EVT 2026-02-22 - Stage D Recommendation Confidence and Primary-reason Priority Policy (Autocomplete)

Session scope

Continue autonomous hardening for recommendation quality by making confidence and explanation deterministic, understandable, and low-complexity.

Confirmed outcomes (auto-recommended defaults)

1. Recommendation confidence is standardized into 3 levels:

• high, medium, low.

2. Confidence is computed by deterministic signal coverage:

• high: strong contextual match (program/form/skill/intent) + availability,
• medium: partial match with controlled fallback,
• low: broad fallback (trending/nearest-ladder broad).

3. Reason label policy:

• show exactly 1 primary reason label per item,
• reason priority order:
  • recovery-critical,
  • goal-aligned,
  • habit-continuity,
  • freshness,
  • trending-fallback.

4. Low-confidence guardrail:

• max 1 low-confidence item per set when inventory is sufficient,
• low-confidence items placed near end of set.

5. Seed/rescue integrity:

• reactivation seed and rescue-pack items should prefer high/medium confidence only;
• if not possible, fallback with short notice.

Open items carried forward

• None for baseline lock.