[jackcosgrove]

You would either need to be able to measure discipline-wide or sub-disclipline-wide performance, or measure unpublished collaborations between researchers.

For the first case, if molecular biologists make a lot of progress, then the molecular biology budget gets increased. Likewise if a discipline stagnates, its budget would be cut. This is probably how it already works to an extent. You still have the problem of a finite budget being shared between disciplines.

I think it's better to measure, and reward, collaboration directly rather than turn the gross knob of aggregate funding. First steps towards this would include publishing negative results, just so there's a data point. Then we reward such results insofar as they contribute to positive results (measured via citations maybe). Obviously positive results would be rewarded much more than negative results, but we shouldn't discount the efforts of the many researchers who diligently find dead ends and tell others to avoid them.