Model techniques map
Techniquespost-trainingagentic post-training

specific method · filed under post-training

Repair-agent environment correction loop

A repair agent fixes identified environment errors, recalibrates evaluation points, and sends the task back for verification.

source
1
model
1
lab adopt it
1
strongest
used

How sources treat it

One count per evidence span, weakest treatment to strongest.

used 2

Documented in

Evidence

2 spans quoted from the sources, strongest treatment first.

If the inspection does not pass, a repair agent fixes all identified errors, adjusts evaluation points that are too easy or too difficult, and the task re-enters verification.

usedunclearin DeepSeek-V4.1-FlashDeepSeek

a repair agent fixes all identified errors, adjusts evaluation points that are too easy or too difficult, and the task re-enters verification

useddata curationin DeepSeek-V4.1 RL training dataDeepSeek

Filed alongside

Other methods under post-training :: agentic post-training.