Metaphor Systems

Warning

Starter clusters are provisional hypotheses. They become findings only when annotations, concordance outputs, analysis artifacts, support ratings, historical corroboration where claimed, and validation checks support them.

The shared starter vocabulary is stored in cases/x-case/config/shared-clusters.json. Case-local starter clusters are stored in each case’s config/case-clusters.json.

How Clusters Are Tested

Each cluster should be tested through annotation rather than assumed from the research question. A cluster gains support when repeated source spans show a stable pattern that contributes to one or more evidentiary-support dimensions: sacred object, sacrificial body, enemy as bringer of death, or historical enactment/alignment.

For textual-symbolic clusters, record:

  • source domain;
  • target domain;
  • entailments;
  • linguistic form;
  • political or moral work;
  • relevant support dimension;
  • Koenigsbergian interpretation, if warranted;
  • register or phase behavior;
  • absence or suppression pattern, where relevant.

For historical enactment/alignment clusters, record:

  • policy, order, decree, institutional practice, mobilization, implementation, social uptake, or outcome being linked;
  • whether the alignment is formal enactment, implementation, social uptake, or historical consequence;
  • whether corroboration comes from established secondary scholarship, additional primary evidence, or both.

Scores of 3 or 4 in textual-symbolic dimensions require repeated evidence. A single striking passage may be marked as high-salience evidence, but it should not by itself establish strong or very strong case-level support.

Revision Rules

Clusters should remain revisable throughout V1.

A cluster may be retained when it repeatedly explains textual evidence or historical alignment better than available alternatives. It may be merged when two labels describe the same mapping or support function. It may be split when one label hides different violence logics, enemy-as-bringer-of-death functions, historical alignments, or register effects. It may be rejected when it does not appear, depends on weak prompt effects, lacks corroboration where historical alignment is claimed, or fails rival-explanation review.

Cluster revision should preserve the distinction between:

  • textual evidence;
  • conceptual mapping;
  • Koenigsbergian interpretation;
  • historical corroboration;
  • support score.

Cross-Case Mapping

Shared clusters should preserve difference as well as similarity. A body, sacrifice, purification, or enemy as bringer of death cluster may perform restorative work in one case and exterminatory work in another. The cross-case map should therefore link cluster evidence without erasing historical and rhetorical specificity.

Cross-case mapping should show:

  • which support dimension the cluster informs;
  • whether support is textual, historical, or both;
  • whether similar language performs different political or historical work;
  • whether the cluster supports, complicates, limits, or challenges Koenigsberg’s Law.

Cluster changes should be reflected in case configuration, concordance indexes, analysis artifacts, support-rating artifacts, and synthesis notes.