Two kinds of problem, and only one of them is yours
Findings are now split by who can act on them. When the platform is the reason a check is wrong, it says so, tells you your half, and files the rest against itself.
A monitoring product that cannot tell the difference between "this machine has a problem" and "we could not measure this machine properly" will eventually report the second as the first. That is how a device ends up marked down for an audit that never ran.
This week's work splits those apart, and the line it splits on is not who is at fault. It is who can act.
Confidence
Some findings are about the estate. Others are about whether the reading can be trusted at all: a check that scored exactly zero because it measured nothing, a score that has not moved in weeks while the data underneath it changed, an analysis type that has never once produced a result on a device that happily produces others.
Those are now a separate surface called Confidence, aggregated by rule rather than by device. If the same integrity problem affects forty machines, that is one entry saying so, not forty findings competing with real ones for your attention.
Deliberate limit, and it is the important part: a confidence problem never touches the health score. A device whose audit is broken is not an unhealthy device. Conflating the two is precisely the error that put a zero into four devices' averages for months.
Service notices
Where the platform is the reason a check is wrong, saying nothing is not neutral. The finding still sits there looking like the customer's problem.
So when the cause is ours, you now get a notice that does two things: it tells you your half, the part you can actually do something about, and it says that we have filed the rest against ourselves. That second sentence is load-bearing. An MSP forwarding a report to a client should never be in the position of explaining a defect in their tooling as if it were the client's estate.
Notices only close when the thing that opened them is fixed. A notice we raised is a notice we own.
Whole-device readings, rolled up
Individual checks answer individual questions. Nobody was answering the question an engineer actually asks, which is "what is going on with this machine, taken together".
Every device now gets a synthesis: all of its checks read at once, with the connections between them stated. It sits at the head of the findings rather than in a separate place, because it is the first thing worth reading and it was not going to be read anywhere else.
That reading rolls up weekly from device to group to organisation to tenant, so the same question can be asked of a client estate as of a laptop.
The device list tells you both
A device now carries two markers, never one: its health, and whether that health can be trusted. They are opposite claims and a single badge cannot make both. A machine scoring 95 that we could not properly measure is not the same machine as one scoring 95 that we could, and the list now says which is which.