Performance tiering and improving-evidence logic

DEC-0004 - Performance tiering and improving-evidence logic

Context

Need common interpretation across different exercise scoring types.

Decision

Performance tiers:

• Chua dat: <20%
• Can cai thien them: 20-59%
• Kha: 60-80%
• Gioi: 81-99%
• Xuat sac: 100%

Improvement evidence can use multiple tracks:

• Consecutive mock-test improvement.
• Week-over-week average improvement.
• Month-over-month average improvement.
• 3-month-cycle comparison.
• Recent 5-day/attempt trend without sharp drop.

Sharp-drop rule:

• Hybrid threshold (% + program-scale threshold), system uses stricter condition.
• Start with shared baseline for all programs, recalibrate per program after 4 weeks real usage data.

Rationale

Need consistent user-facing interpretation while respecting program-specific scales.