prioritization-frameworkdraft
Prioritization: Curator MVP / pilot scope
Applicability Filter Summary
- Ran: ICE (estimates feasible at concept stage), MoSCoW (clear pilot scoping).
- Excluded: RICE — Reach in users/qtr is unknown at concept stage (would be fabricated); revisit post-pilot. Kano — no customer survey yet. Weighted Scoring — ICE suffices.
Items derive from the opportunity tree (07). No real reach/effort data; ICE scores are judgment-based estimates flagged Low confidence. Items:
- A reproducible analysis pipeline (1A)
- B evidence-linked conclusions (1B)
- C editable draft summary (2A)
- D transcript + speaker separation (2B)
- E client dossier (3A)
- F commitment + micro-tracking (3B)
- G consent + depersonalisation (4A)
- H out-of-scope alerting (4B)
- I live in-session suggestions (parking lot)
Per-Framework Scoring
ICE (Impact × Confidence × Ease, 1–10; estimates)
| Item |
Impact |
Confidence |
Ease |
ICE |
Notes |
| G consent/depersonalisation |
9 |
7 |
6 |
378 |
Gating prereq for any external LLM |
| D transcript+diarization |
8 |
8 |
7 |
448 |
Integrate, not invent |
| C editable draft summary |
8 |
6 |
7 |
336 |
Cheap, high felt value |
| A reproducible pipeline |
10 |
4 |
3 |
120 |
The moat, but riskiest/hardest |
| B evidence quotes |
8 |
6 |
6 |
288 |
Boosts trust, moderate effort |
| E client dossier |
7 |
5 |
4 |
140 |
High value, larger build |
| F commitment/micro-tracking |
6 |
5 |
5 |
150 |
Depends on client response |
| H out-of-scope alerting |
7 |
6 |
5 |
210 |
Safety credibility |
| I live suggestions |
5 |
3 |
3 |
45 |
Sensitive, defer |
MoSCoW (pilot)
| Item |
Bucket |
Rationale |
Risk if dropped |
| G consent/depersonalisation |
Must |
Legal/trust gate |
Cannot run safely |
| D transcript+diarization |
Must |
Input to everything |
No analysis |
| A reproducible pipeline |
Must |
The differentiator |
Becomes "just a transcriber" |
| B evidence quotes |
Must |
Trust gate to send |
Output not sendable |
| C editable draft summary |
Should |
Core time-saving |
Less adoption pull |
| E client dossier |
Should |
Continuity value |
Weaker retention |
| H out-of-scope alerting |
Should |
Safety story |
Ethical exposure |
| F commitment/micro-tracking |
Could |
Adds granularity |
Acceptable to defer |
| I live suggestions |
Won't (pilot) |
Sensitive, unproven |
None now |
Per-Framework Ranking Output
- ICE ranking (high→low): D, G, C, B, H, F, E, A, I.
- MoSCoW Musts: G, D, A, B.
Cross-Framework Comparison
| Item |
ICE rank |
MoSCoW |
Agreement |
| D |
1 |
Must |
Strong |
| G |
2 |
Must |
Strong |
| A |
8 |
Must |
Divergent — ICE penalizes A on low confidence/ease, but it is the strategic moat, so MoSCoW forces it to Must. Driving dimension: strategic necessity vs. near-term ease. |
| C |
3 |
Should |
Mild |
Executive Summary with Recommendation
Fund the Must core that makes one trustworthy session loop real: transcript+diarization (D), consent/depersonalisation (G), the reproducible pipeline (A), and evidence-linked conclusions (B). Add the editable draft (C) as the adoption sweetener. Defer dossier depth (E), micro-tracking (F), and especially live suggestions (I). The key divergence is A: it scores low on ICE because it's hard and unproven, but dropping it guts the product — so it stays Must and its risk is bought down via the reproducibility experiment, not by deprioritizing it.
Sensitivity / What Changes the Ranking
- If the reproducibility experiment (A) fails, the whole MVP thesis flips → pivot review (stage 25), not a reshuffle.
- If transcription quality (D) is poor for the target language, D effort rises and may need a different vendor.
Recommendations (Sequencing)
- Fund now: D, G, A, B (+ C).
- Defer: E, F, H-depth.
- Drop (pilot): I.
- Data that would change this: real reach/WTP (enables RICE), reproducibility experiment result.
Limitations and Biases
- ICE scores are estimates, not measured — anchoring risk.
- No RICE/Kano → reach and delight are unmodeled.
- Strategic-necessity override on A is a judgment call, made explicit on purpose.