Evidence anchoring: how to read [A], [P] and [?]

Anchoring is the degree of evidence that backs a statement. It does not measure the confidence of whoever writes it — it measures what anchors the statement in the reality of the code. A statement only rises a level when it gains citable evidence, never when the author “feels more sure.”

The three levels

• [A] Anchored — direct, citable evidence: file and line, a green test, a migration or a commit hash. The obligation is to cite the evidence on the same line. An [A] without a citation is invalid and the sensor downgrades it to [P].
• [P] Presumed — inferred from patterns, names or framework conventions. It’s plausible, not proven. The obligation is to generate a tracked validation item — it’s explicit debt, not certainty.
• [?] Open — not determinable from the available artifacts. It blocks marking the unit as done and becomes a question for a human to answer.

Why ASCII tags

Markers like [A] are greppable: a sensor can scan the text and check, deterministically, whether every [A] carries a citation and whether no unit marked as done still carries a [?]. A colored emoji serves only as visual support in reports — never as the data the machine verifies.

The golden rule

When in doubt, use the lowest level. An honest [?] is worth more than a misleading [P], because engineering begins when you admit what you don’t yet know — and the system stops pretending to know what it doesn’t.