Habit-first dynamic recommendation and adaptive control policy

DEC-0042 - Habit-first dynamic recommendation and adaptive control policy

Context

Recommendation baseline existed but remained too static in count and did not specify user-control persistence, explainability label, or ignore-driven strategy adaptation.

Decision

Recommendation ordering:

• Priority is habit-first.
• Recommend easy-to-start continuation first, then improvement-focused items, then optional expansion choices.

Recommendation size:

• Dynamic set size from 3 to 7 items.

Explainability:

• Each recommendation item shows a one-line reason label.

Refresh behavior:

• Refresh recommendation set after each submission.
• Also support manual refresh by user action.

Adaptive behavior:

• If user ignores recommendations for 3 consecutive times, system changes strategy (for example difficulty/skill mix).

Diversity guardrail:

• Maximum 3 items from the same skill in one recommendation set.

User controls:

• Allow quick controls: skill, difficulty, duration.
• Manual control preference persists for current session only.

Cold-start policy:

• For no-goal/no-history users, source recommendations from trending content in recent 14 days with easy-start bias.

Inventory fallback:

• Use nearest-match fallback ladder before dropping quality.
• Keep program/skill relevance first, then relax in controlled order.

Decision Value

• Improves short-term action rate by reducing start friction.
• Keeps recommendation behavior explainable and tunable.
• Scales without overcomplicating core logic.

Rationale

A dynamic but bounded recommendation model offers better personalization while remaining understandable for users and maintainable for operations.