Unit experience

[sapi]

that's because the bad guys got transferred out leaving only the experienced guys.

[Sinner]

There's a relatively easy if somewhat laborious way to test this: edit the campaign start file so that there's a small stack of rebel peasants in a region belonging to Armenia, Parthia or Scythia - one of the ones that eventually get strong horse archers; then add one of those units with zero experience to the garrison of that region, picking that faction to play when you start the campaign. Use only this added unit of cavalry to attack the peasants.

Before each game, temporarily edit the stats of the unit types involved to set up the required tests since we're not after balance, but trying to intuit the experience mechanism. For all the tests set the peasants' armour and attacks to zero and up the cavalry's hitpoints to 10 - this should guarantee that you won't lose any cavalry, so we shouldn't need to worry about the effects of casualties on experience. In each test, try and kill all the peasants so that results are as common as possible.

Baseline test: Edit the peasants' morale to 16 so that they hopefully fight without routing, eliminating that as a factor in determining experience gained. Do not use missile attacks, thus eliminating that as a factor as well.
Missile test: Simply restart the campaign, leaving all stats as per the first, but this time use only missile attacks. If the experience gained for killing the same number of peasants is less than the baseline, then missile kills are therefore worth less.
Numbers test(s): Close RTW and edit the unit stats file, changing just the cavalry numbers, reducing them to whatever the minimum number the game allows. Then restart RTW and begin a new campaign. This will concentrate the kills among fewer troops, giving a guide to the number of kills required to level up. It may be advisable to use only missile fire and compare with the missile test instead of the baseline so that casualties do not become a factor. The test could be repeated with different numbers of cavalry to try and work out the thresholds for each experience level.
Routing test: Close RTW and edit the unit stats file, changing the cavalry numbers back to normal and editing the peasants' morale back to the normal zero so that they rout quickly. Then restart RTW and begin a new campaign. As per the baseline, avoid using missile attacks. The aim is to try and kill as many peasants as possible while they're routing rather than in combat. If the experience gained for killing the same number of peasants is less than the baseline, then routing kills are therefore worth less.

It's a bit of a bodged methodology, but it should give a fair guide to how experience is working. It does however leave one question unanswered: how do some units gain experience without a single kill? The only ideas I can think of are that units can gain experience simply by being in battle even if they do nothing, although I'd assume that this must be a very small random chance since in many battles my heavy infantry do nothing and I've never seen them gain experience. Alternatively a unit might gain experience if it suffers casualties even without killing - they will after all have seen the elephant for sure in this case. Or perhaps they fought against multi-hitpoint enemies and while they whittled away hitpoints, with each hit counting towards experience, they didn't kill anybody, hence the oddity.

All three possibilities could be tested with a similar test schedule to above, although it'd require two different units on the cavalry side, one as the test subject and the other to actually win the fight.

Random chance: Actually this one could be done with two of the same unit, just have one sit still and do nothing while the other wins the fight. You'd have to repeat this a lot of times, especially if the random chance is something ridiculously low like 0.1%. Personally I don't think this does occur, but it is a possibility so can't be ignored. Perhaps leaving this test to last will allow it to be left out.
Casualties test: Give the peasants their standard attack, but up morale and defence to 10 and hitpoints to 2, while dropping the test cavalry's attack to zero. This should force the cavalry to suffer losses without any kills, but try to guarantee it by provoking an attack from the peasants & then turning the unit away so they get attacked from behind. Make sure the test cavalry are not on fire at will and do not attack the peasants at any time. Wait until losses have been suffered and then get them away before sending in your second unit to kill the peasants. It might be advisable to edit the stats of the killing unit to ensure that you win the battle. This test might have to be repeated a few times, perhaps against multiple stacks with the same units, to be certain whether experience can be gained due to casualties or can be eliminated as a possibility.
Multi-hitpoints test: Give the peasants zero attack and defence, but up morale to 16 and hitpoints to 10, while reducing the test cavalry to a small size and giving them 10 hitpoints as well to avoid having to account for casualties as a factor. Attack with the test cavalry, but don't leave them in combat for too long, so that they should be getting lots of hits on the peasants, thus whittling away at hitpoints, but not get any actually kills. Withdraw the test cavalry and finish off the peasants with the second unit. If the test cavalry gain experience then it's based on hitpoints rather than actual kills.

[Kraxis]

Wouldn't it be easier to do the same, just in Custom Battle???

[Hellboy909]

Detailed examination of unti experience:
https://forums.totalwar.org/vb/showthread.php?t=40864

[Sinner]

Kraxis, I'd personally prefer to do the tests in the campaign setting since this is the only part of the game where the gaining of experience actually matters. There are indications that experience inexplicably alters between the campaign map and the battle map, with chevrons being lost or gained, perhaps due to rounding issues.