Transcript Verification Method
The Transcript Verification Report (Laporan Verifikasi Transkrip) shows how far a verbatim transcript has been checked against its audio recording. This page explains how every number in the report is calculated and how to confirm that a report is genuine.
Checking a report
For supervisors, examiners, or anyone who receives a Transcript Verification Report:
- Scan the QR code on the report. The verbatim.id verification page opens.
- Compare the report number and score on that page with the document you received. If they differ, the document has been altered.
- Check the transcript status. The verification page says whether the transcript was changed after the report was issued.
- Check the recording file (optional). Choose the recording file on the verification page to confirm it is exactly the recording on record. The check runs on your device; the file is not uploaded.
Verbatim Verification Score
The Verbatim Verification Score is the share of words that either have a system confidence of at least 0.60 or have been verified manually.
The score is rounded down to one decimal place, so a transcript that still has unverified words never reaches 100%. The score describes how complete the verification is, not how accurate the transcript is.
Word categories
Every transcribed word falls into exactly one category, so the four categories always add up to the total word count.
| Category | Description |
|---|---|
| High system confidence | Words the transcription engine gave a confidence of at least 0.60. |
| Verified manually, word by word | Words the system flagged, which the user then checked while listening to the recording and accepted or corrected. |
| Edited manually | Words the system flagged that sit in a segment the user rewrote or deleted. |
| Not yet verified | Words the system flagged that have not been checked. |
System confidence levels
The transcription engine gives every word a confidence value between 0 and 1. Words below 0.60, or without a value, are flagged for review.
| Level | Range |
|---|---|
| High | 0.60 and above |
| Medium | 0.40 to below 0.60 |
| Low | Below 0.40 |
| No score | The engine gave no value for the word |
Confidence values are the transcription engine’s internal estimates and have not been calibrated against manual transcripts. With Verbatim Presisi, the recording is processed independently by two engines; inter-engine agreement is the percentage of words both engines recognised the same way.
Review process
| Metric | Description |
|---|---|
| Word-level verifications | The number of times a flagged word was accepted or corrected. |
| Words corrected | The number of words the user changed from the initial transcription. |
| With audio playback | Word-level verifications made while the recording was playing or had been played. |
| Median time per word | The median time between opening a word and deciding on it. |
| Verifications under 1 second | Verifications decided very quickly, which can suggest a less careful review. |
| Segments edited manually | The number of segments whose text was changed outside word-level verification. |
| Verified by | The name of the account that did the verification or editing. |
Audio playback and verification timing are recorded by the user’s device.
Data integrity
The initial transcription is stored when transcription finishes and cannot be changed. Every verification and edit is logged on the server with its time and author. The numbers in the report are calculated from those two records, not from data sent by the user’s device.
| Hash value (SHA-256) | Description |
|---|---|
| Original recording | Calculated on the uploader’s device before the file is compressed for upload. |
| Transcribed audio | Calculated by the server from the file it received and transcribed. |
| Transcript content | Calculated from the transcript content at the moment the report is issued. |
The QR code on the report opens the verification page. That page shows the numbers recorded when the report was issued, shows whether the transcript has changed since, and lets anyone check a recording file against the record without uploading it.
Report coverage
A report can be issued for a single transcript or for a whole session. A session report covers every file in the session: the score and metrics are calculated across all files, a recording hash is shown for each file, and the transcript content hash covers all files. A session report can only be issued once every file in the session has finished transcribing.
Limitations
- The score does not state how accurate the transcript is.
- The report does not assess speaker identification, punctuation, or meaning.
- The report does not prove that a recording is authentic, for example whether it was edited before upload, who is speaking, or when it was made.
- The reviewer’s identity is not verified; what is recorded is the account name.
- Reports are only available for verbatim transcripts.
Version history
| Version | Changes |
|---|---|
| 1.0 · October 2026 | First version. |