Home · Applications · Production readiness
Long-tail guide · claim-safe
Production readiness checklist
Checklist is guidance, not a certification. Passing it is necessary but not sufficient for your plant.
| # | Gate | Why |
| 1 | Strictly positive inputs (or safe offset) | Log-space identity |
| 2 | x* policy documented (frozen vs adaptive) | Wrong normal moves the mean |
| 3 | Shadow run vs EMA on your data | Locked packs ≠ your plant |
| 4 | Delay audit if any command path | Multi-minute delay is a red light |
| 5 | Alarms/detectors not forced through GSRF residual | Fixed-FA residual is a non-claim |
| 6 | MAD tracking not the sole KPI | Spring costs tracking by design |
| 7 | Commercial license executed | Snippet ≠ production grant |
| 8 | Rollback path + monitoring of filter outputs | Operational hygiene |
Pricing: customers.
Limits: safety cage.
Losses: when GSRF lost.
Questions people actually ask
Is the Try snippet production-ready?
No. Evaluation only. Production requires a commercial license and your acceptance tests.
What is the #1 go-live blocker we see conceptually?
Using GSRF as a delayed closed-loop command smoother or as a fixed-FA residual detector — both are published red lights.
Do I need adaptive x* in production?
Only if operating point permanently moves. Adaptive helps multi-regime traces; can hurt flat SpO2-like series. Document the policy.
Request Audit Pack
Try snippet
Evidence
Research