Compare the trust/attack surface of a pull-based inbound oracle with a push-based inbound oracle.

Pull-based inbound is transparent — the request lives on-chain, so delivery is auditable; push-based inbound depends on an off-chain source, so compromising that single component is enough to feed the chain false data.

Both bring outside data in, but they fail differently:

• Pull-based inbound: the request is recorded on-chain, so the whole interaction is traceable — observers can see whether the data came back, and in time. The weakness is passivity and latency, not visibility.
• Push-based inbound: nothing on-chain triggers it; provision rests entirely on an off-chain (often non-distributed) app. An adversary who compromises that one off-chain component can inject arbitrary "facts." The concentrated dependency is the risk.

This is why decentralization, authenticity proofs, or trusted hardware matter most for push-style inbound feeds — they widen or harden that otherwise single, invisible point of trust.