Add deterministic and probabilistic policy input guidance

[cdcavell]

Summary

Adds the Milestone 8 tutorial for deterministic and probabilistic inputs in policy evaluation.

This change introduces an advanced governance tutorial that explains how authoritative host facts and probabilistic or model-derived signals can participate in the same policy pipeline without disguising uncertainty as fact or allowing model output to become execution authority.

The tutorial covers:

• Deterministic host-resolved policy facts.
• Probabilistic and model-derived signals.
• Confidence scores, class probabilities, anomaly scores, and risk estimates.
• The distinction between confidence, probability, and authorization.
• Calibration and the meaning of probabilistic scores.
• Authoritative context versus advisory context.
• Signal provenance, source identity, model identity, and model version.
• Signal timestamps, validity windows, and freshness.
• Thresholds as explicit policy rather than objective truth.
• Threshold versioning and operation-specific thresholds.
• Regional and tenant-specific threshold policies.
• Deterministic policy decisions constructed from captured probabilistic observations.
• Repeatability limits when models or upstream data change.
• Model drift, data drift, and changed model versions.
• Combining multiple signals without hidden or unjustified averaging.
• Unknown, unavailable, stale, and invalid signal states.
• Explicit fail-open and fail-closed behavior.
• Human review for uncertain or high-consequence cases.
• Re-evaluation when probabilistic evidence expires.
• Preservation of exact observed signals in decision provenance.
• Threshold-boundary and model-version testing.
• Separation between model output, policy input, governance decision, scoped authority, and host-owned execution.
• Cases where a simpler deterministic rule remains the better architecture.

The change also:

• Adds the tutorial to Governance navigation.
• Cross-links it from AI Integration navigation.
• Updates the Governance and AI Integration landing pages.
• Links the new material from policy-context, risk-based decision, and typed AI proposal guidance.
• Marks the Milestone 8 Deterministic versus probabilistic policy inputs roadmap item complete under issue [Learning]: Explain deterministic and probabilistic policy inputs #115.

The core architectural boundary remains:

A probabilistic signal may inform policy, but uncertainty should not be disguised as an authoritative fact.

Model-derived evidence can influence a governance decision, while policy interpretation, scoped authority, and protected execution remain explicitly host-controlled.

Closes#115