MITCF Research Platform

Run Experiment

Launch a new MITCF pipeline run. Runs all 5 ablation conditions on the benchmark dataset in a background thread.

Input Path

Seed

Max Revisions

Verification Threshold

Bootstrap Samples

Split

Calibrate DetectorUse Label for Verification

Hallucination weight0.25

Bias weight0.25

Robustness weight0.25

Decision weight0.25

Sum1.00

verification_threshold0.0–1.0Min overlap score to pass verification

max_revision_attempts0–5Max retries when critique finds issues

bootstrap_samples50–5000Bootstrap CI sample count (higher=more precise)

calibrate_detectorboolWhether to calibrate hallucination detector from training split

use_label_for_verificationboolUse gold label as reference during verification