Investigation 371

Dataset: espionage
Model: opus40
Training model: claude-opus-4-20250514
Inference model: gpt-4.1-mini
Example count: 3
Patience: 3
Current round: 543
Best round: 489

Rounds

Round	UUID	Started	Completed	Train Acc	Train KT	Val Acc	Val KT	Min Train/Val	Test Acc	Test KT
124	eef6d64e-9aa9-43a6-b89a-cafc1319b25c	2025-06-25	in progress	n/a	n/a	n/a	n/a	n/a	n/a	n/a
125	18201a65-bd07-4e1c-be25-21774dcbe71c	2025-06-25	2025-07-03 09:28:37.643645+00:00	0.662	0.179	0.350	0.455	0.350	0.750	0.126
481	d107d87e-cc0d-4f3a-8027-b11e8f79bad2	2025-07-03	2025-07-03 09:42:49.700434+00:00	0.719	0.144	0.600	0.222	0.600	0.750	0.126
482	af3b6dc4-3d92-446f-8ca6-d93b43aa5e6d	2025-07-03	2025-07-03 09:58:24.543544+00:00	0.775	0.111	0.700	0.156	0.700	0.900	0.047
483	1bfbc642-d0aa-436c-9d61-8672f74cf206	2025-07-03	2025-07-03 10:15:02.376896+00:00	0.800	0.098	0.700	0.156	0.700	0.900	0.047
484	7a2bda1a-e9bf-4403-afb4-f173309e912b	2025-07-03	2025-07-03 10:31:28.664821+00:00	0.831	0.081	0.750	0.126	0.750	0.900	0.047
485	2518a85d-5a64-4ca9-a2c0-da8f5c88319e	2025-07-03	2025-07-03 10:48:08.487102+00:00	0.831	0.081	0.700	0.156	0.700	0.900	0.047
486	f2086a59-0959-4dca-bd93-cc6ff724061c	2025-07-03	2025-07-03 11:02:29.073456+00:00	0.838	0.078	0.700	0.156	0.700	0.950	0.023
487	aadfd1d9-d350-4083-a164-82554c05f759	2025-07-03	2025-07-03 11:18:03.763044+00:00	0.825	0.084	0.850	0.071	0.825	0.850	0.071
488	a30dc1c7-c123-47d7-abb5-2bd3caf43a6b	2025-07-03	2025-07-03 11:30:50.264315+00:00	0.844	0.075	0.750	0.126	0.750	0.850	0.071
489	1c39255b-f1e4-41a8-a4b6-6b737ea7535e	2025-07-03	2025-07-03 11:43:27.085864+00:00	0.887	0.053	0.850	0.071	0.850	0.800	0.098
490	ded81f5c-b08e-4205-a003-3194530432ca	2025-07-03	2025-07-03 11:58:15.253031+00:00	0.894	0.050	0.850	0.071	0.850	0.800	0.098
491	0bd7c242-c356-47cd-89bf-479f9e0a20cb	2025-07-03	2025-08-02 08:19:17.479466+00:00	0.894	0.050	0.850	0.071	0.850	0.800	0.098
540	4a1ad91c-8277-4f87-b666-7f98111a57e3	2025-08-02	2025-08-02 08:22:11.381714+00:00	0.944	0.026	0.800	0.098	0.800	0.800	0.098

Scores

round scores