, “independent reading”, “self-contained deck”, “without a presenter” and equivalent hidden demands across every input layer. Do not replace these with generic verbosity or mandatory visual instructions.
Review the complete resolved package: global, section, chapter, topic and format text; source-internal directions; optional synthesis; final prompt; native request arguments and adapter defaults. Remove duplication and unintended instruction. There is no compulsory separate editorial source. Producer notes are not a default: verify current support and test separately before adding another instruction layer.
The originating comparisons were limited local deck experiments, not proof of a universal best prompt length, source count, synthesis or cross-media causal result. Review actual OPDA outputs before generalising creative quality.
Skill, agent and default outputs
Installed shared skill locations on this machine:
- Codex:
/Users/henrik/.codex/skills/notebooklm-authoring/SKILL.md - Claude:
/Users/henrik/.claude/skills/notebooklm-authoring/SKILL.md
Use $notebooklm-authoring in Codex or the corresponding Claude skill. Its companion
notebooklm-author agent follows the caller’s selected model/effort; do not silently
choose a transport or assume the skill itself starts an agent. Confirm installed
agent registration in the target host. The skill is authoring guidance plus a thin
workflow adapter, not an installed NotebookLM API, scheduler or renderer.
| Medium | Defaults for commissioned topics |
|---|---|
| Language | British English (en-GB) |
| Video | One Explainer with Auto-select visual style and one Cinematic without a visual-style override |
| Deck | One Detailed Deck and one Presenter Slides; both Long |
| Audio | Brief, Deep Dive, Debate; each native Default length |
| Infographic | Landscape, Auto-select; Concise, Standard, Detailed; two takes each, same selected sources and prompt |
These are overridable owner defaults, not Google recommendations. No invented video-duration control. Resolve actual client labels/values live; do not copy numeric API enums blindly. No flashcards, quizzes, data tables, mind maps or written reports in the default set; ADR-0081’s separately requested briefings can remain explicit additions. If one topic receives every default there are 13 outputs, but format-specific topics mean totals must be calculated from actual requests.
Current owner commission: every chapter, every media family
The owner expanded the implementation commission on 26 October 2026: every chapter must receive videos, decks, audio and infographics using the full defaults above (13 outputs per chapter). This explicitly overrides the initial selective 13-output catalogue. Topics still have useful medium-specific accounts; the curriculum is organised by audience questions, not copied from website routes. The owner also instructed that local output reviews must not hold generation. Source identity/readback, native settings, duplicate prevention and delivery integrity remain necessary; they are not discretionary content-approval gates.
docs/notebooklm/property-pack-media/curriculum.json records the expanded teaching
scope: five sections and 22 chapters, targeting 286 outputs. The sections cover
purpose/history/programme, the Property Pack model, building/governing standards,
participation/implementation, and meaning/evidence/trust. Deeper repository research
adds consumer economics, claims/provenance, identity/roles, time/state/change,
privacy/access, and operational sandbox learning. These are evidenced subjects,
not a quota of additional website pages.
Source packages combine substantial attributed readings with independently useful complete originals: government and industry research, constitutional/programme documents, sandbox records, modelling decisions and concrete candidate artefacts. Native PDF/DOCX/PPTX files retain their original bytes and complete indexed-text readbacks, with visual/table extraction limits stated explicitly. Native readbacks can rotate provider image-URL tokens; identical complete text, unchanged image dimensions and original-file hashes preserve the first frozen readback while fresh observations are retained separately. Other content changes fail inheritance. Older receipts establish provenance and public-source authorisation; current versioned sources retain their own hashes. Private email and newly accessed private repositories are not admitted automatically.
Freezing and submission proceed in bounded source-ready batches. catalogue.json
retains the original 13 requests; catalogue-v2.json adds 39 to complete the first
four chapters, while later catalogues add the expanded subjects. Successor records
inherit unchanged requests, acknowledged provider IDs and originals from their
actual owning run. They do not create another job for unchanged inputs. Separate
run supervisors own disjoint new requests; one index lock serialises rendering,
and the current-run pointer prevents an older completion from replacing the newer
directory. Every delivery refresh also advances that aggregate view, including
late completions inherited from an earlier run. The final
catalogue-complete-v1.json combines 22 readings (22,417 words) with 38 complete
originals. It inherits 156 acknowledged requests and commissions 130 new ones.
