1. Role / Context
You are a junior Research Assistant. The Growth Manager needs a one-page brief on what may be driving low NovaFit D7 retention and what to validate next week.
Manual First Gate
source table first → AI candidate themes → human source verification → brief structure assist → human final draft → AI gap check → human correction log.
Do not:AI writes the final brief first; AI invents sources; AI chooses the strategy; AI certifies the result.
2. Work Order
Use five provided sources: user comments, support categories, activation metrics, competitor notes, and internal meeting quotes.
Boundary: you prepare evidence and validation questions; manager owns product/growth decision.
3. 4-hour Workload Map
| Time | Work |
|---|---|
| 20m | Work order and output contract |
| 45m | Read and label 5 sources |
| 40m | Build evidence table |
| 35m | AI candidate theme extraction |
| 45m | Human verification |
| 40m | Write one-page brief |
| 15m | AI QA + handoff |
4. Source / Input Packet
User comments
"I do not know what to open on day two" / "AI suggestions are too generic" / "I do not know how to fill in goal settings"
Support categories
onboarding confusion 9 / notification annoyance 6 / unclear value proposition 5
Metrics
D7 retention 14%; onboarding complete 48%; first habit plan 33%
Competitor notes
A has 7-day challenge; B calendar; C AI personalization repeats.
Internal quotes
Growth: drops after day two. Support: many ask how to use first week.
5. Skill Level Mapping
original_l4_concept: input/output definition simulation_action: define manager question, source inputs, evidence table, and one-page brief output contract student-visible artifact: work order + source table + research brief
6. AI Windows / AI Throttle
Window 1 — candidate theme extraction
max_uses: 1 forbidden: final conclusion / invented sources / ranking without evidence required_human_verification: accept or reject each theme with source_id
Window 2 — brief structure compression
max_uses: 1 forbidden: add new claims / remove uncertainty / choose strategy
Window 3 — final gap check
max_uses: 1 forbidden: approve truth / certify final work
7. Human Verification
Claim type
Unsupported AI claim
Manager decision
L5 boundary
8. Worked Example
Weak AI-first
“NovaFit should add calendar integration because users need reminders.”
Corrected
“Calendar integration is a validation hypothesis. Evidence: one user asked for schedule integration and Competitor B uses calendar reminders; this does not prove it will fix retention.”
9. Task Cards
Task 1 Work order
Task 2 Evidence row
Task 3 Theme
Task 4 Finding
Task 5 AI log
Task 6 Handoff
10. Quick Checks
11. Guided Practice
Rewrite one AI over-inference as an evidence-backed finding.
12. Independent Practice
13. Common Mistakes
- AI writes brief before source table.
- Hypothesis becomes fact.
- Evidence IDs disappear.
- Junior makes strategy decision.
14. Rubric / Redlines
| Pass | Redline |
|---|---|
| source-backed findings + caveats + AI log | AI-first final brief / invented source / strategy decision |
15. Artifact Builder
Student artifact body: exported Markdown. How this is used in real work: Give the manager one page of research input, without making the strategy decision for them.How this is used in a portfolio: Show source discipline + AI verification.Boundary: The structure check confirms completion only; quality still needs human review.
16. Handoff
D62 will reuse source discipline to synthesize raw user feedback into a theme table.
17. L5 Bridge Marker
This proves bounded junior research readiness under review. It is not job-ready certification or L5 mastery.