workspace | string | ".pandaprobe/reflect"; read-only; Base directory, followed by an identity partition; PANDAPROBE_REFLECT_WORKSPACE overrides the default |
traceVisibilityTimeout | number | 30; read-only; Wait for terminal persisted trace evidence |
evaluationTimeout | number | 120; read-only; Evaluation create/wait budgets |
pollInterval | number | 1; read-only; Poll interval |
evaluationMaxRuns | number | null | null; read-only; Optional creation-attempt cap per UTC day |
evaluationMaxTraces | number | 32; read-only; Maximum trace references for one execution |
evaluationSampleRate | number | 1; read-only; Fraction admitted by deterministic execution-ID sampling |
maxConcurrency | number | 4; read-only; Concurrent evaluation orchestration per runtime |
cooldownSeconds | number | 60; read-only; Repeated failure-signature cooldown |
reflectionTimeout | number | 60; read-only; Host reflection continuation deadline |
shutdownTimeout | number | 5; read-only; Bounded cancellation wait |
reflectionMaxCommands | number | 8; read-only; Tool-call allowance at the reflection checkpoint |
shellTimeout | number | 5; read-only; Interpreter call deadline |
shellMaxCommands | number | 2000; read-only; Interpreter command limit |
shellMaxLoopIterations | number | 5000; read-only; Interpreter loop limit |
shellMemoryBytes | number | 16 * 1024 * 1024; read-only; Interpreter memory |
commandMaxBytes | number | 16 * 1024; read-only; Tool command input |
outputMaxBytes | number | 12 * 1024; read-only; One tool response |
executionOutputMaxBytes | number | 96 * 1024; read-only; Cumulative output, including guard/error replies |
contextMaxBytes | number | 8 * 1024; read-only; Active notes |
ruleMaxBytes | number | 8 * 1024; read-only; One rule |
rulesMaxBytes | number | 512 * 1024; read-only; Active rule corpus |
rulesMaxCount | number | 256; read-only; Active rule count |
evidenceMaxBytes | number | 64 * 1024; read-only; Assigned diagnostic presentation |
thresholds | Record<string, number> | { task_completion: 0.5, coherence: 0.5 }; read-only; Primary metrics and strict failure thresholds |
diagnosticMetrics | readonly string[] | ["tool_correctness", "argument_correctness"]; read-only; Follow-up diagnosis after a notice |
stallWindow | number | 5; read-only; Score-history window |
minimumProgress | number | 0.02; read-only; Stall variation cutoff |
regressionDrop | number | 0.15; read-only; Previous-score drop cutoff |
validationMinCases | number | 3; read-only; Distinct paired cases |
validationMinImprovement | number | 0.05; read-only; Mean non-protected metric improvement |
validationRegressionTolerance | number | 0.02; read-only; Allowed per-case metric decrease |
validationTimeout | number | 120; read-only; Replay collection deadline |