The workflow binding selects this complete catalogue. The 286 target is a
commission, not a claim that every job is complete.
Reference copy and reuse contract
Reference archive contains 20 source files: AI section directory/scaffold/routes and latest run-index adapters, EA shared notebook components, and skill/adapter files. Manifest records exact file hashes, source paths and repository commits. It is a reviewed transfer reference, not a complete runnable dependency closure, installed OPDA application or licence to copy source research into OPDA. It excludes media, research bodies, credentials and runtime DBs.
Extract into a temporary inspection directory when needed. Do not execute its AI-specific adapters unchanged: they reference H&M variants, private run ledgers, source pins and external owners. The latest AI run adapter demonstrates the current three-input inspection view; older generic scaffolding still has legacy synthesis terminology. Follow this ADR’s current behaviour, not those old labels/defaults.
Production implementation must reuse the shared EA renderer at an exact reviewed revision, supplying OPDA data/options. Resolve its transitive components, styles, fonts, media converters and dependencies through the existing shared-renderer mechanism; verify the chosen release, rather than mixing current working files and an older config pin. Keep the archive immutable as provenance. Fix shared behaviour in the shared owner instead of maintaining an OPDA fork. OPDA branding belongs in configuration. The captured files alone do not establish portability or licensing for third-party redistribution; keep this internal reference out of public assets.
Index presentation contract
- Collection directory: section cards with purpose, representative image and chapter links; equal-height cards within a row. The card is clickable while chapter links remain independently accessible; no nested anchors. A compact section jump index points to cards, distinct from chapter-card styling.
- Section page: section overview plus chapters on one page. Do not invent a page per chapter merely because chapter IDs exist. Stable chapter/topic anchors support links. Begin with collapsed accordions and modest indentation.
- Within each chapter: artefact type → topic → output variants. Opening one artefact type closes the other type. Infographics add detail level → two takes. Do not add redundant Audio/Deck/Video anchor lists below each topic heading.
- Use existing deck lightbox, previous/next and slide index. Current-slide download is on slide-navigation level after the index. Whole-deck PDF download is a right-aligned icon in the heading row. Video open/download, audio download and infographic downloads use accessible icon buttons, not extra text rows or a broken gallery button. No notebook links on public media cards.
- Expose Prompt, Source material / direction, Data sources using exact submitted inputs. Render Markdown as safe HTML with raw text available. Separate provenance-only sources from supplied bodies and actual media from pending work.
- Use verified original media and suitable slide/image renditions for HiDPI screens. Never silently substitute a lower-quality video rendition. Validate dimensions, duration, nonzero body, decode and hash; compare the source/download for quality regressions. Label any future upscale separately and retain original.
- Render the page promptly. Lazy-load inactive media; bounded background prefetch must not block render or launch all videos. Honour reduced motion/visibility, release resources and avoid runaway CPU/GPU. Use OPDA Chrome for browser checks.
- Keep immutable versioned run indexes and a run directory with changes, input links and evaluation notes. Ordinary viewers see useful subjects, not diagnostic counts such as “6 / 6 ready” masquerading as topic descriptions.
Artefact indexing pipeline: copied implementation and OPDA binding
The owner additionally requested the indexing pipeline itself. The separate
pipeline source archive
contains 187 files from the AI index/generation integration, SA programme owners
and EA shared rendering code. Its embedded manifest.json records source
repositories, commits, exact copied file hashes and sizes. This complements the
smaller presentation/skill reference above. It is a source handover, not an
activated OPDA runtime. Copied working-file hashes are authoritative where they
differ from the recorded repository commit. Do not serve either archive publicly.
Data flow and ownership
OPDA reviewed catalogue + frozen per-request inputs
→ acknowledged provider jobs and verified downloaded originals
→ media metadata/quality checks + slide/image renditions
→ catalogue projection joined by stable subject/topic/request IDs
→ shared Astro directory + section/chapter/topic views + input dialogs
→ isolated static output + integrity/link checks
→ OPDA release owner uploads media and matching indexes
→ live byte and access verification, delivery receipt
This is presentation indexing, not the semantic/vector knowledge MCP index. Generation, download, rendering and publication are separate stages. Index updates consume acknowledged state; they never create a replacement provider job.
