Suggestion: single game performance ELO rating when analyzing games.

Not sure if this is even workable with single games, but if it is at all viable I would love to see how well I played by ELO rating in certain games (after someone runs the lichess computer analysis), similar to how one gets a performance rating ELO number after playing a tournament.

If this is unworkable as a single number for a single game (as I suspect it is), I would still be happy with a ratings range of perhaps up to 150 or so points, accurate something like 19/20 times as is standard with political polling.

That ain’t work for various reasons. Unnless one accepts an error of 500 or so.

I believe it is feasible. If I play an unrated and unknown opponent in an open tournament, I get a feeling of opponent's strength even before the game is over.

Dr. Regan has researched a similar subject, although you'd need a collection of thousands of games:
cse.buffalo.edu/~regan/papers/pdf/RMH11.pdf

Interesting paper.

With the thousands of analyzed games on record, what kind of accuracy do you figure lichess could manage in this regard, or would deeper computer analysis be needed to get a viable and narrow enough range that people would find useful and interesting?

Or would this simply be too expensive to do with the costs of the additional computational power? Perhaps if expense was an issue it could be a "perk" for donating members?