Layer 1

Bland Webhook node

succeeded

Verified authenticated Webhook-node execution with Neon-backed persistence.

Layer 2

Policy/action regression

50 deterministic policy regression scenarios through the shared gateway. Not a model or telephony evaluation.

Layer 3

HTTP and retry contracts

succeeded

Covered by Vitest: webhook auth, malformed input, idempotency conflicts, tenant fail-closed, reconciliation, and approvals.

Layer 4

Postgres verification

succeeded

Neon correlation recorded the Bland-driven action and requisition.

Deterministic Regression

Business-side safety summary

No eval run recorded yet. The page will surface metrics after `npm run eval`.

Eval Markdown
# Eval Results (deterministic policy regression)

These figures come from a fixed set of scripted scenarios run through the deterministic policy gateway. They are not a model-quality, telephony, or production-reliability measurement.

- Total scenarios: 50
- Passed: 50
- Failed: 0
- Intent accuracy: 100%
- Field extraction accuracy: 100%
- Action accuracy: 100%
- Guardrail accuracy: 100%
- Clarification correctness: 100%
- Audit coverage: 100%
- Policy-bypass defense rate: 100%
- Finance privacy pass rate: 100%
- Duplicate prevention pass rate: 100%
- Unsafe action rate: 0%
- Unsafe action count: 0
- Created requisitions: 5
- Created approvals: 4
- Created escalations: 2
- Missing audit logs: 0
- Readiness status: ready

## Failed scenarios

- None
Readiness Report
# Deployment Readiness Report (deterministic policy regression)

## Summary

This harness ran 50 scripted policy regression scenarios through the shared voice action gateway. "Ready" means those scenarios passed with no unsafe mutation; it does not assess conversation quality or live traffic.

## Readiness Signals

- Intent accuracy: 100%
- Field extraction accuracy: 100%
- Action accuracy: 100%
- Guardrail accuracy: 100%
- Audit coverage: 100%
- Policy-bypass defense rate: 100%
- Finance privacy pass rate: 100%
- Duplicate prevention pass rate: 100%
- Unsafe action count: 0
- Missing audit logs: 0

## Critical Failures

- None

## Recommendation

Ready. No unsafe mutations occurred, audit coverage is 100%, and critical scenarios passed.
External Evidence Index
Run `npm run eval` to generate this report.