A cap yields a floor, never a total

[h4x0r]

From research-method §1.C, A part measured and named the whole:

A vault priced from 1 of the 52 markets it holds. A breadth sample drawn only from the windows that
had not hit the cap, making a 112,000-call public service look like a 4-user private arrangement.

Defence: query the full population first, then window it. State the denominator. A cap yields
a floor, never a total.

And §4 of the grading spec, on which direction the bias runs:

A capped or paged walk yields a floor. State which direction the cap biases the conclusion; a
floor can support "broad" and can never support "narrow".

peira already enforces the sibling rule — #12's PEIR-LINT-WINDOW-EDGE-AS-ONSET catches a temporal
window's edge being read as a behaviour's start. This is the quantitative case: a truncated result
supporting a claim about a total.

Proposed PEIR-LINT-FLOOR-AS-TOTAL — an observation declaring truncated: true supporting a claim
whose quantifier is universal, or whose language asserts a total or an exhaustive count.

The asymmetry is the useful part and should be in the remedy text: the same truncated evidence is
legitimate support for "at least N" and "widespread", and illegitimate for "only N", "all", and
"no more than". The lint should say which direction the claim may be restated in, not merely refuse it.

Related: #12 (the temporal twin, closed), #8 (truncated: lives on the Instrument).

Source: ~/.claude/skills/research-method/SKILL.md §1.C; claim-grading-spec.md §4.