I'd regularly find my to hit rolls in attack chains for alot of attacks consistently hover in 3-5% at times, rarely straying, actually. I'd look at to hit roles all the time. I had other moments where some mob rolled 33.5%, 34.7, then 32% in all three attacks he had on me(hitting EVERY time).
Not saying the to hit roles are biased but really if one attacks fast enough strangely I noticed very strange numbers rolling. I wouldn't say at all in fact that the random number generator was evenly distributed. Each 1% should have a 1% chance of rolling, but just from play experience overall I found that not to be the case. This is from play experience and looking at the numbers in actual combat at the time the game was running.
Edit: I would say, more the hit roles were strange, at best.
If you still have combat logs, I would be more than happy to analyze them for tohit roll anomalies. What I discovered based on analyzing all of my tohit rolls was:
1. Every possible tohit roll showed up equally often within the limits of statistical variance, with one exception: 100.00 and 0.00 showed the obvious inevitable roundoff issue (they each showed up half as often as all other rolls). This was not statistically significant for players.
2. Every possible tohit roll followed every possible tohit roll within statistical limits. In other words, whenever you rolled 5.12, the odds of rolling any of all possible rolls was identical.
3. Every possible range of tohit rolls followed every other possible tohit roll range within statistical limits. In other words, if you rolled under 50%, it was equally likely the next roll would be under or over 50%. The same statement could be made for odd and even rolls, for rolls within the ten decimal bands, the one hundred single point bands, and the 5% bracket bands.
4. Rolls 5% and under and 95% and over happened with exactly the correct statistical likelihood.
5. Rolls on different servers showed the same even distribution, for all the servers examined.
6. Rolls that occurred before or after streakbreaker events showed the same flat distribution.
7. Events designated with a percent chance to hit occurred as often as their defined chance specified relative to all potential occurrences.
8. Using the tohit rolls as a binary source created a file that was basically incompressible.
9. The difference between successive tohit rolls modulo 100 showed the same distribution as the tohit rolls themselves.
10. Plotting tohit rolls using pairs and triples as two and three dimensional vectors showed no signs of clustering.
Also, the developers told me the game engine used the clib rand() or something similar, which while some versions are known to have some cryptographically significant biases, contain no biases that an MMO player could possibly detect through gameplay. The only way for players to see what you were seeing would be if the tohit system used a broken rand. And a broken rand would be something I would have detected almost immediately.
Its like this: I can conceive of someone bringing loaded dice to a game. If the dice consistently rolled 5s more often, you could believe the dice might be skewed. But if someone suggested during a game that the dice were consistently rolling lower numbers when a bus drove by, and higher numbers when a Honda commercial was on television, you'd doubt their observations because dice don't work that way. They can be fair or skewed, but they can't choose when to be skewed. In the same way, the CoH tohit system can't roll different random numbers depending on whose playing or what powersets are being used or what the condition of the characters is, because RAND() cannot be aware of those things.
You might have been seeing *something*. Years of people reporting all sorts of goofy behavior involving defense turned out to be the simple fact that people were relying on the value of lucks being correct in their description. Because they were not, the game did not behave as they expected it to. And because of that, their minds began trying to come up with an explanation for the odd behavior but because they continued to trust the inspiration label, their mental model theories *could not* match reality. So they became bizarre instead. People reported to me that lucks did not work at all, unless you overstacked them. People reported to me that lucks would sometimes simply stop working a few seconds after using them. Some people reported that using reds would cancel out a luck. Some people said that some lucks worked fine while others were "broken" and you could tell which ones were broken because they were a slightly different color (spoiled?).
The truth was that lucks that were labelled as being +25% were actually +12.5% (the others were all mislabeled also). That's all. And once I discovered that, at least some of the observations I thought were insane started to have a potential explanation. If you think two small lucks will floor a minion (+50% defense), when they actually cause the minion to start hitting half as often (from 50% chance to hit to 25%), you might not notice that enough to realize the lucks were doing something. Then you stack two more and suddenly the minion stops hitting all together (because if a minion is hitting one in 20 they might not even live long enough to get the one lucky hit). Ergo, you start believing Lucks don't start working until you overstack them. You might start believing that sometimes they do work (because you pop them and get lucky and the critter starts missing) and then they stop working (because the critter hits twice in a row). You might start seeing "hits one quarter of the time" as "sometimes not hitting at all and sometimes hitting a lot" and start auto-correlating that to other events: the hits start coming when I pop a red, or a green, or the cat walks into the room.
*All* of those weird luck observations were the result of a text typo that no one including the devs caught until I tested them (very pre-real numbers). The players were seeing something, but I guarantee they were not seeing what they thought they were seeing. And interestingly, within a couple months of my posting the lucks and insights article, all such player reports basically stopped.
So I never assume players are completely imagining things. I'm simply saying that whatever you thought you saw, the one thing it almost certainly could not have been was a skewed random tohit roll.