Thresholds published in advance, or the verdict is just a mood.

The rule

The band comes first; the score arrives into it.

The structural health verdict works because a composite of a given magnitude means a fixed thing that was decided in advance. The ordering is the entire safeguard. If the bands are published first, a score of 62 is a fact about the month. If they are set afterwards — even with complete honesty, even by an operator genuinely trying to be fair — 62 becomes whatever the month needs it to be, because the person setting the band already knows what happened and cannot un-know it. This is not a claim about integrity. It is a claim about the impossibility of judging a standard neutrally once the result is in hand.

The failure mode

Threshold drift is quiet, gradual, and always in one direction.

Nobody rewrites a standard in a single move. Threshold drift arrives as a series of individually reasonable adjustments: this month had unusual conditions, so the band was arguably too strict; the previous definition did not anticipate this situation; the metric behaves differently at scale. Each amendment is defensible in isolation and they accumulate in one direction, because the amendments are proposed by the party the standard constrains and only ever feel necessary when the standard bites. Six months of small reasonable adjustments produce a standard that certifies whatever the system is currently doing, at which point the score still exists, still computes, and no longer measures anything.

• Amendments are only ever proposed when the standard is uncomfortable, so the drift has a built-in direction.
• A threshold that has never produced an unwelcome verdict has not yet been tested as a threshold.
• Legitimate revisions exist — they are the ones made when the metric is not currently biting.

The wider pattern

This is the same discipline the whole system applies to itself.

The pre-commitment rule is not local to structural health. The QA layer requires that expected outputs be written as a contract before a scenario is run, because a test whose expected result is decided after the run is not a test. The scenario manager requires the expected behaviour to be recorded before the simulation. The gate ladder publishes its thresholds so that a capital state is never negotiated at the moment it binds. In each case the mechanism is identical: fix the standard while it is still abstract, then meet it or fail it once it is concrete. Structural health is one more instance of a principle the system applies everywhere it can.

How to revise honestly

Change the standard when it is not biting, and record that you did.

Thresholds should not be permanent, and the discipline does not require that they are. It requires that revision happens at the right moment and leaves a trace. The right moment is when the standard is not currently under pressure — a band changed during a month it would have failed is a band changed for the wrong reason regardless of the argument attached. The trace matters because a standard's history is what lets a future reader know whether a run of strong scores reflects an improving system or a softening ruler. A revision made in the open, in a quiet month, with the reasoning recorded, is a legitimate act of maintenance. The same revision made silently under pressure is the failure mode wearing the language of improvement.

The key idea

A standard you can move is not a standard.

The value of a verdict lies entirely in its capacity to be unwelcome. Any mechanism that lets an operator adjust the threshold after seeing the score removes that capacity while leaving everything else intact — the number, the bands, the language of governance — which is what makes the failure so hard to notice from inside it.