EVT 2026-02-22 - Stage D Recommendation Freshness Guardrails and Light Diversification (Autocomplete)

Session scope

Continue autonomous hardening for daily learning UX by reducing recommendation fatigue while preserving current habit-first baseline and simple scalable rules.

Confirmed outcomes (auto-recommended defaults)

1. Keep existing recommendation core unchanged:

• habit_first, dynamic set 3..7, one-line reason, adaptive-on-ignore.

2. Add freshness guardrails (lightweight):

• each rendered recommendation set should include at least 1 freshness item when inventory allows,
• freshness item definition: not attempted in last 14 days OR same skill with new format.

3. Add repetition guardrails:

• max 2 items per same topic in one set,
• if inventory is limited, relax via nearest-ladder and show short notice.

4. Add refresh-intent adaptation:

• if user manual-refreshes 2 consecutive times without click/attempt,
• next set forces at least 1 alternative-skill item.

5. Keep simplicity boundary:

• no heavy ML personalization requirement,
• deterministic rules only, with clear fallback path.

Open items carried forward

• No blocker for baseline lock; copy and visual treatment can iterate later.