| Copied owner | Responsibility and required adaptation |
|---|---|
ai/scripts/notebooks/local-comparison-refresh.mjs | Current delivered-run index, exact input checks, media rendition integration and shared-renderer build. Replace AI variant paths and private run locations with OPDA run configuration. |
ai/scripts/notebooks/local-unified-view.mjs | Three input views, topic hierarchy, infographic detail levels, icon/download controls and safe content handling. Bind OPDA IDs, source projections and catalogue; retain shared behaviour. |
ai/scripts/notebooks/inventory.mjs and ai/site/notebooks/ | Collection directory, section pages and chapter/media projection. Supply OPDA catalogue rather than AI programme records. |
ai/scripts/notebooks/video-quality.mjs | Original-video quality selection/checks; preserve distinction between provider original and rendition. |
ea/site/scripts/media-originals.mjs, pdf-slide-images.mjs, pdf-page-render.py | Metadata, original hashes and passive renditions/slide previews; configure real converter dependencies and target dimensions. |
ea/site/src/components/, styles and scripts | Shared directory, subjects, media players, deck preview/lightbox, infographic carousel, inspection dialogs and visual behaviour. Resolve actual transitive imports and assets through shared owner. |
sa/src/programme/shared-renderer.py and prepare-site.mjs | Pin/resolve shared renderer and prepare static assets; change repository-relative location through binding, not a second renderer. |
ai/scripts/notebooks/local-generation.py, pipeline.py and SA programme modules | Existing generation/delivery integration that requests index refresh as verified completions arrive. Adapt only the OPDA data/ownership boundary after dependency review. |
ai/scripts/notebooks/publication.mjs and ai/scripts/site/unified_n01.py | Examples of exact media/index release joins and receipts. H&M-specific scope and bucket assumptions must be replaced by OPDA release ownership; these are not OPDA publish commands. |
Extract for implementation inspection, preserving the archive:
python3 -m zipfile -e docs/notebooklm/reference/notebook-index-pipeline.zip /tmp/opda-notebook-index-reference
The copied pipeline is deliberately not placed on an executable OPDA production path: its historical run IDs, notebook IDs, H&M destinations and fixed subject counts would be unsafe or incorrect there. Keep those snapshots unchanged as references. Bind reusable code once through the shared owners and implement only OPDA-specific catalogue/run adapters in OPDA’s normal scripts/config locations. Do not copy AI source bodies, source IDs, generation state, cookies or credentials. Do not treat all copied components as required; use the import closure of the actual selected views. Font/asset licences and packages remain separate dependencies.
Incremental index and release behaviour
- Catalogue identity includes run, section/chapter, medium, topic, detail/take, provider ID and input hash; a title change must not select another job’s media.
- Update only changed catalogue entries and affected pages/assets; reuse unchanged compatible renditions by original hash plus renderer/settings fingerprint. A full static build may still be useful, but must not trigger media regeneration.
- A failed item remains recorded while ready items continue through download, validation, indexing and scoped release. No fabricated URLs or empty thumbnails.
- Write metadata/indexes atomically. Publish referenced media before the index that exposes it; verify object hashes and access, then invalidate affected routes. Rollback selects a previous complete manifest, not mixed old HTML/new inputs.
- Superseded and deleted items disappear from the new current catalogue while immutable older run pages remain retained unless separately authorised for removal.
- Input-inspection buttons read the exact frozen package for that output, never the newest prospective authoring file. Otherwise a historic deck could appear to have used instructions it never received.
- Resume from the persisted acknowledged IDs and delivery ledger after interruption. One supervisor owns writes; no competing generation/downloader/index loops.
- Validate an OPDA sample through the full chain before claiming installed: actual media, source/prompt dialogs, section navigation, topic variants, downloads, refresh after another completion, interrupted-run recovery and unchanged rerun. Confirm unchanged inputs create no new provider jobs and preserve original bytes.
Amendment — 26 October 2026
Added the full indexing-pipeline source handover, stage/owner mapping, incremental index/release rules and adaptation instructions at the owner’s explicit request. The requested documentation and source copy are complete; OPDA executable binding, generation and publication remain subsequent implementation work, not completed by this amendment.
Implementation sequence and existing owners
- Inventory current OPDA sources/notebook receipts and audience needs. Read current ADR-0081 and affected standards/website decisions before treating old bodies as current. Design and review sections/chapters/topics; record exact IDs and scope.
