STATE - Progress Metrics

Current truth

Progress uses tiered interpretation, multi-window trend evidence, light-touch decline intervention, and 3-level learner outcome verdicting. Diligence is a weighted hybrid score with anti-idle active-time counting.

Rules

• Tier bands:
  • <20: chua dat
  • 20-59: can cai thien them
  • 60-79: kha
  • 80-99: gioi
  • 100: xuat sac
• For scaled tests, normalize by test max to percentage before applying tier band.
• Goal comparison is program + assessment-form + skill aware:
  • every compared result must include canonical assessment_form_id,
  • each goal carries goal_scale_profile_id,
  • each submitted result carries attempt_score_profile_id,
  • comparison mode is direct | normalized | not_comparable by configured profile map.
• Scale mapping guardrail:
  • normalized is allowed only by explicit approved mapping table,
  • no generic auto-convert-to-percent fallback.
• Goal versioning for comparison:
  • each result stores goal_version_id_at_submission,
  • historical comparison uses version-at-submission.
• Numeric goal-gap confidence gate:
  • show numeric distance-to-goal only when comparable attempts in latest 30 active days >= 3.
• If comparison mode is not_comparable:
  • hide numeric distance-to-goal,
  • keep trend/diligence/progress signals visible with short explanation.
• If sample gate is not met:
  • hide numeric distance-to-goal,
  • keep trend/diligence/progress signals visible with short explanation.
• Objective unanswered items are counted as wrong.
• Improvement evidence can come from mock, weekly/monthly/3-month trends, and recent consistency trend.
• Decline detection (OR logic):
  • 2 consecutive mock tests with score drop, or
  • latest 7-day average drops >=10% versus previous 7-day average.
• If decline is detected, system recommends a “3-bai cuu nhip” pack:
  • easy warm-up,
  • medium stabilization,
  • weakness-targeted practice.
• Rescue-pack touchpoints after decline:
  • immediate prompt on detection,
  • persistent availability in Practice Management.
• Rescue-pack quality and prompt-pressure guardrails:
  • composition: 1 fixed warm-up + 2 personalized exercises,
  • immediate prompt cooldown: 7 days,
  • immediate surface: non-blocking inline card/panel.
• New decline-cycle guardrails:
  • decline signal is considered “new” only after at least one recovery period,
  • if decline is detected during active cooldown, still respect cooldown,
  • cooldown resets when learner completes at least one similar exercise.
• Recovery and matching execution details:
  • similar exercise uses fallback ladder (A -> B -> C),
  • recovery validity uses practice time + completion ratio per exercise as soft thresholds,
  • for Writing/Speaking, cooldown reset only after scoring completed,
  • cooldown reset scope is by skill within program.
• Simplicity and scale guardrails:
  • practice-time threshold is defined by skill,
  • completion-ratio threshold is defined by exercise family (objective vs subjective),
  • no per-program/per-level branching in core recovery rule.
• Numeric defaults (V2 baseline):
  • practice-time thresholds:
    • Reading: 8 minutes,
    • Listening: 8 minutes,
    • Writing: 12 minutes,
    • Speaking: 12 minutes.
  • completion-ratio thresholds:
    • objective: >= 70%,
    • subjective: >= 60%.
  • tier-C fallback cutoff:
    • allow only when combined A+B inventory < 3.
• Reset evaluation timing:
  • evaluate only after result is available,
  • async-scored exercises wait for scoring completion.
• Transparency:
  • show concise user-facing checklist for reset conditions.
• Threshold review cadence:
  • freeze defaults for first 3 months,
  • then review monthly.
• Recovery status visibility:
  • show provisional recovery status after a qualifying result,
  • confirm official recovery in weekly pulse.
• Escalation after non-recovery:
  • if no recovery across 2 consecutive cycles, trigger AI Tutor personalized 7-day plan,
  • intervention remains optional and non-blocking.
