What's up with this anaylsis "best move"?

After I reached  this position (black to move) with the black pieces in a blitz game tonight:

I made the natural/obvious move Bc4. 

Scanning the black pieces quickly once, and the legal white moves quickly after the fork, this seems to be the best move, and the engine confirms when I'm in the 'Analysis' tab with 'Self Analysis' checked.

However, in the "Review" tab, the engine says that the best move is actually letting the knight retreat and losing the QS castle after Rd8 and a rook exchange (albeit only by 4 cp better).

Why the discrepancy between the "best move" in the two tabs?  Shouldn't it be consistent / the same analysis?

Which is computing further / correct?  

Sorry if this is OT here, not sure if "Game Analysis" is just... game analysis or about the site game analysis feature as well.

Just to be on the safe side - someone go ahead and tell me why the fork is not clearly, clearly better!

The fork is clearly better.

I would guess that running "Game Review" on a single position produced the anomolous results.

I never tried that before. I couldn't even get it to run.

When I found the game, and ran the Game Review, it said that Bc4 was the best move.

Ok, freaky.   I guess it's not deterministic!   Because I still get a different result.   Also I since found a position or two where move <A> will say "excellent, but <B> is best", and move <B> says "excellent, but <A> is best"  O_O.

The computer evaluation in the Analysis tab is mostly reliable. (Only mostly, because the depth of the engine analysis is appallingly low).

On the other hand, the "Review" is complete nonsense, no one should use it. Chess.com's attempt at creating an AI that reviews games and tries to teach chess is just ridiculous, chess is way too complex for a simplistic program to understand it, let alone teach it.

The labels "excellent", "brilliant", "best" are just gibberish.