Measured, not claimed

Every page here marks its statements. The marks are not decoration — they encode how much you should trust each claim, and they exist because a manifest is dangerous enough that "it probably works" is not good enough.

Mark	Means
measured	Run end to end on the documenting host — no Docker daemon, 2026-08-27, marketplace at d09eecd. The output was seen and is quoted verbatim.
audited	Run by the maintainer against a real model. Transcripts and guard strings checked against source independently. Not reproduced here.
not built	Deliberately absent — reserved in the schema, refused at runtime, and documented as such.

Why the distinction is load-bearing

Three of the failures found in this project were silent by construction: a guardrail that failed open produced the same clean verdict as a safe bundle; a test suite could not see a Docker-only failure because its CI runner shipped a daemon; a bridge served 200 on every route while its real function was dead. In each case a green result meant nothing, and only someone who distrusted the green result on purpose found the truth.

So the tests that earn a measured mark here are the ones designed to fail loudly:

• For a guard: turn it off and confirm the test fails. A guard never watched failing is unproven.
• For a liveness signal: run a control arm. A value that reads healthy during a real outage teaches its readers to ignore it.
• For a monitoring gap: ask what a green check would look like during the failure. If the answer is "identical", that is the finding.

Nothing on this site is written as working that was only read in the source, and nothing planned is written as if it shipped. Where a claim could only be checked from the outside — the deployed commit, for instance — the honest limit is stated rather than the reassuring version.