# Grid Health — Project Memory

**Client:** Grid Health (healthcare AI governance platform)
**Founders:** Yechiel Engelhard, Gil Shimon
**Page:** https://curiousendeavor.com/grid/
**Channel:** #grid-health (1474042909161558159)

---

## Project Status
- **Current Phase:** 1 (Context) — partially complete
- **Next:** Phase 2 (Landscape) → 3 (Audience) → 4 (Positioning)
- **Methodology:** `/home/clawd/workspace/memory/mission-methodology.md`

---

## Julia's Research (Feb 19, 2026)

### The Core Problem (Validated)
**Grid's claim:** "Health systems face unprecedented pressure to adopt AI but can't evaluate it safely"

**Evidence:**
- Major health systems receiving 50-200+ AI vendor pitches annually
- CMIOs report "AI vendor fatigue" and "evaluation overwhelm"
- <5% of ~600 major US health systems have formal AI governance programs
- Healthcare AI spending: $15B (2023) → projected $102B (2028)

### Documented Failures
1. **IBM Watson Oncology** — $62M investment, discontinued. Worked in trials, failed in practice.
2. **Epic Sepsis Tool** — 67% false positive rate at some institutions
3. **Radiology AI Variance** — Same tool: 94% accuracy at one hospital, 76% at another

### Why "Local Context" Isn't Marketing Fluff
Performance varies because of:
1. **Data/Population Differences** — Diabetes AI trained on urban fails in rural
2. **Workflow Mismatches** — AI designed for different EHR workflows, staffing models
3. **Infrastructure Gaps** — Cloud AI fails at hospitals with bandwidth limits

**Pattern:** 30-50% performance degradation common from trial to real-world deployment.

### Competitive Landscape
| Player | What They Do | What They Miss |
|--------|--------------|----------------|
| Big 4 Consulting | High-level AI strategy | Hands-on technical validation |
| EHR Vendors (Epic, Cerner) | AI marketplace within ecosystems | Cross-platform, independent assessment |
| Academic Medical Centers | Internal AI review committees | Scalable processes for community hospitals |
| Quality/Risk Vendors | Post-deployment monitoring | Pre-deployment evaluation |

**Grid's White Space:** Pre-deployment technical validation with local context testing. Nobody owns this.

---

## CE Positioning Work

### The Hiring Metaphor (strong)
> "Evaluating an AI tool is like hiring a person. Generic vendor claims are like waving around grades from a university you've never heard of — they tell you almost nothing about whether this candidate will actually perform in your environment."

### Old Story vs New Story
**Old:** AI governance = compliance checkbox. Procurement evaluates vendors. IT handles integration. Clinical leadership signs off. Spreadsheets track everything. Takes months, nobody knows if it works on YOUR patients.

**New:** AI governance = operational infrastructure. The question isn't "does this AI work?" but "does this AI work HERE, on OUR data, in OUR workflows, for OUR patients?" Grid is that system.

---

## Files
- `/home/clawd/workspace/public/grid/index.html` — live page
- Marketing doc: Discord attachment (Marketing_-_GRID_Health.docx)

---

---

## Session: Feb 20, 2026

### Status
- Phases 1-7 built on page: https://curiousendeavor.com/grid/
- Phase 7 (Visual) has governance infrastructure refs (Linear, Vercel, Palantir, Stripe Atlas, Plaid) with screenshots
- BUT: Visual research is surface-level — missing positioning intelligence layer

### Gap Identified
Two distinct parts of methodology not fully developed:
1. **Visual Research** — Finding/analyzing references (input)
2. **Visual Direction** — Translating to actionable brand direction (output)

Grid visual work missing:
- Positioning DNA extraction (strategy, market contrast, test question)
- Strategic WHY behind each reference
- Connection to Grid's positioning, not just "they use dark colors"

### Next When Resuming
- Apply improved visual methodology (from other thread)
- Re-do Phase 7 with proper positioning intelligence
- Then Phase 8 Synthesis

### Screenshots Solution
Playwright is free + unlimited:
```bash
playwright screenshot --browser chromium "https://example.com" output.png
```

*Last updated: Feb 20, 2026*