One path.
Six decisions.

STEP 1 / 00:00–01:30

Choose a path, not just a vendor.

Define the desk, direction, endpoint and recorder. This selection changes only the browser example; no call is placed.

STEP 2 / 01:30–03:00

Give the test an authorized window.

Agree synthetic-call consent, counterpart endpoints, operational ownership and a quiet window. Cadence follows your risk assessment, not a website default.

STEP 3 / 03:00–05:00

Follow both sides of the conversation.

Link CALL-DEMO-42 to REC-DEMO-42. Check the expected channels, duration and timestamps. A connected call can still leave one speaker out of the record.

Engine check: each end transcribes the other’s distinct spoken token live. Missing reciprocal audio flags one-way or no audio. The recovered recording must then contain the tokens on the correct legs, in order and without clipping or truncation.

Record the expected and observed answering party too. A human answer triggers a courtesy message, call end, log and reschedule—not a verified test.

STEP 4 / 05:00–07:00

Match the content you expected.

Known phrase: TraderVoice control test, path four, sample forty-two. Inspect the phrase and watermark on each expected leg. A file checksum protects the sample artifact; it is not an acoustic match or proof of caller identity.

Example: outbound marker found; return-leg marker absent.

Beyond the fingerprint: compare the archive or surveillance transcript against the known script for word error rate. Review STT confidence alongside MOS, jitter and loss; compare token spoken/heard times for per-leg latency and, on hoots, every subscriber’s delay.

Surveillance scripts also require the expected alert to fire. Token and answer-classification tags keep those tests separate from human review. See the unique spoken-token and two-sided script example →

STEP 5 / 07:00–09:00

Ask whether the record can be retrieved.

Distinguish retrieval access from preservation. Review metadata, policy assignment and legal-hold state; a successful download does not establish future retention.

STEP 6 / 09:00–10:00

Keep the result and the exception together.

The download is a browser-generated example of your selections, not a live execution record or signed attestation.

Carry answer classification, per-leg tokens, transcript accuracy, quality, timing and surveillance-alert references into the evidence record. Inspect the engine evidence checklist →

These controls explain the engine workflow. This browser walkthrough does not call, listen, transcribe or measure audio.