Convergence Check

convergence-check

Goal:

• Evaluate whether a hypothesis newly entered the current top-k set and update the convergence counter deterministically.

Inputs:

• hypothesis_id
• previous_top_k_ids
• current_top_k_ids
• current convergence count
• caller-owned state/EVOLUTION_STATE.json

Outputs:

• ConvergenceCheckResult
• updated convergence count
• when consumed by the evolution loop, updated state/EVOLUTION_STATE.json

Context Loading:

• Open skills/shared-references/schema-index.md.
• Read packages/agent_contracts/pipeline_control.py and confirm the exact EvolutionStateContract shape before writing state/EVOLUTION_STATE.json.
• Treat the top-k sets as caller-supplied frontier inputs. This skill only evaluates the rule and updates the counter.

Execution Contract:

• This skill is deterministic and must not call an LLM.
• Use from tools import evaluate_convergence as the stable invocation surface.
• The exported helper is implemented in packages/agent_mechanics/convergence_check.py.
• The helper signature is evaluate_convergence(hypothesis_id, previous_top_k_ids, current_top_k_ids, current_convergence_count) -> ConvergenceCheckResult.

Execution Steps:

1. Open skills/shared-references/schema-index.md, then read packages/agent_contracts/pipeline_control.py before writing state/EVOLUTION_STATE.json.
2. Read the candidate hypothesis_id, the previous and current top-k sets, and the current convergence count.
3. Call tools.evaluate_convergence(hypothesis_id, previous_top_k_ids, current_top_k_ids, current_convergence_count).
4. Return the ConvergenceCheckResult to the caller.
5. When used by the evolution loop, persist the returned entered_top_k and convergenceCount values into state/EVOLUTION_STATE.json.
6. Validate any updated state/EVOLUTION_STATE.json artifact before declaring completion.

Artifact Rules:

• The convergence rule is fixed: entering the top-k frontier resets the counter to zero; otherwise the counter increments by one.
• Do not fold additional stopping logic into this skill. Stop decisions belong to evolution state management and completion verification.

Completion Rule:

• This skill is complete only when the deterministic result has been produced and any caller-owned state/EVOLUTION_STATE.json update matches that result exactly.