tbf the dev end of Deadlock is effectively a black box, barring the proffered tidbits (which we, in good faith, meekly assume to be exhaustively true), and so OP's speculation isn't entirely unreasonable, certainly not to the extent that many posts in this thread seem to suggest. For what it's worth, I play almost exclusively against bots, which I'd wager is unlike most Deadlock players, and so I'd like to add my disproportionately-unique experience to the conversation:
After the star system came out, I added Seven to my roster for the first time, as one of several prioritized heroes. Game assigned me to Seven for my first match back after a modest break, i.e. my first game after the star-system dropped. Anyways, I played that first game, had a blast, rated it five stars. Lo and behold, for the next several games, I kept getting placed as Seven, despite having 4-5 other heroes on the same priority tier, all of whom I had played many more matches as, but none of whom had any star-system submissions at that time. A second tidbit: since that unprecedented string of games as Seven, I've had several "Dynamo stretches" which seemed, at least anecdotally, to correspond with my continued 3-to-5 star ranking of matches, and which I think only ended after I bit the bullet and told The Director that, damn it, I was tired of laning against McGinnis as Dynamo by providing a post-match assessment of 1-2 stars. And I think that broke the streak the very next match.
I'd like to once again emphasize that mine were all bot matches, and they were the only matches that I'd played -- against bots or players -- since the star-ranking-system had been introduced. As such, the matchmaking AI had more flexibility to assign me to whatever hero it (thought that I) desired, since there were only 5 other players to satisfy as opposed to 11 in a PvP match. And thus, in a sense, bot matches offer we matchmade sheeple a "purer" insight into the abovementioned "black box"... in theory. So, IMO, there may just be something to what OP said.
...and now to actually get tin-foil-hatty for this final textblock: I played a KILLER game against bots as Seven, in which four players DCed within the first few minutes. Didn't realize it for a bit, then noticed that only myself and a virtuous Haze remained. Long story short, we met up, fought, pushed, and eventually won. It was probably the most entertaining match of Deadlock that I've ever played, and so it got five stars from me. Following this highly-rated 2v6 bot match, for a surprisingly lengthy stretch of matches, at least one player would DC immediately or within the first few minutes. Which seems statistically unlikely, cf. mine observed rate of DCing players in bot matches. End of day, doesn't matter, though it'd be prudent not to assume that Valve only wants data from player-experience-optimized/predictably-structured matches. That wouldn't provide an especially varied dataset, would it..?
ratings are saved as feedback to somewhere else and used for different purpose.
