Written guidance, however correct and however clearly flagged as required reading, only works if whoever hits the problem already believes an answer exists somewhere and decides to go find it. A problem that has been classified as new (something nobody has apparently dealt with before) doesn’t produce that belief on its own, even when the same problem has already been solved and written up in detail, more than once, in exactly the place a search would have found it.
Why a document sitting there isn’t enough
The failure isn’t ignoring a document known to exist. From inside the moment a problem looks new, no document is believed to exist at all, so there’s nothing being ignored. The step that’s missing is the one where someone stops and asks whether this has been solved before. That step doesn’t fire automatically just because the answer would help. Being told directly that a problem has been solved before doesn’t reliably fire it either, if the response to being told is to try something new anyway rather than to go look at the record of what worked last time.
What actually closes the gap
The only thing that reliably closes it is guidance that arrives automatically at the exact moment the risky action is about to happen, rather than guidance that has to be remembered and retrieved. A refusal placed directly at the point of the action, stopping it and naming the safe alternative right there, needs no one to recall that the alternative exists, and no one to have decided in advance to go looking for it. The wrong reach itself produces the right answer.
Building the gate well
A gate like this holds up better when it keys on the general shape of what makes an action dangerous rather than a fixed list of specific names. A new but similarly shaped case then gets caught automatically instead of waiting to be added by hand after the fact. Matching on something as loose as a plain substring of a name is a known failure mode: a differently pluralized variant of an already-covered case can slip through a substring match untouched. And a gate like this should be proven before it’s trusted, deliberately, against a harmless target: confirm the block fires, revert it, confirm it’s open again, rather than ever being tested by actually attempting the unsafe action it exists to stop.
The limits of the claim
A structural gate only ever covers the specific point it’s built into. It doesn’t generalize to every other place the same underlying risk could show up. A different tool, a different transport, a different kind of silent failure each need their own gate, or their own judgment, and a gate that works well at one point of action says nothing about whether the next one has been covered yet. Placing guidance at the point of action raises the floor substantially. It doesn’t close every gap at once, and treating one working gate as proof the surrounding risk is handled is the same mistake this whole approach exists to correct.
The night this was learned on, including the detours that preceded it and the different near-miss that showed the fix’s own limit hours later, is told in full in Act 28, Guidance that never fires on its own.