Skip to content

Releases: dlescanogithub/voice-receptionist-consent-lab

Release list

v0.1.0

Choose a tag to compare

@dlescanogithub dlescanogithub released this 22 Oct 23:38
0a4a3fc
  • Package the offline acoustic measurement and the credit-gated WebSocket probe under src/consent_lab. The old tools/ commands delegate to it. The prompt, the sanitized config, the case files, and the saved audio are unchanged.
  • Environment-only settings. The process does not read a dotenv file. LIVE_ENABLED must be the string 1. Anything else leaves the probe off.
  • JSON logs. Bounded retries (at most 5) on the signed-URL fetch and on the socket connect, with jitter and a cap. A turn that has started is not retried. A turn with no non-silent interval is skipped instead of aborting the CSV.
  • Tooling: ruff, mypy (strict), pytest with coverage, pre-commit, a small Dockerfile, and a secret scan. CI for Python 3.11 and 3.12 is .github/workflows/ci.yml.
  • Docs in English: README, docs/HOW_IT_WORKS.md, discovery, risks, runbook, and cost. Labels used below and in the README are MEASURED, [EST], [SYN], SIMULATED, and UNVERIFIED.
  • Local pytest result: 100 passed, 97.42% line coverage, Python 3.12.3. MEASURED.

The v17 simulation score (8 of 10) and the untested v18 consent re-ask are historical lab records. This package does not re-run them.

Also on this tag: CI is active (.github/workflows/ci.yml, Python 3.11 and 3.12).