Multi-Attempt Response

A response containing multiple distinct attempts to answer the prompt, used as primary metric for ESR

Neighborhood — ranked by edge-count

concept

Endogenous Steering Resistance
implements
The central phenomenon introduced by this paper: inference-time recovery from irrelevant activation steering in LLMs

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Multi-Attempt Rate (metric)concept0.820
Secondary metric: percentage of responses containing multiple attempts, separating surface from actual self-correction
Five judge models agree 90-96% on multi-attempt detection and ESR direction for same responsesfinding0.755
Validation that ESR findings are not artifacts of any particular judge model's evaluation methodology
Stress Responseconcept0.744
Multiple-choice evaluation method for PM trainingmethod0.741
Using language model log probabilities of answer choices (A)/(B) to produce preference labels.
Multi-Agent Reflectionconcept0.739
Approach using multiple LLM agents for generation and critique, a key prior approach to improving reflection.
multiplicative scoringconcept0.737
Score = (sum of completed quartet values) × (number of quartets), making portfolio composition consequential.
Multi-attempt rate peaks at 2.7% around -0.3σ below threshold in boost sweep experimentfinding0.732
Quantitative characterization of ESR operating regime in boost level sweep
Responsivenessconcept0.731
Requirement that answers to questions be responsive as well as truthful; requires knowing that questioner will know the answer after receiving it.