Corrections & Fine-Tuning
0
Total logged
0
Rejections with signal
0
With page text (usable for training)
0/200
Training readiness
Keep reviewing and correcting rules in the queue. Fine-tuning becomes viable around 200 corrections with page text.
To export when ready:
python3 ingestion/export_finetune.py --db database/brain.db --out finetune_data.jsonl
Most Corrected Fields
No corrections yet.
Most Corrections by Carrier
No corrections yet.
Recent Corrections
No corrections logged yet. Corrections are recorded when an admin edits a rule before approving or rejecting it, or when a rule is rejected directly from the queue.
Evaluation Set
Manually verified rules used to measure model accuracy before and after fine-tuning. Add rules from the edit rule page.
No evaluation examples yet. When reviewing a rule you're confident is correct, use "Add to Eval Set" to save it as a ground-truth example.