Release v4.2.0: Canonical governance loop, expanded HealingIntent, DPT thresholds, and backward‑compatible event model

[petter2025us]

This release introduces a mathematically rigorous governance loop, significantly expands the HealingIntent model, adds Deterministic Probability Thresholds (DPT), and makes the event model fully backward‑compatible.

🔍 What’s new

• Canonical governance loop (governance_loop.py) – a single entry point that orchestrates policy evaluation, cost estimation, reliability signals, and Bayesian risk scoring into a comprehensive HealingIntent.
• Expanded HealingIntent (v2.1.0) – adds 30+ optional fields for:
  • Bayesian uncertainty (posterior_mean, credible_interval, epistemic_uncertainty)
  • Evidence provenance (evidence_lift, contradiction_score, evidence_sources)
  • Human factors (ambiguity_score, operator_attention_required, explanation_depth)
  • Learning hooks (feedback_expected, replay_key, experiment_tags)
  • Full backward compatibility – old intents still deserialize correctly.
• Deterministic Probability Thresholds (DPT) – new constants DPT_LOW = 0.2 and DPT_HIGH = 0.8 for transparent approve/deny/escalate decisions.
• Backward‑compatible event model – ReliabilityEvent now accepts legacy fields (component, latency_p99, error_rate, etc.) and maps them into the canonical metrics dictionary. Existing code that reads event.latency_p99 continues to work via property aliases.
• Metric resolver – a new helper resolve_metric(event, name) safely retrieves numeric values even when the attribute is a property, eliminating type errors in comparisons and formatting.
• Improved test coverage – overall coverage increased to 50%, with critical paths (governance loop, healing intent serialization) now at >80%.
• New requirements.txt – lists all core dependencies for easy installation.
• Citation file (CITATION.cff) added for academic use.

🧪 Testing

All 71 tests pass (including previously failing ones). The CI pipeline is green.

📦 Version

Bump to 4.2.0.

📝 Documentation

Updated README.md with a “What’s New” section and added CHANGELOG.md.

🔗 Related

• Closes #... (if any issues)
• Prepares for integration with the research probe (epistemic risk signals)