Scope firstOne scenario and agreed constraintsKPIs and baselines agreed upfrontClear boundaries: public site is intentionally non-enabling
Threat model (evaluation oriented)Eclipse style neighbor capture, targeted victim isolationRouting table poisoning, gradual or burst injection of biased entriesChurn, mild to severe, including orchestrated churnSybil note: meaningful evaluation requires an explicit identity or admission control postureNo operational attack steps on the public site (evaluation only)Details and variants available under NDA after scope alignment
What we measure (examples)Giant component under churnEclipse susceptibility proxyRecovery time after partitionMixing or propagation proxyTail behavior under targeted removal
TrackBackbone failures: reduce worst-case impact postureTail behavior: p95 to p99.9 under comparable regimesOperational safety: guard rails, budgets, rollback expectations
How it worksScope alignment: scenario, constraints, KPIs, baselinesTime-boxed execution: stress suites and comparable regimesDecision report: pass / conditional / no-go
Typical engagement modelScope alignmentReproducible run planExecutionDecision reportNon-enabling on the public site. Results may vary. No deployment instructions.
DeliverablesScope sheet: assumptions and scenario definitionResults summary: distributions, limitations, variability postureOptional NDA addendum after scope alignment (when warranted)
Decision report includesDistributionsWorst-case seedsBudget consumptionLimitationsPass / conditional / no-go
Reproducibility (where applicable)Controlled run steps and runbook postureConfigs and logging expectationsComparable baselinesSample redacted report
Clear boundariesNo enabling algorithmic disclosure on the public siteNo production deployment on a public scopeResults may vary with constraints and operating conditions