D542 - Six records said nothing was known while holding a measurement, and the axis closes
decided - 2026-09-04
The plan's one open question: known_by was wrong on an
entry for weeks and nothing noticed. Check 10 is gated in one direction
- a measured entry may not list open questions. Is the
converse worth a gate?
Check what it would fire on before building it was the instruction, and it was the right one, because the obvious rule is wrong.
The obvious rule fires on 22 and is wrong about 16
An entry with a hardware measurement may not be
assumed. Twenty-two entries are in that state. Sixteen
of them are correct: a console answering one behaviour
does not establish the rest, and check 10 exists precisely to stop one
measured fact promoting an entry past the questions it still lists.
known_by describes the entry, not its strongest single
fact.
So the rule would have been a guard that fires mostly on things that are fine - which is a guard somebody weakens the first time it costs them an afternoon.
Six are a contradiction, not a judgement
The other six list no assumptions at all, so their records print the sentence an entry gives when it rests on a guess and itemises nothing:
Nothing about this entry has been established.
Two lines above, in the same record:
Measured on hardware: obSCEne ... reported pass, value ...
sceKernelGetSystemSwVersion sceGnmDispatchDirect sceGnmDispatchInitDefaultHardwareState
sceSysmoduleIsLoaded sceVideoOutGetResolutionStatus sceVideoOutSetFlipRate
That is D537 again - reasoning in the code, nothing in the record -
and it survived five ticks of auditing this exact
thing. questions ranks by call count; these have
between none and eleven, so they sat at the bottom of every list read
from the top. The audit had a blind spot shaped like its own
ranking.
And the reasoning waiting in the code was not thin.
sceKernelGetSystemSwVersion documents the whole structure,
that the number is not the firmware, that obSCEne
measured 13.090.001 across three runs while the same
console's banner says 12.40, and that wiring it to the firmware would
answer a value hardware refutes (D420).
sceVideoOutSetFlipRate discards the rate
entirely - it validates the handle and throws the value away -
which no reader of the record could have known. All six are now written
down.
The guard is the narrow one
an_entry_holding_a_measurement_does_not_say_nothing_is_established.
Not a claim about how entries should be labelled - a claim that a record
must not contradict itself. Made to fail by emptying one of the six.
It says what it cannot do: it does not look at the sixteen, and it
cannot see a measurement described in prose that does not begin
Measured on hardware:, because that prefix is what the
absorption writes and this checks the absorption's own output.
The axis closes here
Six ticks, D537 to D542, on the list this project hands to a console.
What it bought. questions --premises,
which turned a queue nobody could see the shape of into 142 sentences
with their weight attached. One question that was written 149 ways is
now written once - a fifth of the list. Nine claims that were
false are gone, and the one they were hiding worst
(sceKernelWaitSema ignoring two of its three arguments) is
recorded. One ask was withdrawn from the console queue
because hardware had already answered it and the entry said so in its
own open-question list. Six records stopped contradicting themselves.
Seven guards now hold all of it, and every one has been watched
failing.
What it cost. Six ticks and no movement on the wall, which has not moved since D528 named it as needing a console. That is the honest accounting, and the defence is narrow but real: the ask list is the handoff to hardware, hardware is the only thing that can move the wall, and a handoff that asks for a function that does not exist, or for an answer already given, spends the scarcest resource this project has on nothing.
What is deliberately left. The sixteen ambiguous
labels above - known_by is a judgement there and check 10
already governs it. Three terse arity questions and two unpunctuated
ones: cosmetic, and saying so is cheaper than fixing them. The 29
records that still say nothing has been established hold 86 calls
between them and no measurements, which is the honest state of a
declaration nobody has exercised.
The next axis is not this one. If it is the ask list again, it should be because a console run came back, not because the prose can be improved further.