Refusal is a third state: completeness must be asserted, not assumed

[h4x0r]

From research-method §1.B, Absence read as evidence — the system said "I don't know" and the
record says "no":

Refusals counted as zeros, repeatedly. A total assembled from only the query windows that
succeeded, when the ten that refused were the busy ones — understating by 30×. A producer
that filtered out refused readings and wrote an artifact with no trace that any had been dropped.

Defence: a refusal is a third state. Count it, carry it into the artifact, and make any claim
resting on completeness assert refusals == 0 rather than assume it.

peira already applies exactly this rule to itself and nowhere else. GateResult::Unassessed exists so
that "could not reach a verdict" never renders as a pass, and peira gates on an absent vault exits
2 rather than 0 so that "found nothing" cannot be read as "could not look" — control C in
docs/validation.md exists to prove that distinction holds.

What the graph cannot currently express is the same distinction about a cited tool. An
Observation records what was seen; nothing records that 10 of 40 queries behind it were refused.

Proposed PEIR-LINT-COMPLETENESS-ASSUMED — a claim resting on an observation whose instrument
declares no refusals: count, where the claim asserts a total, a census or an exhaustive search.

Absent is not zero. A missing refusals: field must read as unmeasured, never as none — which
is the same reason a missing author: leaves PEIR-LINT-SELF-GRADED silent rather than guessing.

Depends on #8.

Source: ~/.claude/skills/research-method/SKILL.md §1.B.