D635 - The branch that printed nothing

Status: measured Date: 2026-09-09

Two states, one silence

D634 found a 44% import regression in PPSA28061 that no instrument reported. The reason is one line, and it had a comment explaining itself:

// Nothing to say. The record already holds something as good, which is the ordinary
// outcome of a run that changed nothing.
Ok(Kept::NotBetter { .. }) => {}

NotBetter covers two states. The record holding something as good is indeed nothing to say. The record holding something better is a regression - and it printed for neither, discarding the previous status it was handed in the same pattern.

So a title that lost ground kept its old number in the file, said nothing on screen, and every run afterwards compared itself to the previous run and reported same.

What it says now

below the best ever recorded for this title: 26 imports and 334 calls, against 47 and 933 on 2026-08-23
reached entered where the record reached entered
this run had 20s against the record's 12s, so time is not the explanation
the record is not overwritten - it is a best-ever, and this run is not one

Four lines, and the third is the one that makes it usable. A shorter run legitimately reaches less, so a bare "you reached fewer imports" would cry wolf every time somebody tried a quicker run - and would be ignored by the third time. Printing both limits settles that reading before anybody has to go and look, and here it makes the finding stronger: twenty seconds against the record's twelve, and still short by twenty-one imports.

The last line exists because the natural next question - "has it overwritten my record?" - has a reassuring answer that the reader should not have to hunt for.

The rule, split out and tested

is_below_best is a pure function of two statuses: fewer imports, or the same imports and fewer calls. The printing is a thin wrapper over it (principle 8).

Standing is deliberately not consulted. Implementing a function the guest already called raises standing while moving no import and no call, so a run can be better on standing and worse on reach

The test asserts all four cases, including both negatives: a run equal to the record must stay silent, because a line that fires on equality is one people learn to skip, and that is precisely how the silence it replaces survived. A rule that only ever answered true would satisfy any one assertion alone.

Watched firing, and watched staying quiet

PPSA28061 prints it. PPSA21564, which is at its own best, prints nothing. Both run before this was written, because a guard nobody has watched reject something is a guard nobody knows anything about.