This is a very productive response.
Thank you. It's where I was trying to take the discussion about 1 or 2 hundred posts upthread.
I think roles are a result of the way the mechanics and fiction intersect. You can't talk meaningfully about roles without talking about mechanics; simply talking about the fiction won't do. Because in the fiction of Marvel Heroic RP there are skirmishers (eg Wolverine, or even morseo Spiderman and Daredevil) and there are "tanks" (eg The Thing, Colossus), but in the play of that game these characters do not occupy distinct roles because the mechanics don't distinguish between skirmishing and tanking at the level of mechanical minutiae. The distinction between them, rather, goes simply to colour and to fictional positioning for action declaration.
Because of the difference between AD&D, the middle editions, and 5e, both 5e and AD&D had a similar wide expanse of characters who could be equally effective with defensive tactics
<snip>
Much like the earliest editions, 5e is less dependent on build
<snip>
So 5e has no roles independent of a few class specific abilities. Just tactics.
I think this is where we disagree about AD&D, and perhaps therefore also about 5e.
In AD&D, in my experience, there are four Gygax PHB character classes who can be effective in melee: clerics, fighters, rangers and paladins. (Mid-to-high level monks also, perhaps, but I think there is relatively wide recognition that there are issues with the class design of monks.)
Thieves, assassins, MUs, illusionists are not very viable in melee. They default to poor ACs (unless at high level with a good item kit-out) and have poor melee attacks. Druids are something of an intermediate case, having the same AC issues but better hit points and to-hit.
In 4e, of the 8 PHB classes there are four who are perfectly viable in melee: clerics, paladins, fighters and warlords. Clerics and warlords will tend not be be very sticky, because of the default non-stickiness of 4e melee - in that respect they will resemble their 5e brethren. But they can play a tactical role every bit as important as that of a fighter or a paladin.
Rangers and rogues in 4e can be viable in melee if played with care - they can have good AC and do good damage but tend to be prone to fall over when hit too hard. (And don't have the self-buffing/healing options of a cleric or warlord.) They can play a defensive tactical role if required - I've seen this done by a sorcerer plenty of times, who as a primarily ranged striker is even less suited to this then a ranger or rogue - but they wouldn't be your first choice.
Warlocks are shakier again in melee, and wizards shakier still. In 4e, if your wizard is holding your defensive line by physical prowess then something has gone wrong. I think 5e is pretty similar in this respect. If the 4e wizard is holding the defensive line because concentrating on a zone or conjuration of some sort (in 4e concentration manifests itself by the requirement to use an action each round to sustain an effect) then that is of course a different kettle of fish, just as it would be in 5e.
Bounded accuracy has been mentioned.
4e is also fairly bounded, especially at heroic tier: the gap is 5 to 7, between an 8 or 10 stat and a 20 or 22 stat. A magic item might add another 2 to the gap. This is not very different from the gap in 5e between proficiency and an 18 or 20 stat, and no proficiency and an 8 or 10 stat.
The target numbers behave a little differently across the two editions. Typical AC in 5e ranges between (say) 12 and 19 - which gives success rates at the highest and lowest end of around 85% (+8 vs 12) and 5% (-1 vs 19). Matching up in the opposite direction gives success rates of around 40% (-1 vs 12) and 50% (+8 vs 19). In 4e, at 10th level typical defence is 23, which gives a success rate of around 10% (8 stat gives -1, +5 from level gives net +4) and 55% (22 stat gives +6, +5 from level gives +11, +2 from item gives +13). The variation in expected defences won't be as great in 4e as 5e - closer to +/- 2 rather than +/- 3 to 4.
It seems to me that the difference between 5e and heroic tier 4e isn't so much bounded accuracy as an overall drop in typical target numbers relative to typical PC bonuses, so that those who are strong come close to auto-success, and the 45-odd percentage point gap between the strong and the weak reduces success rates from near-auto to "a bit less than 50/50" - as opposed to 4e, where the strong have a success rate between one-half and two-thirds, and the similar gap reduces the weak to the neighbourhood of "natural 20 to succeed".
A further factor that I think hasn't been mentioned is that the default buff, in 5e, is advantage rather than +2. With percentage chances of success for the weak already higher than in 4e, rerolls have an even bigger affect. Eg +2 with a 10% chance of success takes it to 20% - you'll still probably fail. Whereas advantage with a 40% chance of success takes you to a 64% chance - you've got a real likelihood of success.
These changes to the maths open up certain viable options in 5e that are harder to activate in 4e or 3E (or AD&D, for that matter, where a MU's chance to (say) grapple effectively is mostly negligible). I still don't think that it means there are "no roles". I think the phrase "a few specific class abilities" is quickly glossing over a lot of fairly significant mechanical features of the game: healing and ranged AoE damage (the trad cleric and trad MU) being two of them.