experience-memory-tolerates-an-imperfect-self-judge CLM-0517observationsingle paperReviewed by AI
When an agent labels its own past attempts as successes or failures to build an experience memory, the labels can be substantially wrong without much cost: the LLM judge used was 72.7% accurate against ground truth on WebArena-Shopping, and simulating judges anywhere from 70% to 90% accurate gave similar success rates, with ground-truth labels best — so for distilling lessons, a rough self-judgment is close to enough, although a perfect one still helps.
Evidence for: No one should be judge in their own cause (narrows)
Capability: Improving its own scaffolding from failures
Observed on
2025, Gemini-2.5-flash as judge, one WebArena subset. Autonomous agent.
Applies when
- Effect on the capability:
- helps, narrowly
- Fails when:
- Judge accuracy falls below about 70%, the bottom of the range found harmless; the simulation flips labels at random, so a judge whose errors are systematic rather than random is not covered.
Sources
- Figure 8 and its text: judge calibrated against ground truth at 72.7%, then labels flipped at random to simulate 50% to 100% accuracy. One subset, random rather than systematic errors.
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For LLM agents that score their own actions during exploration in text-based games and then reuse that experience, the accuracy of those self-scores against the environment verifier does not predict later gains on held-out stricter configurations: block-level correlations between judgment agreement or calibration and the next checkpoint's score change sit near zero, so recognizing a good action is not the bottleneck — converting it into a transferable policy is.Improving its own scaffolding from failures · reviewed by AI
- When an LLM judge is validated against expert annotators on multi-dimensional rubrics for open-ended agent outputs, a single pooled agreement figure overstates reliability: per-dimension human-human agreement varies widely (substantial on a recall-style Completeness dimension, only slight to fair on Relevancy, Structural Clarity and Scope Adherence), the pooled statistic is inflated by between-dimension variation, and the judge's residual disagreements are directional (consistently one step stricter on Completeness, one step more generous on Structural Clarity) rather than noise.Whether the measurement made the finding · reviewed by AI
- In LLM multi-agent systems that reflect after failure, restricting reflection to the single agent responsible for the first incorrect step — located by a trained step-level correctness scorer and paired with a counterfactually generated corrected step — raises task success more than having every agent reflect on the trajectory, because reflections written by agents that behaved correctly enter their memory as misleading guidance for later tasks.Getting more from several models than from one · reviewed by AI
- When an LLM judge is given a prior evaluation score as context metadata on a revision, its rubric scores shift toward that uninformative prior — seven of eight tested models (API and open-weight, 0-5 scale, 20 fixed texts) show a negative total shift, and on a categorical compliance task the metadata blocks 48% of error corrections and flips about 10% of previously correct judgments to the anchored wrong label.Judging another's work as well as a careful reviewer would · reviewed by AI
- For LLM web agents on cluttered pages containing many concurrent interface components, success falls sharply as component count grows, and having the agent first emit a localized sub-goal plus a task-relevant sub-observation before choosing an action recovers much of the loss, because self-assessment of the action is more reliable in the reduced context.Using the tools it is given · reviewed by AI
Notes
Bears on the adage that no one should judge their own cause, which holds for judging a single output: for distilling lessons across many attempts, a self-judge's errors appear to average out. The scope condition is the random-error simulation; a judge with a consistent blind spot would write the same wrong lesson every time.