411. The record could not see the work
2026-09-04 - directed, while the Agc probe set is out for hardware
What was done
Closed the gap worklog 409 and 410 both ended on. A day that took
PPSA02664 from 35 unanswered functions to 20, and from
914 stubbed calls to 32, moved no recorded number at
all: standing is an integer percentage of calls, and 0.22%
and 0.008% both round to 100.
The fix was not more precision. A percentage of calls belongs to whatever the guest loops on - this title spent 12,924 calls in one wait - so it swings on where a title spins while the thing a person acts on barely moves. What moved is distinct imports called with nothing behind them: it counts functions, is stable against a hot loop, and is literally the work list.
Status::unanswered now records it,
answered() ranks on it, and the table has an
Answered column. PPSA02664's honest record reads
197 imports, 177 answered.
The bug found on the way, which mattered more
Three copies of the ranking existed and two had already
drifted. beats decides what gets recorded,
frontier decides what a shim shows,
render_markdown decides the table - and neither of the last
two gained frames when D558 added it this
morning. The rung I added today ranked one way in the record
and another in the table.
frontier's own doc had already warned about it, in as
many words: a table that disagreed with the thing deciding what to
record would be the more convincing of the two and the wrong one.
It was right. There is one ranking_key now, and a guard
that the three cannot disagree - watched failing by putting each drifted
copy back.
That is the second time today a warning written into this codebase turned out to be describing something that had already happened (D523's per-thread record was the first).
Where it ranks
Below imports, which is D182's shape a third time: a
guest that gets further calls more imports, and some of
what it newly calls will be unimplemented - so a run can reach further
and have more unanswered. Ranked above imports, going further would
report as going backwards.
Optional, and why that matters
None is not measured, not nothing
unanswered. Collapsing them would let a stale record claim a
perfect score and refuse every honest run after it - the exact trap
propping sprang this morning, when 33 records deserialised
as honest. An unmeasured record ranks as though nothing was answered,
which is not a claim about the run but about the record, and it
self-heals on that title's next run.
Guards
Nine, each watched failing: answered dropped from the
ranking; answered ranked above imports; unmeasured reading
as nothing-unanswered; the count taken from calls instead of functions;
the count taking every import rather than the unanswered ones; the value
never reaching the record; and the table and the frontier each keeping
their drifted copy.
Surprise
Every record in the table now shows a dash. That is correct - none of them measured this - and it is also a plain statement of how much of the compatibility record is older than today. It will fill in one title at a time as each is re-run, which is the honest way for it to fill in.