Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2607.22682.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

A paper proposing a vocabulary for describing multi-agent automated research systems, covering agent identity, operations, communication, visibility, action selection, initialization, and evaluation. It defines a trajectory as a record of one run and distinguishes generative taste (rate of novel proposals) from evaluative taste (gap between proxy score and true quality).

Provides a shared framework for comparing autoresearch system designs and splits the vague complaint about lacking taste into two distinct, testable failure modes.

Cited by

No citations on record.