Skip to content
Quality framework

How we keep bad data out

Every release goes through four checks before it ships.

Hybrid QANative-speaker audits
QA pipeline

Four checks before a release

If a batch fails any step, it does not go out.

01

Does the file even look right?

Required fields, file structure, and metadata are checked before the batch enters deeper review.

02

Automated checks

Deduplication, outlier detection, language ID, and transcript sanity checks run across every segment.

03

Native speakers review samples

Bengali, Assamese, Odia, and Meitei reviewers listen, read, and flag what machines miss.

04

Release gate

Passing batches receive a version number and scorecard. Failing batches do not ship.

Metrics and thresholds

What the scorecard shows

These sample numbers show the checks tracked on each release.

Transcription accuracy

97.4%

Bengali speech median across recent releases

Dialect label confidence

91.2%

Checked on each release

Rejected segments removed

4.8%

Removed before buyer export

Versioning and change control

  • Every release receives a version ID, manifest hash, and change-note summary.
  • Corrections are reissued as new versions instead of silent overwrites.
  • Buyers receive changelog visibility when affected fields or files change.

Issue handling

  • Issue identified through QA or buyer feedback
  • Scope impact reviewed against the current export record
  • Correction batch prepared and re-audited
  • Replacement package issued with an updated scorecard

Open the QA scorecard sample.

See the metrics, thresholds, and issue flags.