1Principles
ValuesBeliefsControls
Measure, don't opine
- 210 features tracked. Zero computed from tests.
- Manual L-levels drift within days of editing.
- FAVV Build Contracts now declare test artifacts — the data exists, the bridge doesn't.
| Problem | Question | Decision |
|---|---|---|
| States are opinions dressed as measurements | What would it take to compute every state from test output? | Build the parser, runner, and writer as one pipeline |
The Vision
A commissioning engineer reading gauges on a dairy factory control panel — not guessing, measuring.
2Performance
ValuesBeliefsControls
Zero false L3s
| Metric | Target | Now |
|---|---|---|
| False L3 rate | 0 | Unknown — no verification |
| Unmapped visibility | 100% flagged | 0% — hidden at L0 |
| CI overhead | ≤3 min added | N/A — no script |
| Regression detection | 100% caught | 0% — one-way ratchet |
| Problem | Question | Decision |
|---|---|---|
| A feature marked L3 that doesn't work destroys trust in the matrix | Is under-reporting better than over-reporting? | Yes — conservative: Safety Test violations cap at L2 |
The Vision
A green dashboard with one red cell — the Safety Test caught a counterfeit pass.
3Platform
ValuesBeliefsControls
Parser exists, bridge doesn't
| Layer | Built | Wired | Working |
|---|---|---|---|
| PRD FAVV tables | Yes (3 PRDs) | No | Partial |
| feature-matrix.json | Yes | Yes | Hand-edited |
| Vitest in CI | Yes | Yes | Not scoped to features |
| FAVV parser | Yes | Yes | Doesn't extract test mapping |
| Problem | Question | Decision |
|---|---|---|
| All the pieces exist but nothing connects them | What's the minimum new code to close the loop? | ~60% new code: index builder + level computer + writer |
The Vision
Two islands: PRD specs on the left, test files on the right. A broken bridge between them.
4Protocols
ValuesBeliefsControls
Ten days, four sprints
| Sprint | What | Effort | Unblocks |
|---|---|---|---|
| 0 | FAVV parser + index builder | 3 days | Know which tests map to which features |
| 1 | Test runner + L-level computer | 4 days | Computed states exist |
| 2 | Matrix writer + receipt + CLI | 2 days | States written, evidence logged |
| 3 | CI integration | 1 day | Automatic on every merge |
| Problem | Question | Decision |
|---|---|---|
| Manual commissioning doesn't scale past 50 features | Can we validate with Identity & Access features first? | Yes — Sprint 1 acceptance: computed states match manual for AUTH-* features |
The Vision
A factory production line with four stations, each producing a piece of the commissioning pipeline.
5Players
ValuesBeliefsControls
Commissioners verify, scripts compute
| Job | Struggling Moment | Hidden Objection |
|---|---|---|
| Commissioner verifies states | 30+ min manual inspection, usually skipped | "What if the script is wrong and I trusted it?" |
| Engineering agent updates matrix | Forgets to update JSON after merge | "What if CI takes too long?" |
| Problem | Question | Decision |
|---|---|---|
| L4 requires human judgment — script handles L0-L3 | Where does automation end and human judgment begin? | L4 = commissioner sign-off. Everything below = computed. |
The Vision
Two people: one running tests at a terminal, one walking through the live app with a checklist.
1 / 5