propose_duplicate_merge_plan

OPT-IN, METERED second opinion: a server-side LLM reads a duplicate group (keys + translations + glossary + context) and PROPOSES a plan - recommendation merge | split | keep_separate + a well-formed merge_plan (null when keep_separate) + a one-line reason. PROPOSE-ONLY: it does NOT change status; apply via set_source_duplicate_merge_plan + apply_new_key_cluster / mark_cluster_applied / set_duplicate_status. You are already an LLM and decide locally at 0 credit - use this only as a tie-breaker or when the dev wants the AI's take. Costs 1 credit per generation UNLESS a valid cached proposal exists (cached=true → free); regenerate=true forces a fresh one. 402 when credits are insufficient. The persisted proposal also surfaces on list_source_duplicates (ai_proposed_plan / ai_diagnosis, free to read).

Parameters

Name	Type		Description
project_uuid	string	optional	Project UUID. Required when multiple projects are configured (SONENTA_PROJECTS); defaults to the single configured project otherwise.
source_value	string	required	The shared source value identifying the duplicate group.
namespace_slug	string	optional	Required only if the value spans namespaces (groups are intra-namespace).
regenerate	boolean	optional	Force a fresh diagnosis (1 credit) even if a valid cached one exists. Default false.
version_id	string	optional	Optional version UUID; defaults to the live version.

Example

{
  "tool": "propose_duplicate_merge_plan",
  "arguments": {
    "source_value": "<source_value>"
  }
}