EVT 2026-02-22 - Cross-stage UX Optimization Idea Pack (Draft Only, No Decision Promotion)

Session scope

Continue autonomous ideation for DOL English V2 with a single goal: deliver best-possible UX that is full enough to be useful, but lean enough to scale and operate long-term.

Working assumptions (explicit)

1. Lane is DOL English V2 (not DOL KID).
2. Existing confirmed decisions remain active unless explicitly replaced later.
3. This event is draft-only: no DEC promotion in this cycle.

Diverge - 12 draft ideas across 4 directions

Direction A - Frictionless Entry and Fast Activation

1. Pre-auth smart preview: allow browse + sample difficulty preview before login.
2. Intent-preserving auth handoff: user always returns to the exact interrupted context after login/signup.
3. One-tap quick-start set: offer short (5-10 min) exercises for first actions.

Direction B - Program-adaptive Learning Guidance

4. Program-aware recommendation mix: prioritize items aligned with user goal program or latest active program.
5. Assessment-form continuity: after each attempt, suggest nearest follow-up form first.
6. Adaptive starter ladder by account age (day 0-7): progressively increase depth while staying optional.

Direction C - Outcome Clarity and Motivation

7. Single primary reason label per recommendation (one-line explainability).
8. Progress without goal mode: show improvement trend even when target is missing.
9. Lightweight confidence indicator for recommendation quality (high/medium/low).

Direction D - Low-friction Monetization and Fair Entitlement

10. Paywall intent matching: show only relevant plans for the blocked feature.
11. Post-payment instant unlock in-place: no journey break after successful checkout.
12. Lock teaser quota: cap locked items in one recommendation set to avoid frustration.

Converge - recommended direction for next deepening

Top 1 (recommended): Direction B - Program-adaptive Learning Guidance

Why:

• Impact: strongest leverage on learning outcomes and daily habit quality.
• Feasibility: can be implemented with rule-based v1 before heavy ML.
• Differentiation: aligns with DOL philosophy and program-specific pedagogy.
• Risk: manageable if fallback ladder is explicit.

Optional Top 2: Direction A - Frictionless Entry and Fast Activation

Why:

• Conversion and first-week activation upside is immediate.
• Complements Direction B (good guidance is useful only if users enter quickly).

Concept draft (for Top 1)

Value proposition (1 sentence)

Give each learner a next step that feels immediately relevant to their program and current level, while keeping every suggestion optional and easy to act on.

Scope in

• Program-aware recommendation priority.
• Assessment-form continuity after submission.
• Optional week-1 adaptive starter for new users.
• Deterministic fallback ladder when inventory is weak.

Scope out

• Full ML ranking model.
• New scoring framework changes.
• New mandatory tasks or penalties.

Core flow (6 steps)

1. User finishes an attempt (or opens practice management/home recommendation area).
2. Engine resolves context: goal program, recent activity program, and account age window.
3. Engine generates candidate pool and applies availability + entitlement filtering.
4. Engine ranks candidates by continuity -> habit-fit -> freshness/diversity guardrails.
5. UI renders dynamic set with one-line reason per item and quick controls.
6. User action (start/swap/ignore) is logged to adapt next refresh.

Must / Should / Could

• Must:
  • Recommendations stay non-blocking and optional.
  • Fallback always produces actionable items.
  • Reason label is visible and concise.
• Should:
  • Week-1 starter adapts by program and recent submission.
  • Manual controls (skill/difficulty/duration) remain low-friction.
• Could:
  • Add lightweight confidence badge when signals are weak.

Key risks and mitigations

1. Risk: over-concentration in one skill/program.
   • Mitigation: enforce per-skill cap and freshness guardrails.
2. Risk: low inventory for a specific assessment form.
   • Mitigation: nearest-ladder fallback with transparent reason.
3. Risk: user perceives recommendation as repetitive.
   • Mitigation: ignore-streak adaptation after repeated skips.

Validation draft

Hypothesis H1

Program-adaptive ordering increases first-click-to-start rate vs non-adaptive ordering.

• Experiment E1.1: A/B test adaptive ordering vs baseline ordering on post-result surface.
• Pass: +8% relative lift in recommendation_start_rate with no drop in day-7 retention.
• Fail: <3% lift or retention drops >2%.

Hypothesis H2

Assessment-form continuity improves completion depth in first week.

• Experiment E2.1: A/B continuity-first vs neutral-mix for users with at least one submission.
• Pass: +10% lift in week1_second_submission_rate.
• Fail: lift <4%.

Hypothesis H3

One-line reason labels improve trust and reduce random skipping.

• Experiment E3.1: show reason label vs hide label.
• Pass: recommendation_ignore_rate decreases by >=6% and manual_refresh_rate does not spike.
• Fail: no ignore-rate improvement or refresh spikes >10%.

Open items carried forward (for next question cluster)

1. Program priority tie-breaker when user has mixed behavior across programs in same week.
2. Minimum candidate count policy under strict entitlement filter.
3. Confidence label exposure rule (always-on vs only low-confidence situations).