How do you know whether the numbers you are getting reflect something about the real world? How do you know that your data hasn't been tampered with intentionally or unintentionally?
Data provenance is extremely difficult, and that is only half the problem. The other half of the problem is determining whether you have actually measured what you think you have measured. (Note: there may be some additional halves lurking around as well.)
To give some concrete examples. You find a picture of a CEO groping someone, how do you know whether it is real? Does it matter if you want to short the stock of the company, given that you know that other people will not verify it? What if the photographer's camera automatically pushed a stream of hashes of sensor contents and GPS locations to a source with a verifiable timestamp?
How about the equivalent for a microscope sensor to ensure that the raw data from the sensor was verifiable, and that all transformations of that data needed to be published in a form that would allow anyone to verify that the raw data could be transformed to the final data?
How about a contract that specifies that if a freezer temperature rises above X degrees C for more than 3 minutes then the freezer owner must pay Y dollars to the owners of the contents of the freezer. What does it take to build sensors that the owner of the freezer would trust enough to enter into such a contract?
tl;dr Verifiability of data from oracles is extremely difficult, and much a bigger challenge than building a blockchain that could make use of it. GIGO will be a big stumbling block.