Since I started getting on-board video (Mobius), making it easier to tell just when the deployment event occurs (frame-by-frame analysis, also correlated with an A2 for most of the flights below), my observation has been that AT delays (DMS or RMS) can be fast, and CTI delays are more likely to be late.
Going through all my data so far, I have (these were all SD motor-eject flights):
AT DMS I280DM-14A (-2s drill): Fired @ 13.43s post-ignition (11.8s after burn-out @ 1.61s)
AT DMS I500T-14A (no drill): Fired @ 13.73s post-ignition (12.6s after burn-out @ 1.15s)
AT RMS-38/360 I218R-M (no drill): Fired @ 9.03s post-ignition (7.6s after burn-out @ 1.46s)
AT DMS I280DM-14A (-4s drill): Fired @ 12.30s post-ignition (10.5s after burn-out @ 1.77s)
CTI Pro38-4G I287SS-15A (-3s drill): Fired @ 14.67s post-ignition (13.1s after burn-out @ 1.58s)
CTI Pro38-5G I297SK-15A (-5s drill): Fired @ 11.53s post-ignition (9.8s after burn-out @ 1.70s)
CTI Pro38-4G I255RL-16A (-3s drill): Fired @ 16.40s post-ignition (14.8s after burn-out @ 1.56s)
So two of the AT's were about right (interestingly the two I drilled, also both I280DMs), two were very short (the two I didn't drill, one DMS and one an RMS "M" delay). One of the CTIs was on, the others were a second or more longer than what I had supposedly drilled them to, and that was with multiple turns of the drill, blowing-out the hole, cleaning the bit, repeating until no more loose delay chunks in the hole or on the bit.
I flew an I140W-14A (-2s drill) in December, but I didn't get video for that flight (the SD Card was still plugged into my computer
). The A2 data there suggests an early deployment, reporting 2.21s burn time, 9.7s coast-apogee and -1.4s apogee-eject (i.e. 8.3s burn-out to eject, 10.5s ignition to eject). That flight came in 230' less than the 2720' OR predicted, when my flights generally go about 10% over what OR predicts, so that would also seem to agree with an early ejection (about 500' lower than the flight probably should have gone with the proper delay).