- Extend
docs/notebooklm/*.yamlthrough their existing owners, preserving accepted original receipts and historical IDs. Prepare complete prospective topic packages with new source identities for changed bodies. Do not mutate frozen past inputs. - Reuse
scripts/build-notebooklm-prepared-sources.mjs,scripts/run-notebooklm-preparation.mjsandscripts/verify-notebooklm-execution.mjsplus their library modules. Read their actual flags before use: preparation can upload/run prompts and is not an offline dry-run. Existing verification includes historical fixed counts; extend deliberately for new catalogue versions. - Bind an OPDA media supervisor/delivery/index adapter to existing native tooling. Adapt the proven shared owners rather than writing another harness. Existing OPDA preparation does not establish that media generation/delivery already exists. Declare SDK/interpreter, authentication profile, Node/Astro, shared renderer, PDF renderer, ffmpeg/ffprobe and required media dependencies explicitly.
- Supply
config/notebook-workflow.jsonwith schemanotebooklm.project-workflow.v1, OPDAcwd, explicitrunId, dependencies and argv arrays fordry-run,prepare,validate,generate,download,index. Use the skill’sreferences/project-workflow.mdandscripts/run-workflow.py. The adapter’s--checkverifies paths/commands, not authentication or API health. - Review resolved inputs, then authorised upload with full indexed source readback and membership checks. Record what fidelity the readback proves. Freeze exact request/source hashes and notebook identities before provider submission.
- Generate a small representative acceptance slice, inspect complete outputs and repair concrete issues. Then run the commissioned set with automatic delivery and index refresh; do not make the owner babysit jobs.
- Before create, persist attempt; immediately save acknowledged provider ID. Reconcile uncertain submissions before retrying. Track actual starts, poll active jobs, distinguish slow/stalled/failed/quota states, and restart through supported same-ID recovery where available. A stalled item does not block others. Saturate actual provider capacity within safe local resources, back off to the provider’s reported reset, resume automatically and stop when the finite manifest completes.
- One supervisor/lock owns each run; downloads continue as completed jobs arrive. Validate bytes and media before atomic catalogue updates; no incomplete assets advertised as complete. Preserve failures and publish useful completions without endless rerolls. Review slides/images visually and actual audio/video playback.
- Implement index data projection and shared components under a separate OPDA
artefact build scope. Test desktop/mobile, keyboard/lightbox/downloads, detail
accordions, input dialogs, safe HTML and missing-media handling. Run
make testand appropriate build before committing code. Use OPDA’s existing CI/AWS owners, not H&M buckets/Cognito/hosts or a new hosting stack.
Acceptance and publication boundary
Done means an OPDA catalogue and all commissioned input packages are reviewed; selected uploaded sources match their recorded bodies; native defaults match consumed arguments; one representative generation/recovery/delivery flow passes; indexes and download bytes are verified; failed/quota/held states remain honest; and subsequent finite runs need no manual babysitting. Retain scoped evidence and appropriate native project-memory readback through the OPDA owner, never this AI project’s connection. Do not write OPDA ADR metadata into another project’s index.
The original ADR commissioned design and reference handover only. The subsequent owner instruction explicitly commissioned generation and local index integration; it did not add media routes to the public OPDA website. Implementation uses the existing repository main-only workflow and scoped commits. Publication follows ADR-0081’s source/sharing boundaries and the actual release approval; no private notebooks, prompt receipts or research reference archive become public simply because an index has been built. Test local first and verify exact served routes before handing over URLs.
Implementation evidence — 26 October 2026
- The initial
docs/notebooklm/property-pack-media/catalogue.jsonfreezes one section, four chapters and four deliberately different media topics: the transaction account (two videos), reading the connected model (two Long decks), semantic ownership and reuse (six infographics), and validation/judgement/release (three audio formats). This initial 13-output slice is retained unchanged; the expanded commission above supersedes its selective scope. - Four attributed readings and three complementary complete originals are selected. All seven private notebook sources passed complete indexed-text readback and exact source membership checks. Verification preserves ordered, case-sensitive words; it does not assert byte-identical Markdown/Turtle/JSON formatting. Frozen inputs retain original, uploaded and indexed hashes and exact consumed native arguments.
