Recommendation interaction, adaptation, and control contract
DomainsDOL EnglishProduct226 words1 min read
DEC-0045 - Recommendation interaction, adaptation, and control contract
Phần tiêu đề “DEC-0045 - Recommendation interaction, adaptation, and control contract”Context
Phần tiêu đề “Context”DEC-0042 locked habit-first baseline, but runtime interaction semantics (ignore, feedback, adaptation order, and control behavior) remained under-defined.
Decision
Phần tiêu đề “Decision”Recommendation execution is finalized as:
- Ignore definition includes:
- no click, and
- click then leave without a submission.
- Ignore streak is evaluated at end-of-session (
30midle or explicit session exit). - If ignore streak threshold is hit, adaptation order is:
- difficulty -> format -> skill.
- Feedback model is like/dislike at recommendation-cluster level.
- Cluster dislike persists for next clusters.
- Feedback interaction does not reset ignore streak.
- Dislike + leave without attempt still increments ignore by 1.
- Default recommendation count: 5.
- Low-inventory behavior: shrink to minimum 3.
- Default structure for 5 items:
- 2 habit,
- 2 target,
- 1 explore.
- Manual controls (
skill,difficulty,duration):- soft-priority ranking (not hard strict filter by default),
- no-match uses nearest-ladder relaxation with short notice,
- scope persists across pages in same session.
Decision Value
Phần tiêu đề “Decision Value”- Makes recommendation behavior predictable for users and measurable for product.
- Prevents silent ambiguity around ignore/feedback signals.
- Preserves personalization quality while avoiding empty-state dead ends.
Rationale
Phần tiêu đề “Rationale”Behavioral adaptation should be transparent and lightweight. This contract keeps logic simple enough to scale while retaining enough signal fidelity to improve relevance.