Containment claim
DetCordon is designed to let hostile web payloads execute inside a disposable, time-boxed sandbox while evidence flows outward to a separate store-only sink. The HTTP tap remains DetectionOnly: it observes the interaction rather than claiming to make the payload safe.
Trust boundaries
- The victim writes only to sandbox-scoped ephemeral storage.
- Sandbox egress is default-drop and limited to required evidence and name
- Event telemetry and sample upload use separate channels.
- The sink computes sample digests and encrypts samples to an analyst-held age
- The sandbox and sink are expected to run on different hosts.
- Dashboard access defaults to a private boundary and requires authenticated
resolution paths.
recipient; the analyst private key is not present on the sink.
mediation when used remotely.
Assurance and residual risk
Automated tests check important Docker and Firecracker deployment invariants, including DetectionOnly operation, ephemeral victim writes, default-drop egress, channel separation, sink-service separation, encrypted storage, and time-boxed execution.
Those tests do not prove live kernel isolation, physical host separation, firewall state after operator changes, or future evidence availability after a sink compromise. Every pilot therefore includes live-host preflight, separate-host confirmation, evidence checks, and an explicit residual-risk review. DetCordon is containment-first observation, not a formal proof that arbitrary third-party payloads are safe.