The experiment evaluates Hierarchical supervisor-agent under a permanent byzantine / adversarial agents injection. Fault onset is randomized between environment steps 40 and 60.
Expected evaluation record
Metric
Recorded value
Degradation-area
55.65 ± 5.21 (corrected; mean ± 95% CI, n=216)
Safety-violation excess
327.2 ± 36.9 (weighted, control-subtracted; mean ± 95% CI, n=216)
Fraction of time healthy
0.411 ± 0.028 (mean ± 95% CI, n=216)
Relapse count
0.25 ± 0.07 (mean ± 95% CI, n=216)
False-recovery rate
0.246 ± 0.006 (mean ± 95% CI, n=216)
Intervention recall
0.761 ± 0.009 (mean ± 95% CI, n=216)
Values are means with 95% confidence intervals across every target × duration-fixed combination shown on this page (see graphs below), 8 evaluation seeds each, for each of 3 training seeds. See the Results page for the cross-mechanism comparison.
Experimental trajectory graphs
One representative run (evaluation seed 0 of 8, training seed 1 of 3) per target × mode 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 (fixed malicious mode)
Hierarchical supervisor-agent - Sorting Agent Target (fixed malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Pressing Agent Target (fixed malicious mode)
Hierarchical supervisor-agent - Pressing Agent Target (fixed malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Both Agents Target (fixed malicious mode)
Hierarchical supervisor-agent - Both Agents Target (fixed malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Sorting Agent Target (random malicious mode)
Hierarchical supervisor-agent - Sorting Agent Target (random malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Pressing Agent Target (random malicious mode)
Hierarchical supervisor-agent - Pressing Agent Target (random malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Both Agents Target (random malicious mode)
Hierarchical supervisor-agent - Both Agents Target (random malicious mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Sorting Agent Target (worst action mode)
Hierarchical supervisor-agent - Sorting Agent Target (worst action mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Pressing Agent Target (worst action mode)
Hierarchical supervisor-agent - Pressing Agent Target (worst action mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
Both Agents Target (worst action mode)
Hierarchical supervisor-agent - Both Agents Target (worst action mode) - evaluation trajectory under permanent byzantine / adversarial agents (fault onset randomized between steps 40-60), seed 0 of the 8-seed sweep, training seed 1 of 3.
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.
