1 · This appliance, right now
Read live from the container serving this page.
2 · Where it runs
- A container on your infrastructure. The licensor hosts nothing and receives nothing: no telemetry, no call home (EULA §3.1).
- Runs as a non-root user. Multi-stage python:3.11-slim image.
- Loopback by default. The evaluation compose file binds 127.0.0.1:8000 only, because API-key checks are off for evaluation.
- Your credentials stay yours. Model and key-verification credentials are read from your environment and never sent to the licensor (§3.4).
3 · What leaves your network, endpoint by endpoint
This edition carries the three deterministic engines only. None of them makes a network call with your data, and the image contains no AI model client that could (EULA §3.2).
| Endpoint | Network | What leaves the container |
|---|---|---|
| POST /v1/cdr/validate | In-process | Nothing |
| POST /v1/ingest/inspect | In-process | Nothing |
| POST /v1/claims/triage | In-process | Nothing. This edition triages structured facts only. |
docker run --rm --network none -e ENVIRONMENT=local -e AUTH_DISABLED=true \ apexdatarouting/cdr-gatekeeper:latest
The three in-process endpoints keep working inside that container.
4 · What the licence commits to
The fees you paid in the preceding twelve months, or £100, whichever is greater.
Loss of profit, revenue, data, regulatory fines, and indirect or consequential loss.
Pre-flight linting only. No underwriting advice, clearing metrics or Velonetic compliance sign-off.
5 · Evidence
Test counts, extraction accuracy with confidence intervals, container checks from the last verification run, and the known limits sheet.
Every endpoint, try-it-out. Every response carries the disclaimer and how its shape was enforced.
The automated checks verify the statements above against this appliance and report pass or fail.
Read the workbook's Known limits sheet before relying on the accuracy figures: they come from a small synthetic evaluation set.