Corpus and Sources

Warning

This page documents the provisional corpus-selection and source-promotion process. Candidate sources are not approved corpus records until they pass source discovery, rights review, metadata validation, case-fit review, and support-dimension relevance review.

Structured starter candidates live in config/starter-corpus-candidates.json. Those candidates are a discovery queue, not a corpus.

Promotion Path

A candidate source becomes an approved corpus document through this sequence:

  1. Confirm the source edition, archive, URL, or bibliographic citation.
  2. Review rights and decide whether the raw text may be committed, stored locally, referenced metadata-only, or marked unavailable.
  3. Record document metadata in cases/<case>/metadata/document-manifest.json.
  4. Record source provenance in cases/<case>/metadata/source-registry.json.
  5. Add a verification block to the document manifest entry, including a minimum word count and anchor phrases when feasible.
  6. Acquire permitted raw text at the expected path under cases/<case>/corpus/raw/ using scripts/fetch-corpus.py, a source-specific download skill, or documented manual placement.
  7. Run python3 scripts/verify-corpus.py --case <case-id> and resolve any failures before normalization.
  8. Run the case pipeline to normalize and segment the document.
  9. Review the document’s register, phase, analytical priority, support-dimension relevance, and risk flags before annotation.
  10. Record whether the document may contribute to document-level support scores, historical enactment/alignment, or both.

Inclusion Criteria

Documents should be included when they contribute to at least one of these needs:

  • balanced-core coverage for a case phase or register;
  • evidence for one or more support dimensions: sacred object, sacrificial body, enemy as bringer of death, or historical enactment/alignment;
  • close reading of a high-priority sacrificial, sacred-object, enemy as bringer of death, or historical-alignment pattern;
  • negative evidence or absence analysis;
  • comparison with another case;
  • testing of a starter cluster or rival explanation;
  • historical corroboration of a textual-symbolic pattern through policy, implementation, social uptake, or durable outcome.

Documents should be deferred when provenance, rights, translation, OCR quality, or authorship confidence cannot yet be documented well enough for audit.

Support-Dimension Review

Before promotion, each candidate should receive a brief support-dimension note. The note should identify whether the document is expected to contribute to:

Dimension Corpus Relevance Question
Sacred object Does the document represent a collective object as transcendent, ultimate, sacred, immortal, or worth dying/killing for?
Sacrificial body Does the document represent bodies, suffering, killing, or dying as meaningful for the collective object?
Enemy as bringer of death Does the document represent an enemy as the agent, carrier, embodiment, or sign of death, doubt, mortality, dissolution, unreality, or collective destruction?
Historical enactment/alignment Does the document directly document or help corroborate policy, implementation, social uptake, mobilization, killing, exclusion, purification, or historical outcome?

The note is not a finding. It is a selection rationale that helps explain why a document belongs in the balanced core, extended corpus, historical corroboration set, or discovery queue.

Documents that directly express policy, orders, decrees, institutional programs, or implementation may serve two roles: textual-symbolic evidence and historical enactment/alignment evidence. When that happens, the two roles should be marked separately to avoid circular reasoning.

Corpus Status

This page is a discovery and promotion guide, not the live corpus status page. Promoted corpus records live in the case manifests and source registries. For the current project snapshot, see CURRENT_STATUS.md.

The pipeline can still produce stub artifacts when annotation inputs are empty, but those outputs are readiness indicators rather than analysis. Completed corpus claims require approved document records, traceable annotations, support-dimension scores where applicable, and historical corroboration where claimed.