Each synthetic participant selects between two reference-coded routes. The beneficiary set and demand remain fixed by the evaluator.
SYNTHETIC SERVICE-RECOVERY BENCHMARK
Synthetic service recovery
Recorded synthetic routing cases with message poisoning, unreliable verification, action denial and prescribed capacity loss.
Recovery requires the declared error, service, beneficiary-coverage and violation criteria for ten consecutive updates. The rule is not a general alignment measure.
Recorded casesMethods, results and limitationsProgrammed changes affect messages, verification, action authority or capacity. Scenario labels do not establish simulation of the corresponding real-world phenomena.
A separate scorer applies declared thresholds. Threshold satisfaction establishes an operational result, not safety or ethical adequacy.
Scope and limitations
No novelty claim. Comparative added value is unestablished.
Evidence. Finite synthetic cases generated internally. Post-freeze cases were generated after code/scorer freeze, without external blinding or independent validation.
Normative choices. Beneficiaries, demand, reference routes, thresholds and violation rules are evaluator stipulations, not established ethical truths.
Comparison. Methods are comparable only within matched suites. Bundled changes to information, gates, gains and topology prevent attribution of improvement to RETA or recursion.
Evaluator effects. The coordinator designed the generator and scorer. Thresholds shape what counts as recovery; repeated dependent review is not additional independent evidence.
Available resources limited testing to simplified simulations. Separate diagnostics use a one-step greedy search over three scalar payloads, reconsidered every ten updates. No declared outcome changed across 128 matched comparisons; stronger adversarial testing is not demonstrated. Resources for broader empirical validation remain insufficient. Diagnostic method and limits
RECORDED CASES
Recorded cases
Graph layout is illustrative. Curves display recorded or reconstructed measurements.
Recorded service outcome
Interpretation
—Useful service delivered
—Least-served beneficiary
—Violations this update
Separate cases retain distinct initial conditions. Continuous playback does not merge causal histories or modify data.
Exact measurements and source references
FINITE RECORDED OUTCOMES
Outcomes under the operational definition
Guarded fixed control met the final rule in 48/48 post-freeze internal cases; 20 contained earlier violations. In the distinct development-exposed persistent-extreme suite, the method met the rule in 50/144 cases.
All benchmark countsMethods and limitationsConditional theorem and scopeProvenance recordData and review website
Finite scalar-service results do not establish AI alignment, ethical adequacy, physical resilience, causal superiority or unrestricted adversarial robustness.