PWM roadmap
Phases gate on observable signals, not calendar dates. Calendar estimates are advisory; the only real trigger is the exit criterion. A phase is “done” when every exit signal is true. Source of truth: pwm-team/coordination/PWM_PRODUCT_VISION.md (private repo; this page mirrors the published phase tables).
Phase U1a — 2 anchors verified
CompleteCASSI + CACTI principles, specs, benchmarks, reference solvers, matcher, demos, walkthroughs all shipped.
| Exit signal | Target | |
|---|---|---|
| ✓ | CASSI L1/L2/L3 live on testnet | ✓ |
| ✓ | CACTI L1/L2/L3 live on testnet | ✓ |
| ✓ | Reference solvers (GAP-TV, PnP-ADMM) | running |
| ✓ | Faceted matcher on /match | ✓ |
| ✓ | /demos with InverseNet KAIST-10 + SCI-Video samples | ✓ |
| ✓ | /walkthroughs/{cassi,cacti} | ✓ |
| ✓ | /contribute page (six paths) | ✓ |
| ⏳ | G4 gate — ≥ 20 non-founder L4 submissions | pending UTSW cohort |
Phase U1a polish — make CASSI + CACTI work well
NowThe near-term focus. Solver tuning and real-data validation before any new anchors. Driven by NEAR_TERM_CHECKLIST.md.
| Exit signal | Target | |
|---|---|---|
| ⏳ | CASSI PSNR on KAIST-10 | ≥ 24 dB (current 14–19 dB) |
| ⏳ | CACTI PSNR on SCI Video | ≥ 24 dB (current 4–13 dB) |
| ⏳ | Real-data validation | ≥ 1 IRB-approved scene reproduced end-to-end |
| ⏳ | Solver runtime, 256×256, commodity CPU | < 60 s |
| ⏳ | Pre-mainnet UX bugs | all resolved |
| ⏳ | G4 gate (TWO_ANCHOR_MVP_LOCKED) | ≥ 20 non-founder L4 submissions |
Phase U1b — 3 more anchors + translator marketplace
NextOnce Phase U1a polish exits, expand the catalog and let third parties propose new specs/benchmarks via stake.
| Exit signal | Target | |
|---|---|---|
| · | MRI reconstruction anchor (L1-005) | live + ≥ 5 L4 submissions |
| · | CT reconstruction anchor (L1-006) | live + ≥ 5 L4 submissions |
| · | TEM anchor (L1-007) | live + ≥ 5 L4 submissions |
| · | Translator marketplace | ≥ 3 accepted non-founder spec/benchmark proposals |
| · | USD/PWM reference exchange rate | methodology published; first rate live |
Phase U2 — third-party platforms
Later3rd-party platform SDK + full contributor mining UI. Multiple branded platforms operating concurrently.
| Exit signal | Target | |
|---|---|---|
| · | 3rd-party platform SDK | v1.0 released, audited |
| · | Reference 3rd-party integration | ≥ 1 live (e.g., UTSW clinical workflow) |
| · | Contributor UI in explorer | /mine/{solution,benchmark,spec,principle} shipped |
| · | Independent 3rd-party platforms | ≥ 3 operating, ≥ 10K PWM/month settlement each |
| · | End-user volume | ≥ 100 unique users/platform/month |
| · | USD/PWM reference rate | updated daily, accepted by ≥ 2 platforms |
Phase U3 — mature ecosystem
LaterMulti-domain coverage; sustainable contributor economics; visible competition between platforms.
| Exit signal | Target | |
|---|---|---|
| · | Principles in catalog | ≥ 50 active beyond Genesis 500 |
| · | Domain coverage | medical imaging + materials science + structural mechanics + fluid dynamics |
| · | Top-decile L4 miners | sustainable income ≥ TBD USD/month |
| · | 3rd-party platform competition | visible price/UX/niche differentiation across ≥ 3 platforms |
Why signals, not dates
Research-pace projects can't hit calendar deadlines without either cutting corners or pretending. By gating on observable signals (PSNR ≥ 24 dB, ≥ 20 non-founder submissions, etc.), we keep the schedule honest: a phase is done when the work is done, not when the calendar says so. Failure modes (solver tuning plateaus, cohort doesn't mine, etc.) have explicit contingencies — see the vision doc.