Threshold Ranking

Thresholds are ranked, not guessed. The budget for instrumentation is finite, so the ranking method matters more than the list it produces. A ranked threshold list survives contact with a real deployment. A guessed list rots the first time the fleet changes shape.

Why guessing fails quietly

A guessed threshold list has a tell: it mirrors the org chart, or the last incident, or the loudest stakeholder. None of those survive composition drift. The fleet changes weekly, the doorways change with it, and a list with no method behind it has no way to update itself without becoming a different list of guesses.

The ranking method is the update mechanism. The list is just its output.

The method

Four properties, weighted by consequence. Irreversibility first: can the action be undone. Blast radius second: who is touched that did not consent. Authorization distance third: the gap between granted and used permission. Rate of change fourth: how much signal lives in a sudden deviation.

Each doorway gets scored, the scores get ranked, and the instrumentation budget flows down the ranking until it runs out. The cutoff is honest in a way no hand-picked list can be: it says, in public, here is what we are watching and here is what we are not.

The cutoff is a decision too

The bottom of the ranked list is the uncomfortable part. Every budget ends somewhere, and the doorways below the line are the ones nobody is watching. The method's value is that the gap is explicit, owned, and revisited on a schedule.

The ranking method ships with the harness. The list it produces today is already stale. The method is not.