FaultBench-Industrial Research documentation
Research documentation ยท Experimental record
Agent dropout / Permanent

LLM-agent replanning

Permanent agent dropout evaluation

Fault: Agent dropout
Duration: Permanent
Recovery mechanism: LLM-agent replanning

Experiment definition

The experiment evaluates LLM-agent replanning under a permanent agent dropout injection. Fault onset is randomized between environment steps 40 and 60.

Expected evaluation record

MetricRecorded value
Degradation-area39.96 ± 13.72 (corrected; mean ± 95% CI, n=24)
Safety-violation excess208.8 ± 73.5 (weighted, control-subtracted; mean ± 95% CI, n=24)
Fraction of time healthy0.450 ± 0.074 (mean ± 95% CI, n=24)
Relapse count0.08 ± 0.12 (mean ± 95% CI, n=24)
False-recovery rate0.013 ± 0.014 (mean ± 95% CI, n=24)
Intervention recall0.541 ± 0.114 (mean ± 95% CI, n=24)

Values are means with 95% confidence intervals across every target × duration-fixed combination shown on this page (see graphs below), 8 seeds each. See the Results page for the cross-mechanism comparison.

Experimental trajectory graphs

One representative seed (seed 0 of 8) per target combination from the full sweep. See the Results page for the multi-seed aggregate comparison across mechanisms. The graphs for every seed are archived at doi:10.5281/zenodo.23304785 (files available from 10 November 2026).

Sorting Agent Target

Agent dropout permanent fault, LLM-agent replanning, Sorting Agent Target
LLM-agent replanning - Sorting Agent Target - evaluation trajectory under permanent agent dropout (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep.

Pressing Agent Target

Agent dropout permanent fault, LLM-agent replanning, Pressing Agent Target
LLM-agent replanning - Pressing Agent Target - evaluation trajectory under permanent agent dropout (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep.

Both Agents Target

Agent dropout permanent fault, LLM-agent replanning, Both Agents Target
LLM-agent replanning - Both Agents Target - evaluation trajectory under permanent agent dropout (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep.

Notes

This page is intentionally separated from the paper-facing results so that full-resolution plots, additional seeds, and diagnostic figures can be documented without overloading the publication.