C.2.8:4.5 - Reading results, estimates and robustness

A design walkthrough, an actual reading and a formal-model estimate can concern the same characteristic. State which grounds the assertion and the conditions at which it applies.

An observed response establishes the structure recovered in that trial and can ground an estimate of what is extractable. One success does not establish a maximum or reliability for a population; one failure does not establish impossibility. A computational procedure can support an exact bounded result when its operation and selected structure make that result derivable.

Repeat or perturb a reading when the receiving use needs a claim about robustness or transfer. Preserve the first response before giving a changed condition or further help. Success under the new condition strengthens the corresponding changed-condition claim; it does not add structure to the original result by itself.

For a cold-reader check, obtain the instruction from the publication, its declared prerequisites and the knowledge presupposed for the intended reader. The task may supply case facts. If it supplies the missing explanation of the instruction, the response shows use of that added explanation. Use F.19 to repair the publication before claiming that the publication itself supplied the action.

Absence of human validation restricts empirical claims about people. It does not restrict ESI to computational observers or prevent a conditional author estimate about a human reading situation.