I think you mean photos over 500kb
Anyway… concerning the point “I don’t know whether I have been attacked with lvl 7 or lvl 10 arbs”, on a replay you still wouldn’t know (I mean there might be a slight color change on lvl-up for troops, but that’s not enough to see their exact lvl either. And probably you would already “know” that in high lvl you won’t be attacked with lvl 1 arbs even without looking at the troop color tint).
Anyway, the thing with a replay is that there’s accuracy limits for everything being saved. Those inaccuracies add up quickly and may easily generate a completely different outcome on the replay compared to the original raid. E.g. the difference between a king barely surviving at a point with very low health, and the king dying there, that lies very close together. Even minimal inaccuracy will make the replay possible show a wrong death or wrong survive. And that is just one little example, many things can go slightly different. Also e.g. a hostile froster barely passing your across-path troops (slightly out of range) and moving further along his way, or barely reaching the first of your troops, slowing them, and himself staying in place to continue firing at your following troops, can change the further development of a raid. And can you garantuee that 10 quickly summoned knight at the start always pop up in the right order and positions and move in the exact same fashion (including small collisions between the troops)? Can you guarantee that a firebolt hits the exact same targets to set them on fire, will they run around in panic in the exact same fashion? What about mummies (spawned “near the king”? They have a “spawn radius” of about 1 tile, and even during the raid it’s not easy to guess where exactly one particular mummy will pop up, even less so for the replay. What about e.g. lag during the game, changing framerates, time of input execution, troop spawning, etc… in the replay this is most likely different…
I guess by now you get the point, an accurate and RELIABLE replay with hundreds of units, thousands of actions and reactions etc… is hardly possible. And an inaccurate, unreliable, possibly (most likely) completely wrong replay is no use.
All of that does not even think about video recording, space for saving the replay data, alternative visualization of replays or any other aspects.
What I meant in my post above was, you can reconstruct e.g. the time and place where the attacker died, from victory percentage and rest time - as he moves along mostly linearly through your base, you can just say, “if he reached 45%, he was shortly before reaching the gate (about right before the last barricade/blockade, in most designs)”, without further knowledge of the raid. And with the time left, you can see what waves where near the place where the player died, at the time of his death (as the defensive waves always appear in the same fixed order, at the same fixed times).
Sure, this doesn’t tell you which of your 5 archers and 2 ogres actually dealt the final, deadly blow, nor does it tell you “neither of them! it was the firebolt tower nearby!”. But this detail isn’t important, as it is a matter of luck / randomness, which of your units deals the final hit. Important is “this area is where most of my attackers have trouble and die, so this seems to be an effective combination there”. Also, you see which units (and roughly in which ratio) an attacker uses. If most people with e.g. archer knight cannon combo beat your base with 30s left, while those with knight froster ogre also beat your base but only with 2s left, then you know your base isn’t very effective against the 1st combo, but decent against the 2nd combo. Whether the attacker used 1 more or less knight on a particular raid, is completely irrelevant to you as defender.
So that’s why I say rest time (in addition with already existing information) would already help a lot for analyzing attackers’ raids. Sure, not every detail caught, but could be implemented easily and without any problems, and would help a lot already, while being compact enough not to waste space, time and computation power.