Ideally, oracles would have an incentive to find decentralized sources for the feeds they provide. They should be paid to have unique ways of gathering information. But it's difficult to identify the difference between clever sourcing and faked feeds. It seems like there should be watchdog oracles that can do a meta-analysis of the feeds, to validate them. Something like a sigma statistical significance.
In specific fields such as particle physics and manufacturing, statistical significance is often expressed in multiples of the standard deviation or sigma (σ) of a normal distribution, with significance thresholds set at a much stricter level (e.g. 5σ).
At some point, real-world information feeds will break down. Imagine the scenario of SMT Oracles providing real-world information feeds for specific solutions. For example, several oracles start to provide feeds for satellite information they gather. The information they provide to the blockchain includes things like the temperature, geological particulars, and elevation of a specific area.
It's a decentralized blockchain with decentralized feeds of centralized data.
If the raw data provider ever interrupts their service, all of the feeds are interrupted right along with it because none of the oracles operate the satellites. Instead, they subscribe to the information feed and pay a monthly fee, or something. Several of them do this, so it's decentralized, in a way. But the reality is, the information source is centralized. It's repeated to the blockchain and appears decentralized because the oracles all run their feed updates at different times.
That's a single point of failure that needs to be identified before business rules are designed on top of them. There should be an incentive to manage proper feeds and a disincentive to provide garbage. But a single garbage feed might not be enough of a worry for an oracle to think about.
I think the net result would be slow decay of confidence in all feeds unless some compensating control is considered.