Convergence Profile Migration Guide

This guide explains how to move safely between strict, balanced, and robust convergence profiles.

Profile Contract

Profile	Goal	Typical use	Risk posture
strict	deterministic diagnostics, bounded recovery	CI reproducibility, bug triage, regression isolation	lowest tolerance for implicit recovery
balanced	broad robustness with bounded cost	default program gate (Gate B/C)	moderate adaptive behavior
robust	maximize convergence on hard cases	deep stress debugging and extreme scenarios	highest adaptive tolerance

Migration Path

1. Start in strict and ensure baseline reproducibility and typed telemetry coverage.
2. Move to balanced only after Gate A is stable.
3. Enable Gate B budgets (--phase-key gate_b) and require stable-class non-regression.
4. Move to robust only for targeted stress classes and keep artifacts isolated.

Python API Snippet

import pulsim as ps

opts = ps.SimulationOptions()
opts.fallback_policy.convergence_profile = ps.ConvergenceProfile.Balanced
opts.fallback_policy.policy_dry_run = True
opts.fallback_policy.anti_overfit_check = True
opts.fallback_policy.anti_overfit_stable_budget = 0

Gate Expectations by Profile

• strict
• prioritize deterministic failures over aggressive recovery
• require complete typed diagnostics for each fallback event
• balanced
• require Gate B pass with stable-class guard metrics
• allow bounded policy recommendations and dry-run comparison
• robust
• allowed for targeted stress tracks
• must still satisfy cross-class non-regression budgets before promotion

Rollback Guidance

Rollback immediately when any of these occur:

• policy_stable_mismatch_rate regresses over budget
• policy_stable_anti_overfit_violation_rate regresses over budget
• typed_convergence_schema_coverage_rate drops below baseline budget

Recommended rollback sequence:

1. revert profile to strict
2. capture fresh artifacts (benchmark_runner, stress_suite, kpi_gate)
3. inspect fallback trace deltas before re-enabling balanced