Skip to contents

Ask the AI to propose a partition of the current leaves, or to converge

Source: R/theme_algorithm_v2.R
ai_propose_clustering.Rd

One AI call per pass. The AI sees ALL current leaves at once (full picture, no chunking). For pass 1, leaves are codes. For pass 2+, leaves are clusters from the prior pass, each rendered with its member codes visible so the AI can reason about content, not just cluster IDs.

Usage

ai_propose_clustering(
  leaves,
  pass_index,
  prior_history,
  codes,
  provider,
  research_focus = "",
  concept_str = "",
  reference_text = "",
  reflexivity_block = "",
  audit_log = NULL,
  response_cache = NULL,
  methodology_override = NULL,
  walk_state = NULL
)

Details

The AI returns either a partition of the leaves into new clusters or verdict='converged'. The schema (.clustering_schema()) forbids name/description fields during clustering – labeling is a separate pass (per C-tenet 5).

Returns the parsed proposal record:

  • verdict: "continue" or "converged"

  • cluster_assignments: list of clusters (NULL if converged)

  • overall_rationale: AI's top-level explanation

Records the call in walk_state and the audit log.