Load-test export files

Export every call and step of a finished load-test run from its report page (Export data) or the API. You get a zip with calls.csv, steps.csv, calls.jsonl (the complete record), run.json and a README.md. The README inside each export is the authoritative description of that file.

Reading the files

Join steps.csv to calls.csv on call_id.
An empty cell means "not measured", never zero.
Times are ISO 8601 in UTC.
Recordings are not in the zip: download one by recording_id with GET /audio/download-url?audioId=. Recordings older than 30 days are archived and may not download.
Runs transcribed through Telnyx have no per-word confidence or response latency.
Exports are kept for 30 days. You can export a run again at any time.

Export with the API

Endpoint: /testing/load-tests/{id}/exports. Needs the testing:read scope.

bash

# 1. Start an export of a finished run
curl -X POST -H "x-api-key: YOUR_API_KEY" \
  https://api.nopaque.co.uk/testing/load-tests/RUN_ID/exports
# -> 202 {"export": {"exportId": "EXPORT_ID", "status": "pending", ...}}

# 2. Poll until status is "ready" (usually under a minute)
curl -H "x-api-key: YOUR_API_KEY" \
  https://api.nopaque.co.uk/testing/load-tests/RUN_ID/exports/EXPORT_ID
# -> 200 {"export": {"status": "ready", "downloadUrl": "https://...", ...}}

# 3. Download the zip (the link is valid for 1 hour; poll again for a fresh one)
curl -L -o export.zip "DOWNLOAD_URL"

calls.csv

Column	Unit	Meaning	Empty means
`call_id`		Unique id of this call. Joins to steps.csv `call_id` and to `id` in calls.jsonl.	Never empty.
`load_test_run_id`		The load-test run this call belongs to (the same for every row).	Not reported for this call (for example, the call was never answered or the field predates this data).
`test_config_id`		The test configuration the call executed.	Not reported for this call (for example, the call was never answered or the field predates this data).
`status`		Call record status: completed, failed, cancelled, running, or pending (created but never dialled).	Never empty.
`outcome`		Test verdict for the call: PASS, FAIL, ERROR or INCONCLUSIVE.	The call did not finish.
`error_message`		Why the call failed or errored, when known.	No error was recorded.
`passed_steps`		Number of steps that passed.	The call did not report step results.
`failed_steps`		Number of steps that failed or timed out.	The call did not report step results.
`dialed_at`	ISO 8601 UTC	When we dialled.	Not reported for this call (for example, the call was never answered or the field predates this data).
`answered_at`	ISO 8601 UTC	When the far end answered.	Not answered, or not reported.
`ended_at`	ISO 8601 UTC	When the call ended.	Not reported for this call (for example, the call was never answered or the field predates this data).
`dial_to_answer_ms`	ms	Time from dialling to answer.	Not answered, or not reported.
`answer_to_first_utterance_ms`	ms	Time from answer until the system under test first spoke.	Not reported for this call (for example, the call was never answered or the field predates this data).
`call_duration_s`	s	Length of the call.	Not reported for this call (for example, the call was never answered or the field predates this data).
`to_number`	E.164	The number dialled.	Not reported for this call (for example, the call was never answered or the field predates this data).
`caller_id`	E.164	The caller ID presented on this call.	Not reported (the platform default number was presented).
`caller_id_from_pool`		true when the caller ID was leased from the workspace's number pool for this call. The run-level split by source is in run.json stats.	Never empty.
`terminal_status`		How the call ended, from the tester: completed, failed, no_answer, aborted_by_duration, aborted_by_inactivity, and so on.	Not reported for this call (for example, the call was never answered or the field predates this data).
`hangup_cause`		Carrier hang-up cause (e.g. normal_clearing).	Not reported for this call (for example, the call was never answered or the field predates this data).
`hangup_source`		Who hung up: caller (us) or callee.	Not reported for this call (for example, the call was never answered or the field predates this data).
`sip_hangup_cause`		SIP response code behind the hang-up, when the carrier gave one.	Not reported for this call (for example, the call was never answered or the field predates this data).
`estimated_mos`	1 to 4.5	Estimated call quality (MOS) from the platform's own model (ITU-T G.107), using jitter and packet loss. This is the figure the run report uses.	Not enough audio statistics to estimate.
`telnyx_mos_inbound`	1 to 4.5	MOS as reported by the carrier (Telnyx) for audio we received. Close to 4.5 on nearly every call, so prefer estimated_mos.	Not reported for this call (for example, the call was never answered or the field predates this data).
`telnyx_mos_outbound`	1 to 4.5	MOS as reported by the carrier for audio we sent.	Not reported for this call (for example, the call was never answered or the field predates this data).
`inbound_jitter_max_variance_ms`		Maximum jitter variance on received audio, exactly as reported by the carrier. The unit is the carrier's; treat it as relative between calls.	Not reported for this call (for example, the call was never answered or the field predates this data).
`inbound_packets`		RTP packets received.	Not reported for this call (for example, the call was never answered or the field predates this data).
`inbound_packets_lost`		RTP packets lost on the way to us.	Not reported for this call (for example, the call was never answered or the field predates this data).
`outbound_packets`		RTP packets sent.	Not reported for this call (for example, the call was never answered or the field predates this data).
`outbound_packets_lost`		RTP packets we sent that were lost.	Not reported for this call (for example, the call was never answered or the field predates this data).
`codec`		Audio codec negotiated on the call (e.g. PCMA).	Not reported for this call (for example, the call was never answered or the field predates this data).
`audio_coverage`	0 to 1	Share of the call during which we received audio frames.	Not reported for this call (for example, the call was never answered or the field predates this data).
`recording_id`		Id of the call recording. Download it with GET /audio/download-url?audioId=<recording_id>. Recordings older than 30 days are archived and may not download.	No recording was saved.
`recording_duration_s`	s	Length of the recording.	No recording was saved.
`stt_provider`		Speech-to-text path used for this run: deepgram (Deepgram direct) or telnyx (Telnyx, Deepgram engine). Telnyx returns no per-word data, so steps.csv word confidence and response latency are empty for telnyx runs.	Not recorded: the run was started before the platform recorded the provider (Deepgram unless an operator had chosen Telnyx).

