D059 - A third kind of record: measurements, which assert nothing
Status: decided, on the user's suggestion, correcting a limitation in the earlier design.
Every record until now carried a judgement, and a judgement needs an
expectation. Where a document supplies one that is spec;
where nobody does, this project supplies it and marks it
assumed.
There is a third option and this program did not have it: record what happened and assert nothing. A duration is a fact. It needs no expectation and does not go stale when a guess turns out wrong.
The earlier objection - that timing a sleep means picking a
threshold, and picking one is inventing a specification - was aimed at
the wrong thing. It argues against asserting on a measurement,
not against recording one. OBS|measure records
carry a quantity, a value and a unit, and no verdict.
Assertions stay. A loose bound marked
assumed is still worth having, because it is correctable
and because "no expectation at all" is not a stronger position than "a
stated guess". What changes is that the number sits beside the verdict,
so correcting the guess from hardware is reading a figure and editing a
constant rather than re-deriving what the expectation should have
been.
The measurement is the calibration. Forty-seven
checks are assumed and none is hardware. That
is the ratio a console visit has to move, and a measurement makes each
move cheap and unambiguous.