YazSes v0.4 — Capability Prioritization Matrix
Scoring Rubric¶
Each capability is scored on four dimensions, each 1–5:
| Dimension | 1 | 3 | 5 |
|---|---|---|---|
| user_pain | Nice-to-have; workaround exists and is adequate | Frequent friction; workaround is suboptimal | Blocking; no workaround; users disengage |
| feasibility | Very hard; requires novel research or unvalidated tech | Achievable with significant engineering; key questions remain | Straightforward; technology validated; integration is the main work |
| novelty | Widely available in competing tools | Differentiated but has partial alternatives | Unique; no competing tool offers this combination |
| production_value | Prototype-only; breaks under real use | Works reliably for primary use case; some edge cases remain | Works reliably across all documented scenarios; directly enables daily-driver use |
total_score = user_pain + feasibility + novelty + production_value (max: 20)
Matrix¶
| cap_id | title | user_pain | feasibility | novelty | production_value | total_score | recommendation |
|---|---|---|---|---|---|---|---|
| cap-001 | Offline LLM Intent Routing Layer | 5 | 4 | 5 | 5 | 19 | MVP |
| cap-002 | Code-Aware Voice Dictation via LSP | 5 | 3 | 5 | 5 | 18 | MVP |
| cap-003 | EMG Silent Speech Backend | 4 | 4 | 5 | 4 | 17 | MVP |
All scores are tagged [HYPOTHESIS] unless noted otherwise — see CapabilityCard scoring rationale sections for evidence citations per dimension.
MVP Selection Rationale¶
The three capabilities above are all recommended for the v0.4 MVP. This selection is justified by three converging reasons:
Score separation. All three capabilities score 17–19/20, placing them in a distinct cluster. No other gap from the gap analysis (gap-004 through gap-009) scores above 16/20 on the same rubric (estimated: gap-004 ≈ 15, gap-008 ≈ 13, gap-009 ≈ 11). The gap between cap-003 (17) and the next-best gap is sufficient to exclude lower-priority gaps without risk of under-delivery. [HYPOTHESIS]
Strategic sequencing. cap-001 (LLM routing) is a prerequisite for several future capabilities (gap-004 soft gaze, gap-005 gaming protocol) and an enhancement for cap-002 (LSP context consumed by the SLM prompt). Building cap-001 first maximises the value of subsequent capabilities. cap-003 (EMG backend) is fully independent of cap-001 and cap-002 and can be developed in parallel, making it a natural parallel workstream for a two-developer team. [EVIDENCE gap analysis §Sequencing Constraints]
User population coverage. cap-001 benefits all YazSes users. cap-002 specifically targets the developer cohort — the highest-value users who are most likely to become evangelists and contributors. cap-003 opens two new user populations (open-office workers, accessibility users) who currently have no viable voice daemon option. Together, the three capabilities address every documented persona from the research scope without overlap or redundancy. [HYPOTHESIS]
Excluded from MVP:
- gap-004 (Soft Gaze via OS Accessibility): Depends on cap-001; deferred to v0.4.1 after cap-001 is proven stable. Score estimate: ~15/20.
- gap-005 (Gaming Voice Protocol): Niche use case relative to developer and office personas; deferred to v0.5. Score estimate: ~14/20.
- gap-006 (AAC Mode): High moral value; depends on cap-003 being proven first; deferred to v0.5. Score estimate: ~14/20.
- gap-007 (Dysarthric Speech / LoRA): Engineering cost is high and total addressable user count is lower; deferred pending additional SoA research. Score estimate: ~12/20.
- gap-008 (Ambient/Wake-Word Mode): Independent but lower urgency; can be added as a NEXT release item. Score estimate: ~13/20.
- gap-009 (Voice Error Correction Protocol): Low delta over existing rollback mechanism; DEFER. Score estimate: ~10/20.
Study: [[yazses-future-voice-hci/input/research_scope|yazses-future-voice-hci]]