[noggin]

I keep meaning to do this with VALID. I'd need to ingest a VALID test clip in FCP or Avid and trust that the Kona capture cards we use don't do anything nasty (the audio will come in embedded in the HD-SDI) If I then play out the recording via HDMI at 1080i25 (or possibly 720p50) and assuming our HDMI to HD-SDI converters only add a small delay (and I don't route through a synchroniser) I could get a basic VALID answer (ignoring AC3) for PCM stereo.

I'd be interested in the results of this for Pi (as well as other platforms).
There is a patch in fernetmenta's branch that adds an extra one or two (display) frames of latency to audio (depending on whether renderer is double or triple buffered).
I have a suspicion we should be doing that (especially after recent VideoPlayer changes have added at least one frame of latency to video).
Would be nice to know if the right number to add is 0, 1, 2, 3...

If you do test, then running display at 24Hz and 60Hz would be interesting (possibly the 60Hz case will be more accurate if the error is, say, two vsyncs).
Also test with "adjust display refresh rate to match video" enabled/disabled (which could affect A/V latency).
And on Pi run with OMXPlayer enabled/disabled (different code paths so potentially different settings).
Testing a nightly build would be more useful (the answers may well be quite different from stable builds - but we want a fix for the future, not the past).

Suspect it will be 1080/50i or 720/50p only I think I only have 50i VALID available. It's not something I'll be able to do tomorrow. Will take a bit of work to get everything sorted - and to ensure I'm not adding delays elsewhere.