Fixture-only sample report

Benchmark Snapshot

This page shows the current Voice Bench SA sample report format for a future paid benchmark deliverable. It uses fake fixtures and local checker outputs only.

It is not evidence of live users, buyer revenue, production South African speech data, a live WhatsApp collection loop, live Google Cloud product use, or a live Gemini API call.

Accepted

5 media-intake candidates

Exported

5 audio rows

Manifest

5 smoke rows

Excluded

3 dry-run rows

Buyer value

Show the benchmark shape before live collection starts

A future buyer can use this format to choose the first language scope, task area, review depth, and failure modes to evaluate. The current sample proves the reporting shape, not market traction. The fake-only sample now includes isiXhosa, Afrikaans, South African English, and mixed South African phrase tags.

Mode: fake_fixture_only_offline
Sample report: ops/xprize/generated/benchmark-snapshot-sample-report-v0.md
Evidence appendix: ops/xprize/generated/benchmark-snapshot-evidence-appendix-v0.json

Quality notes

Accepted rows pass the current evidence gates

Consent, provenance, task, audio, transcript, correction, review, and export gates are represented.
Self-correction and self-review are excluded from export-ready rows.
Withdrawal, rejection, and needs-more-review paths stay visible.
Raw transcripts, raw manifest rows, private contact details, and secrets are not printed here.

Current hashes

The fixture proof can be checked from the repo

Smoke manifest SHA-256: e03756f6ce3fdfd4327adb95ffdd3b3d523d0835002e85b98fcdeacc0c7732c6
Dry-run manifest SHA-256: 187639fb2d4af95f36f93b2f6295f86f925134fc9830e8410a417cb9dba7760b
Proof commands: npm run xprize:benchmark-snapshot:report; npm run xprize:benchmark-snapshot:check; npm run xprize:local-proof:check

Go-live gates

The sample becomes useful only after real gates clear

Buyer-approved question, language scope, task set, price, delivery date, and allowed use.
Legal, consent, withdrawal, deletion, retention, reviewer, payout, and POPIA handling.
WAHA/account-health checks and lawful contributor provenance.
Google Cloud bucket/object proof, deployed Gemini call, provider usage, and cost evidence.
Accepted and independently reviewed records for the agreed buyer scope.

Boundary

Sample first

Use this page for review of the deliverable shape.
Do not treat this page as revenue, user, provider, or production-data evidence.
Use the paid pilot page for buyer scope and pricing conversations.
Questions go to agent@updates.voicebench.co.za.

Discuss a pilot View readiness