Making it usable
Cutting the noise
Sixty one per cent of one real estate's findings were true, unarguable and not work. A list where the ones that matter are hard to find is a list nobody finishes.
A hole in a package nothing imports
Half of every dependency finding on a real estate is exactly that: 8,769 of 17,236, against 1,642 the code demonstrably reaches. The default view leaves those out, says how many it left out, and offers one click to put them back. A finding with no verdict at all still shows, because nobody having worked it out is a different claim from unreachable.
A rule the engine itself is unsure about
86% of code findings carry low confidence from the engine. Where the severity is also low, that combination is an observation rather than a problem, and those are graded to info as they arrive: kept, searchable, out of the way. Only where both are low. Low confidence about something serious is exactly what an engine's uncertainty flag is for.
Output nobody serves
On one repository, 22,048 of 22,054 code findings were one low confidence rule on one kind of file: the header of an HTML report a test runner writes fresh on every run. The rule is not wrong, and nobody serves those pages. The code pass takes a list of paths to skip, as a setting rather than a constant, and names the list in the scan log whenever it uses it.
A password a person chose
Three filters were throwing real credentials away. "Passw0rd!" is nine characters at entropy 2.95 and was dropped; "your_password_here" is eighteen characters at 3.46 and was kept. A password a person picked is short and low entropy because people are, and a placeholder is wordy because it is trying to explain itself. Where a finding was found by the name of the variable holding it, length and entropy no longer apply. Where the shape is the only evidence there is, nothing changed.
Nothing deleted, nothing silent
Every one of these is a filter with a count beside it and a way back. A filter nobody can see is indistinguishable from a scanner that missed something, and that is a worse problem than the noise.
What it does not do
The claim is not that any of this is wrong. Most of it is true. The claim is that it was making the findings that matter harder to find, and somebody who wants all of it back is one click from it.
Every feature page carries one of those. The comparison page scores the same ground against five products people pay for, including the rows that go against us.