steps.csv

Column	Unit	Meaning	Empty means
`call_id`		The call this step belongs to. Joins to calls.csv `call_id`.	Never empty.
`step_index`		Position of the step in the test, from 0.	Never empty.
`step_id`		Id of the step in the test configuration.	Not reported for this call (for example, the call was never answered or the field predates this data).
`step_name`		Name of the step in the test configuration.	Not reported for this call (for example, the call was never answered or the field predates this data).
`step_type`		What the step did: listen, audio (play a clip), dtmf, hold or wait.	Not reported for this call (for example, the call was never answered or the field predates this data).
`outcome`		PASS, FAIL, TIMEOUT (nothing matched before the listen window closed) or ERROR.	Never empty.
`error_message`		Why the step failed, when known.	No error was recorded.
`expected_transcript`		What the step expected to hear.	The step does not listen.
`actual_transcript`		What was transcribed.	Nothing was transcribed for this step.
`similarity`	0 to 100	How closely actual matched expected. Can come from an early partial match (the step passes as soon as enough of the prompt is heard), so a correct prompt can score below 100.	The step does not listen.
`threshold`	0 to 100	Similarity needed to pass.	The step does not listen.
`started_at`	ISO 8601 UTC	When the step started.	Not reported for this call (for example, the call was never answered or the field predates this data).
`ended_at`	ISO 8601 UTC	When the step ended.	Not reported for this call (for example, the call was never answered or the field predates this data).
`duration_ms`	ms	How long the step took.	Not reported for this call (for example, the call was never answered or the field predates this data).
`response_latency_ms`	ms	For an audio step: time from the end of the system's last word to our reply starting.	Not an audio reply, or the run used the telnyx speech-to-text path (no word timings).
`avg_word_confidence`	0 to 1	Speech-to-text engine's average confidence over the words in this step.	No words were transcribed, or the run used the telnyx speech-to-text path.
`min_word_confidence`	0 to 1	Lowest single-word confidence in this step.	No words were transcribed, or the run used the telnyx speech-to-text path.
`word_count`		Number of words transcribed in this step.	No words were transcribed, or the run used the telnyx speech-to-text path.
`digits_sent`		DTMF digits we sent.	Not a DTMF step.
`hold_secs`	s	How long a hold step actually stayed on the line.	Not a hold step.