safe-to-publishlisted

# Safe to publish Two questions, and they are the same question at different distances. **Is this correct enough to publish?** That is a gate: a check between authored and published whose failure mode is refusal. **Could the wrong part of it escape?** That is a whitelist: an explicit list of what may cross the boundary, and a test that fires if anything else does. Both fail the same way — quietly, through convenience, at the moment somebody is in a hurry. --- # Part one — the gate A gate is a deterministic check that runs between authored and published, and its failure mode is **refusal**, not a warning line in a log. A warning that can be ignored eventually is. ## The pattern **1. One command, exit non-zero on any blocker.** Whatever publishes the material runs the check first and stops. In the project this came from, the seeding command re-verifies every case and *unpublishes* any case with a critical finding, and a separate preflight command refuses to call an instance safe while anything would raise in front of an audience. **2. Check before you write, not after.** A script that writes the files and then prints "checks failed" has already published. Compute, check, and only then write. If a check fails, say plainly that whatever is on disk is now out of date and must not be handed out. **3. Allow an override, and make it cost something.** A blocked build with no way past it gets worked around. Require an explicit flag *and* a written reason, so the override is