Posts Tagged ‘false positive’

The Ockham Sound Analysis Criteria

September 16th, 2015 No comments

Yesterday a headline on a tool vendor blog caught my eye “… meets NIST high assurance standards”. What is this high assurance standard that I had not heard about before?

It turns out to be the Ockham Sound Analysis Criteria (which is not a standard, but is sort of connected to assurance at some level). I have been following the NIST group (now known as SAMATE) behind this work since their first meeting (the only one I have attended); their Static Analysis Tool Exposition (SATE) work is a great idea, but I imagine it has been an uphill battle convincing tool vendors to publicly expose the strength and weaknesses of their tools.

Passing the Ockham Sound Analysis Criteria requires that a static analysis tool detect “…a minimum of 60% of buggy sites OR of non-buggy sites.”, with no false-positives (which I take to mean that no incorrect warnings generated, i.e., the tool cannot incorrectly say “there is a fault here” or “there are no faults here”).

The obvious low cost tool implementation is to pattern detect the known problems in the known test suite (called the Juliet test suite) and output warnings about them. The only way for SAMATE to stop companies doing this (or at least tuning their tools to pass the suite) is to regularly change the test cases used.

I think I understand the rationale being the no false-positive requirement (SAMATE are using the marketing term “sound”). NIST want static analysis tools to be usable by people who don’t know anything about software; a strange idea I know, but the Nuclear Regulatory Commission have wanted to do this in the past.

Not being willing to accept false-positives kills innovation. New analysis techniques invariably start out being unreliable and improve over time.

I suspect that few vendors will have any interest in claiming to meet the Ockham Sound Analysis Criteria (apart from the ones that pattern match on the tests to satisfy some contract requirement). There is too much downside (new tests could put a vendor in the position of having to make a big investment just to continue to meet the criteria) for almost no upside (does anybody make purchase decisions based on this criteria?)

I think that the tool vendor (TrustinSoft) found they could make the claim and being relatively new in the tools market thought it might mean something to customers (I doubt it will, as their sales people are probably finding out). Of course what customers really want tool vendors to tell them is that their code does not contain any problems.