You're hitting on some really hard questions :)

the gist is that you use a "record system" that has some transport guarantees, for example relying on a dht has certain kind of properties, as opposed to relying on pub/sub over trusted nodes. IPFS has this part as pluggable, though we're focusing on a large public DHT first. DHTs are pretty robust today, though yes have weaknesses we're working on.

now, the key is that, on top of that, you build assurances around cryptographic freshness (i.e. "trust this record for a certain amount of time"). of course, "certain amount of time" varies with your notion of time (e.g. NTP, vs blockchain times, etc), so the user gets to set that.

if you're interested in how the "record system" works and will evolve, take a look at https://github.com/ipfs/specs/tree/master/records -- though admittedly this is not complete or exhaustive, as we have A TON to do and are focusing on pushing out reliable code.