2026-08-21 - Run conditions and the discount on the headline (D181)
A verdict compared two traces without knowing whether they were
comparable. Two settings break the comparison and both are one line of
TOML: the wall-clock --limit, which means a faster machine
reaches further on the same build, and
default_return = "ok", which makes every unimplemented
function claim success.
The second is a reward hack one line wide. Every number improves at once - the guest stops checking, runs on, reaches imports it never reached - and nothing has been implemented. Anything steering by a call count finds it within a few iterations, not from malice but because it is the cheapest change with the biggest effect.
Recorded rather than forbidden. Answering
ok everywhere is a legitimate bisection technique and
principle 5 exists to keep that loop cheap; what makes it a hack is
doing it unlabelled. Conditions rides on the trace and
compare() reports what changed underneath a verdict instead
of refusing to render one - the numbers are real, they just are not
evidence about the emulator. Build is recorded and deliberately not
compared: it would fire on every release and drown the two that
matter.
Every run also prints what it stands on now - 787 of 933 calls answered by an implementation, 15% on stubs. A call count is progress only to the extent something real answered.
Demonstrated rather than argued. Same title, same build, one line
changed: same / nothing moved became
FURTHER / reached imports it could not reach before, with
the caveat printed directly under the verdict. The hack still works; it
can no longer be read without being told.