Data Quality Tools

LLM-based verification and streaming integrity scans. For quick in-process checks and review queues, use the Quality Hub.

2960

Words in database

1280

Example sentences

View Batch Jobs

Given a sentence, uses an LLM to identify the vocabulary lemmas it contains, then creates database links between the sentence and those lemmas. Use Single Sentence for one-off additions or From File to batch-process a text file (one sentence per line). Translations into the selected target languages are generated at the same time.

Sentence

Source Language

Target Languages

Lithuanian

Chinese (Simplified)

French

Spanish

German

Japanese

Korean

Portuguese (Brazilian)

Russian

Context (optional)

Mark as verified

JSON output

File Path One sentence per line

Source Language

Target Languages

Lithuanian

Chinese (Simplified)

French

Spanish

German

Japanese

Korean

Portuguese (Brazilian)

Russian

Mark as verified

JSON output

CLI Equivalent:

PYTHONPATH=src python src/agents/bebras.py --sentence "Your sentence" --languages lt zh fr es

Scan the database for structural problems. Choose a specific check below or run them all at once. Checks marked with Fix Issues support auto-repair (currently: missing punctuation and sentence levels). Results stream in real time so you can watch progress on large databases.

For a read-only pass you can run anywhere, see Database Integrity and Stale Audio on the Reports page.

Check Type

All Checks
Run every check below in one pass

Orphaned Records
Derivative forms or word frequencies pointing to deleted lemmas/tokens

Missing Fields
Lemmas missing a definition, part of speech, or difficulty level

No Derivatives
Lemmas that have no inflected forms (e.g. no plural, no conjugations)

Duplicate GUIDs
Two or more lemmas sharing the same GUID identifier

Duplicate Words
Lemmas with the same text and part of speech that may need merging

Invalid Levels
Difficulty levels outside the valid 1–100 range

Missing Punctuation
Sentence translations not ending with a period, question mark, or exclamation mark

Sentence Levels
Sentences whose minimum_level doesn't match their linked word difficulties

Audio Mismatches
Audio files whose expected text no longer matches the current translation

Pronunciation Fields
IPA fields that look like leaked prompts/instructions instead of pronunciations

Output File (optional) Save detailed JSON report to file

Fix Issues
Auto-fix missing punctuation and sentence levels

CLI Equivalent:

PYTHONPATH=src python src/reports/integrity.py

Ask an LLM to review existing translations and judge whether they are correct. Translations are not regenerated — the model only gives a correct/incorrect verdict so you can spot problems. Use Verify Words for lemma translations or Verify Sentences for sentence translations. A specific word or sentence runs immediately. Bulk selections are queued as one independently retryable job per record.

Word ID (optional) Leave empty to verify multiple words

Limit Max words to verify

Languages to Verify

Lithuanian

Chinese (Simplified)

French

Spanish

German

Japanese

Korean

Portuguese (Brazilian)

Russian

Only unverified translations

Report only (don't update database)

JSON output

Sentence ID (optional) Leave empty to verify multiple sentences

Limit Max sentences to verify

Languages to Verify

Lithuanian

Chinese (Simplified)

French

Spanish

German

Japanese

Korean

Portuguese (Brazilian)

Russian

Only unverified translations

Report only (don't update database)

Also verify linked lemmas

JSON output

CLI Equivalent:

PYTHONPATH=src python -m verification words --languages lt zh fr es --limit 100

Find lemmas that share the same headword and part of speech. These are likely accidental duplicates that should be merged. Lemmas intentionally distinguished by a disambiguation tag (e.g. "mouse (animal)" vs "mouse (computer)") are not flagged.

Same scan from the command line: PYTHONPATH=src python src/reports/integrity.py --check duplicate-words