Jiang Corpus Impact Pass

Jiang Corpus Impact Pass

Use this after a newly published episode or interview needs source-to-lens accounting. Do not wait for fixed batches. Every new independent source applies to the corpus; the question is what level of mutation it justifies.

This skill records the impact decision. Downstream public edits belong to jiang-provenance-linker, jiang-lens-concept-writer, jiang-lens-atlas-maintainer, or jiang-canon-promotion.

Operating Modes

Use this skill in two modes:

• Steady-state intake: for a newly merged episode or interview. This is mandatory before optional lens patching. The pass may be compact, but it must account for the source.
• Maintainer-directed backfill intake: for already-published sources that predate the impact requirement. Historical backfill is currently paused for budget conservation unless the maintainer names a bounded source or concept cluster. Do not bulk-write shallow files and do not use the missing-count backlog alone as permission to work.

Budget mode is recorded in content/workflow/tasks/source-processing-policy.json. In budget mode, scheduled agents should process fresh missing impact from newly merged independent sources or explicitly requested clusters only. Known duplicate reuploads and blocked raw artifacts do not create new impact obligations.

Impact intake is not the end state. It creates evidence-backed obligations for public lens work. After compact impact intakes, the next worker pass should switch to downstream synthesis with jiang-lens-concept-writer, jiang-lens-atlas-maintainer, or jiang-provenance-linker when a concrete target is ready. If no fresh source or concrete synthesis target is ready, go idle instead of spending tokens on historical backfill.

Backfill priority:

1. Newest merged sources with no impact file.
2. Tight concept clusters where two or more missing impacts point to the same public lens mutation, page split, chronology revision, or canon candidate.
3. Sources whose public read has strong signature moments but no lens_points links.
4. Sources repeatedly referenced by source-light public lens pages.
5. Sources that add pressure to bloated pages that may need splitting.
6. Ordinary historical reads that likely require only evidence-only or link-existing-lens.

Do not choose by slug order alone. Interviews and classroom episodes should be mixed when they carry the same mechanism; otherwise the impact queue can become an interview-only accounting treadmill instead of a lens-building loop.

For backfill, prefer a useful compact decision over a long essay. The goal is to close the source's accounting and create clear downstream tasks only when the source actually pressures the lens.

Inputs

• content/lens/episodes/<source-slug>/read.json
• content/lens/evidence/videos/<source-slug>.semantic.json
• content/sources/videos/<source-slug>/metadata.json
• content/sources/videos/<source-slug>/transcripts/v1/transcript.clean.jsonl
• existing lens docs under website/src/content/docs/lens*.md
• generated link data from website/src/data/lens/link-index.json
• existing proposals, reviews, promotions, glossary, canon, and ledger files under content/

Output

Write one impact proposal:

content/workflow/proposals/<source-slug>/corpus-impact.json

Use this shape:

{
  "source_slug": "<source-slug>",
  "source_title": "<episode title>",
  "source_date": "YYYY-MM-DD",
  "source_ref_base": "video:<source-slug>@transcript:v1",
  "impact_level": "evidence-only",
  "summary": "One sentence on how this episode changes or confirms the corpus.",
  "existing_lens_links": [],
  "concept_extensions": [],
  "new_lens_seeds": [],
  "atlas_mutations": [],
  "chronology_notes": [],
  "canon_candidates": [],
  "next_actions": []
}

impact_level is the highest justified level:

• evidence-only
• link-existing-lens
• extend-existing-concept
• create-new-lens-seed
• morph-atlas-structure
• canon-candidate

Multiple arrays may be non-empty. The highest level says how strong the source impact is, not the only action allowed.

Pressure Checks

Before deciding impact_level, compare the public read and semantic bundle against:

• signature_moments in content/lens/evidence/videos/<source-slug>.semantic.json,
• existing lens points and public concept pages,
• topic aliases and near-duplicate topic clusters,
• recent corpus-impact proposals,
• the source date and any older/newer formulations of the same mechanism.

If a source has vivid signature moments that are absent from both episode marks and public lens surfaces, record them as unresolved pressure in next_actions or new_lens_seeds. Do not let a source disappear just because the episode page itself is readable.

Mutation Ladder

1. Evidence Only

Use when the episode adds useful refs or examples but does not change any public structure.

Record:

• which existing concepts it supports,
• exact refs,
• why no public mutation is needed.

2. Link Existing Lens

Use when the episode directly invokes an existing lens-point:*.

Record:

• target lens point,
• episode phrase or beat that should link,
• exact source refs.

Then use jiang-provenance-linker if the link should be applied immediately.

3. Extend Existing Concept

Use when the episode adds a distinct mechanism, example, contradiction, chronology note, or strong Jiang formulation to an existing concept page.

Record:

• target concept page,
• proposed addition,
• source refs,
• whether the addition is evidence, mechanism, example, contradiction, or chronology.

Then use jiang-lens-concept-writer for the actual page edit.

4. Create New Lens Seed

A single video can create a new lens seed if it introduces a reusable mechanism, not just a topic.

Threshold:

• the idea explains more than the episode itself,
• future episodes would naturally link back to it,
• the source includes a strong Jiang phrase, scene, definition, reversal, or causal model,
• it does not fit cleanly inside an existing concept without losing force.

Record:

• proposed slug and title,
• core mechanism,
• why it is reusable,
• what existing concepts it touches,
• exact refs,
• whether it should be public now as an early seed.

Then use jiang-lens-concept-writer if the seed should become a public page immediately.

5. Morph Atlas Structure

Use when the source shows that the current atlas nesting is misleading.

Examples:

• a new concept does not belong under any current family,
• two families should merge,
• one family should split,
• a concept belongs under a different chapter,
• an older source processed later changes the apparent foundation of a family.

Record:

• current atlas structure,
• proposed mutation,
• reason,
• exact refs,
• affected concept pages and lens points.

Then use jiang-lens-atlas-maintainer. For structural changes, run jiang-lens-judge before handoff.

6. Canon Candidate

Use when the source adds material that may belong in canon, glossary, or ledger.

Record the candidate, but do not promote it directly. Promotion requires jiang-canon-promotion with review and dated grounding.

Bias Control

Processing order is not intellectual order. Always reason from source date and source context, not from when the episode was processed.

If an older lecture is processed late, it may:

• become the earliest known formulation of a concept,
• change a concept chronology,
• reveal that a later source was an extension rather than an origin,
• force atlas language to become less biased toward newer processed episodes.

Do not protect the current atlas just because it already exists. Do not rewrite it casually either. Mutation should be justified by source-backed pressure.

Do Not

• Treat one episode topic as one lens by default.
• Wait for arbitrary batches before recording impact.
• Promote agent interpretation into canon.
• Create a public lens page for a mere topic label.
• Erase older positions when newer ones exist.
• Treat quoted literary material as Jiang's own claim unless Jiang's gloss supports it.

Validation

Run:

node ops/scripts/validate-corpus-impact.mjs content/workflow/proposals/<source-slug>/corpus-impact.json
node ops/scripts/compile-content.mjs
node ops/scripts/validate-content.mjs

To audit fresh actionable missing impact files in budget mode:

node ops/scripts/audit-corpus-impact.mjs

To inspect the full historical backlog only when the maintainer explicitly asks:

node ops/scripts/audit-corpus-impact.mjs --all-missing

For downstream website-visible mutations, the downstream skill should also run:

cd website && npm run build