- The successor commission is linked from
property-pack-ontology.yaml. It uses a separate owner-only notebook because the historical portfolio was already shared. The previous portfolio, August receipts and frozen bodies are retained unchanged. config/notebook-workflow.jsonbinds the skill’s workflow runner to the OPDA inputs/media/index adapters. Source import reuses AI’sprepare.py; submission, acknowledgements, quota and reconciliation reuse the BA/SA native owners. Video admission reuses AI’s advertised-resolution and complete-decode checks.config/shared-renderer.jsonpins EA revision1a19439285822e852f45cf3ad50bbe22740e85a4through SA’s verified export owner. The local index reuses EA directory/subject/media components and original/PDF rendition owners plus AI’s exact-input readers and topic hierarchy. Its thin adapter supplies OPDA data, navigation, URLs and file-bound review findings.- Runtime inputs, source readbacks, durable provider IDs, media originals and
delivery evidence are retained under the ignored
.local/property-pack-media/runs/property-pack-v1/. The local site is isolated under.local/property-pack-media/site/; it is outside OPDA’s public build. - Four representative jobs completed first. A full 17-slide visual review found
material generated diagram/terminology errors, recorded against the original
hash in
property-pack-media/review.jsonand shown beside that deck. One complete infographic was visually reviewed and Brief audio playback was exercised. These are bounded review observations, not factual approval of the whole collection. - The owner then explicitly instructed “no local reviews” and “generate everything now”. The local review hold was removed and all nine remaining requests were acknowledged without waiting for delivery, rendering or review. No automatic rerolls or duplicate creates were commissioned. The inherited local concurrency cap is disabled for this finite set; provider quota and uncertain-create handling remain with the shared owner. Additional content review was stopped.
- Verified loopback directory:
http://127.0.0.1:4340/development/property-pack/learning-media/. Section:http://127.0.0.1:4340/development/property-pack/learning-media/property-pack/. The equivalent/runs/property-pack-v1/routes preserve exact frozen inputs. Input dialogs distinguish supplied bodies from provenance-only evidence. - Seven focused identity/source/cache/review tests pass.
make testpasses (1,106 passed and one skipped, then 19 passed);make buildpasses. Loopback directory, section and versioned routes and their local resource links resolve. Media and HTML install atomically; unchanged compatible renditions are reused. The live local availability count remains the delivery authority during the run.
Public release authorisation and implementation — 26 October 2026
The owner subsequently explicitly authorised public publication of all 286
artefacts and their index at https://opda.org.uk/presentations/, using OPDA’s
existing S3/CloudFront release owners. This supersedes the earlier local-only
delivery boundary. Historical frozen commissions retain publish:false; the
separate config/notebook-publication.json records the new publication scope.
node scripts/notebooks/property-pack-index.mjs --public verifies the same
provider/file evidence and reuses the pinned shared renderer. It stages all 286
unchanged originals, their renditions, five section pages, the directory and the
corresponding versioned run. Public input readers preserve exact prompts and
supplied text. Native PDF/DOCX/PPTX sources link to the unchanged original files;
their private indexed readbacks and temporary image URLs are excluded. Public
source keys replace provider source IDs. Private notebook links, raw receipts,
renderer checkout paths and reference archives are excluded. Earlier factual
correction notices remain bound to their original media hashes; additional
output review remains waived under the owner’s explicit instruction.
The finite stage is .local/property-pack-media/public-release/site/, with an
external file-hash manifest at public-release/manifest.json. Large immutable
media and supplied native originals use /presentations/platform/media/ and
the existing resources bucket through a dedicated CloudFront behaviour. This
namespace preserves the shared slide-player admission contract. The small
config/presentations/public-site.zip contains only public pages, input texts,
CSS, JavaScript and fonts. config/presentations-release.json pins every file
and the archive; the normal site build installs it through the existing release
process. Source-archive synchronisation preserves the separate media prefix.
Public staging requires all 286 deliveries and rejects private host/transport references. Verification covers current/versioned routes, selected source bodies, original file hashes, production canonical links and the absence of private notebook/source identities. Staging is not a claim of successful S3 upload or live activation; those require the AWS deployment owner’s separate evidence.
Consequences
Reuse preserves the mature media experience and avoids another renderer/scheduler. OPDA retains control of its audience, terminology and sources. Costs are source curation, dependency binding and actual media review; a shared skill does not remove them. Completion history remains auditable while future packages can improve.
Comments
Loading comments…
Sign in to post a comment