docs(vrs): define fail-closed once, where subsystems can cite it

Three subsystems state a fail-closed rule independently and none
references the others, so for a safety property the drift risk is the
failure mode that matters.

Definition and obligation are split along the existing doc-class line:
the ontology says what fail-closed means, and each subsystem's
requirements say which of its surfaces must be. The obligation stays
where it can be argued with; the definition stops being restated.

Records the asymmetry that makes the term operationally useful —
widening what counts as proven needs evidence, widening what counts as
unproven is always safe — and that fail-closed is a property of a
mechanism's unknown path, not a wish about it.

Names fail-open too, since it is usually reached by removing a check
that was accidentally load-bearing rather than by writing a permissive
one.

agent-session-id: e5217740-eef6-48eb-a794-3e5e11939e4d
agent-tool: Claude Code
agent-tool-version: 2.1.220
agent-model: claude-opus-5
agent-runtime-profile: /nix/store/i8y8b542cyqi385ywcjw5fvsq24f75v4-coding-agent-runtime-profile/share/coding-agents/profile.json
agent-skills-manifest: /nix/store/i81qxhzlrzcxrrdwpp6i8hagka2gby8y-agent-skills-corpus/share/agent-skills/manifest.json
tooling-profile: dotfiles@unknown-dirty

Author: schickling-assistant

docs/vrs/ontology.md

@@ -161,6 +161,29 @@ These terms are refined per harness; see the DING subsystem's nodes.
 - **Resource** — a durable, listable record of high-value output an agent
   produced, so a peer can find it.
 
+### Dispositions
+
+Words for how a mechanism behaves when it does not know. They are defined here
+once because several subsystems depend on them and each would otherwise state
+its own version.
+
+- **Fail-closed** — when a mechanism cannot prove the condition it needs, it
+  resolves toward refusal rather than toward action. Absence of evidence that
+  something is unsafe is not evidence that it is safe. The consequence worth
+  stating plainly: widening what counts as *proven* requires evidence, while
+  widening what counts as *unproven* is always safe. This is a property of a
+  mechanism, not a wish about one — a mechanism is fail-closed only if its
+  unknown path is the refusing path.
+- **Fail-open** — the opposite, and always a defect where it is not a stated
+  choice: an unproven condition resolving toward action. Naming it matters
+  because it is usually reached by removing a check that was accidentally
+  load-bearing, not by writing a permissive one.
+
+This document defines the terms; it does not impose them. Which surfaces are
+obliged to be fail-closed is stated in those subsystems' own requirements, so
+the obligation stays where it can be argued with and the definition does not
+drift across restatements.
+
 ### Health and evidence
 
 - **Doctor** — the on-demand health check for one catalog as seen from one host.