Trust is not a profile and not a prompt. It is a verifiable state. DarwinX sorts every access into three states, defaults to deny, and records the reason. A boolean gate forces every uncertain case into a wrong answer. A three-state gate does not.
Being on the same network is not a reason to trust. Recognition comes from an explicit allowlist, never from a shared subnet. Every refusal is logged exactly as an approval is, because a register of permits alone proves nothing was stopped.
The impact is auditability. No silent trust, no decision without a named reason, and a suspicion that does not quietly expire. This is sustainable by design: the knowledge is encoded once and adapts, instead of a rulebook that has to be maintained until it rots.
Runs fully in your browser. Allowlist here is 10.10.0.0/16 and 127.0.0.1; denylist is 203.0.113.0/24.
This is the governed flow DarwinX runs, and the flow our agents run on: a decision that permits, refuses while naming the rule it violated, and leaves a trace either way.