• AI Tutor 7-day plan operating model:
  • daily structure: 1 required exercise + 1 optional suggested exercise,
  • regenerate next plan only after 7-day cycle if learner is still high attention.
• AI Tutor plan sourcing and cycle stability:
  • required slot prioritizes weakest skill in current program,
  • low-inventory handling reuses A -> B -> C ladder in current program,
  • mid-cycle program change does not rewrite active plan; update applies next cycle.
• AI Tutor daily execution semantics:
  • missed-day required is not carried forward,
  • required completion uses submission-based counting,
  • optional item remains bonus-only and does not alter recovery/exit decision rules.
• High-attention threshold:
  • mark as high attention after 2 consecutive cycles without recovery.
• High-attention exit:
  • exit only after one official recovery confirmation in weekly pulse.
• Tier-C guardrails:
  • allow only when A/B inventory below minimum threshold,
  • no cross-program fallback.
• Intervention is light-touch:
  • recommendation/encouragement only,
  • no forced gate or hard block.
• If rescue pack is skipped, re-prompt only on a new decline signal.
• Verdict cadence:
  • weekly tracking pulse,
  • monthly official outcome verdict.
• Learner-facing outcome levels:
  • can co gang,
  • dang on,
  • dang tien bo tot.
• Learners with goal:
  • evaluate by trend + distance to goal,
  • monthly verdict does not rely on one attempt only.
• Learners without goal:
  • evaluate by trend + diligence,
  • completed-count only is not enough.
• Diligence score:
  • 40% active days
  • 40% completed exercises
  • 20% active time
• Active-time auto-pause at 90s inactivity; auto-resume on first interaction.
• All submitted exercises are counted into learning metrics regardless of account tier.
• KPI minimum set:
  • primary: habit retention (>=4 study days/week in 4 consecutive weeks),
  • secondary growth: locked-feature upgrade conversion,
  • secondary quality: recommendation-to-attempt rate.
• First-week activation leading indicators (new-user onboarding quality):
  • activation_first_attempt_24h,
  • activation_second_attempt_48h,
  • activation_week1_active_days_ge3,
  • activation_card_cta_click_rate,
  • activation_week1_completion_rate.
• Activation diagnostics by source/continuity:
  • activation_program_resolution_source_distribution,
  • activation_week1_completion_by_program_source,
  • activation_step2_same_program_continuity_rate.
• First-week activation is a leading indicator only:
  • does not replace primary KPI priority (habit_retention),
  • used for early-friction diagnosis in first 7 days after signup.
• KPI denominator/eligibility normalization:
  • habit retention uses rolling 4-week valid completed-day signals only,
  • recommendation-to-attempt only counts recommendation items rendered and visible,
  • locked-feature upgrade conversion measures locked-entry exposure to checkout success within attribution window.
• KPI operation cadence:
  • daily D-1 monitor,
  • weekly KPI review,
  • monthly KPI policy review.
• KPI default guardrails:
  • habit retention drop >5pp WoW -> investigation,
  • recommendation-to-attempt drop >8pp WoW -> recommendation audit,
  • locked-feature upgrade conversion drop >10pp WoW -> paywall/offer audit.
• KPI escalation:
  • 1 breach week -> diagnosis note,
  • 2 consecutive breach weeks -> experiment ticket,
  • 3 consecutive breach weeks -> mandatory policy review.
• KPI mandatory cuts:
  • program,
  • tier,
  • cold-start vs returning users.
• Streak attribution:
  • day boundary uses user local timezone,
  • completion uses submit timestamp,
  • single freeze auto-consumes on first missed day.

Decision trace

• DEC-0001
• DEC-0002
• DEC-0004
• DEC-0007
• DEC-0014
• DEC-0015
• DEC-0016
• DEC-0017
• DEC-0018
• DEC-0019
• DEC-0020
• DEC-0021
• DEC-0022
• DEC-0023
• DEC-0024
• DEC-0025
• DEC-0026
• DEC-0035
• DEC-0051
• DEC-0054
• DEC-0066
• DEC-0071
• DEC-0072
• DEC-0073
• DEC-0079
• DEC-0080