So, if you've ever watched TV, you know that an expert with a microscope can tell if a mangled bullet from a crime scene came from a certain gun, or if the casings match.  Or can they?

Seriously, these guns are mass produced.  There were 10,000 barrels made on the same production run.  Bullets are soft, malleable, and get squished when they hit things.  Does anyone know what degree of accuracy is actually common?  I'd buy "This bullet was fired through a S&W barrel, four grooves, probably 6" long."  I'm sure there's enough distinctiveness in the combination of groove depth, land width,, pitch, and rifling shape to make that determination.  But to say "this damaged bullet matches the test shot from the defendant's gun with 100% accuracy"?  No way I buy that.

Likewise with the casings.  S&W mass-produced tens of thousands of J-frame revolvers in .38 special last year, largely using the same design they have for years, if not decades, before that.  How can there possibly be enough distinctiveness in a firing pin strike to identify one gun amongst hundreds of thousands?

Anyone have an insight into how this really works?

There are always imperfections in the barrel left from the manufacturing process.  These imperfections have a reasonable chance of leaving marks on bullets in a consistent pattern.  When testing a gun for matching, the lab tech fires 2-3 rounds.  They compare the rounds fired for consistent markings that make the pattern unique.  If there are no consistent marks to compare, then matching the evidence bullet to the gun is less likely.

The type of ammo used can improve or reduce the appearance of the unique marks, too. 

Since the bullet used for evidence is usually hitting a solid target, it is going to be less than perfect and more difficult to match with a test round.

So, the answer to you question on whether ballistic bullet matching is conclusive or not is ....  Maybe.   If the firearm tested actually creates a unique pattern consistently, then it is possible to match another round if that one is in good enough condition.

You know, fingerprint matching doesn't look at the entire print.  The process looks for a number of markers which taken together make the print unique.  Since so many prints left at crime scenes are partials, that's the best way to find matches.

"If you've ever watched TV" pretty much says it all.

I came to the conclusion long ago that matching a bullet to a specific barrel is highly unlikely. Even the tooling marks mentioned by MD are going to be consistent across an entire production run. At best they can say that the profile matches a specific brand of firearm, unless there is a unique flaw in the barrel.

Mythbusters!
Get 2 guns from the same lot and see if a lab can determine which gun shot which bullet.

"If you've ever watched TV" pretty much says it all.

I came to the conclusion long ago that matching a bullet to a specific barrel is highly unlikely. Even the tooling marks mentioned by MD are going to be consistent across an entire production run. At best they can say that the profile matches a specific brand of firearm, unless there is a unique flaw in the barrel.

Yup.  I would think the best potential for matching something to a particular barrel would be damage to the crown on a barrel.  Last thing that touches the bullet on the way out.  Also easiest thing to damage and leave a unique mark on the bullet.  Some food for thought.

There's going to be more characteristics in a barrel besides manufacturer tooling marks I can think of...
differences in the chrome plating
corrosion
wear and tear after hundreds/thousands of bullets
leading and copper left in the barrels
damage to the crown
damage from cleaning

And a lot of these differences are microscopic.  If bullet matching has been used in courts this long against thousands of lawyers and expert witnesses, then it's probably accurate.  Not 100%, but very good chance.  And you're not comparing it to hundreds of thousands of barrels out there, you're just saying it most likely came from the gun found.