media-to-wiki-pipeline ยท diff
git:20260924.a7abae7 to git:20260928.53ba292
20 added, 14 removed. Audit A to A.
---
name: media-to-wiki-pipeline
- description: Design, run, or repair a connected audio/video-to-wiki pipeline spanning transcripts, essays, researched speaker profiles, organization guides, and incremental publication. Not for isolated transcription, article editing, or profile research.
+ description: Design, run, or repair a resumable source-to-wiki publishing workflow across recordings, interviews, articles, documentation, papers or social-post corpora, producing source readers, person/organization profiles and topic pages. Use for connected corpus ingestion, generation and subset refreshes, not isolated transcription or single-page editing.
---
# Media to wiki pipeline
- Preserve accepted content, update affected entities, publish once, and verify the release with sampled live checks.
+ Preserve accepted content, update affected entities, publish once, and verify the release at the required coverage. Use this existing workflow for written and mixed-media corpora too; neither a conference archive nor a particular site's output is a prerequisite.
- Read [references/pipeline-contract.md](references/pipeline-contract.md) for stage contracts and change handling. This describes intended behavior, not proof that an existing implementation meets it. When adapting a project, map its actual acquisition, transcript, identity, writing, ledger, enrollment, publishing, and cache drivers; label verified defects separately from proposed improvements.
+ Read the references relevant to the task:
+ - [Page contracts and quality](references/page-contracts.md) before selecting page types, drafting or accepting a new corpus. It supplies portable editorial defaults, including organization and topic pages; examples illustrate quality rather than substitute for these criteria.
+ - [Source adapters](references/source-adapters.md) when selecting or normalizing sources. Recording clocks and media recovery apply only to recorded input.
+ - [Stage and release contracts](references/pipeline-contract.md) when implementing, operating, resuming or refreshing a corpus. Map actual project drivers before repairing them; intended behavior is not proof of implementation.
+
## Decisions that govern progression
- - Freeze recording IDs per batch; queue later arrivals separately.
- - Resolve identities early, then let essay, person, and organization work proceed independently.
- - Accept substantive bios and guides before enrolling new public entity pages. Brief source-backed writing is valid; placeholder name/affiliation pages are not. Attempt research and repair within budget, then report unresolved enrollment explicitly.
- - Keep historical conference affiliation separate from independently sourced current employment.
+ - Freeze source IDs and versions per batch or subset refresh; queue later arrivals separately.
+ - Resolve identities early, then let source readers, person, organization and topic work proceed independently where their dependencies permit.
+ - Enroll supported identities and relations early; public page readiness is a separate content decision. Brief supported writing is valid; placeholder identity pages are not complete profiles.
+ - Keep dated source/event roles separate from independently sourced current employment. Research disputed identity within budget; do not guess a canonical link.
- Keep accepted writing visible while replacements are reviewed. Holds attach to disputed fields or replacements; correct demonstrated falsehoods or privacy defects locally.
- - Reuse unchanged approvals. Review affected meaning and dependencies rather than every downstream artifact.
- - Recover from existing evidence first; automatic targeted paid transcription may use an authorized provider within the explicit available batch budget.
- - Use one per-entity ledger, with receipts as evidence beneath it.
- - Batch completion uses settled generation/publication and sampled desktop/mobile checks. Held items remain unresolved; sampling does not establish individual live verification or full acoustic certification.
+ - Reuse accepted stages whose relevant inputs remain valid. Select a corpus subset explicitly and expand only to its affected dependents, not unrelated pages.
+ - Recover from existing evidence first. Targeted paid transcription requires an authorized provider and available budget; written sources need no ASR.
+ - Use one per-entity ledger with receipts as evidence. Retain interrupted work and reconcile ambiguous deliveries before retrying.
+ - Completion requires settled required generation/publication and live checks at the requested coverage. A subset's completion is not whole-corpus completion; sampled verification is not individual proof or full acoustic certification.
- Use `video-talk-to-essay` for reader craft, `person-profile-writing` for biographies, and `smart-entity-resolution` for ambiguous identities. Their defaults do not override these decisions or the user's instructions.
+ Use the project's existing canonical path with the portable page contracts as defaults. Batch scripts select entities and supply evidence, not alternate narrow writing prompts. Projects specialize audience, schemas, presentation and release mechanics without replacing source fidelity or editorial usefulness. Archive guides are a distinct requested artifact, not the default organization profile.
- Use configured models, concurrency, providers, and spending limits. Parallelize useful ready jobs; concurrency ceilings are not quotas. This skill does not authorize extra providers, spending, cloud services, or external publication.
+ Pair `research-grounded-writing` and `swyx-writing` for evidence and prose, `video-talk-to-essay` for recorded-source readers, `person-profile-writing` for biographies, and `smart-entity-resolution` only for ambiguous identities. Do not copy project-specific routes, models, paragraph counts or examples into unrelated corpora.
- Stop when the frozen scope has settled outcomes and the release's sampled checks pass. Report exact completed and unresolved counts; continue independent work when one entity is blocked. Do not add optional polish or repeated fruitless recovery to the critical path.
+ Use configured models, concurrency, providers and spending limits. Stream useful ready jobs; concurrency ceilings are not quotas. This skill does not authorize additional providers, spending, cloud services or publication.
+
+ Stop when the frozen scope has settled outcomes and required checks pass. Report exact completed and unresolved counts; continue independent work when one entity is blocked. Optional polish or repeated fruitless recovery should not prolong the critical path.