On-Chain Data
Information recorded directly in a blockchain's blocks, readable by anyone who runs or queries a node.
This covers transactions, balances, contract state and emitted events, fees, block times, and validator or miner activity. Its strengths are unusual in finance: it is complete for that chain, verifiable independently, and available without anyone's permission. Its limits are equally real. Addresses are pseudonymous strings, not names, so any statement about users, exchanges, or institutions depends on labels produced off-chain by inference. Intent is invisible, so a transfer cannot be distinguished from a sale, a custody migration, or an internal move without additional information. And raw node data must be indexed before it can answer analytical questions at any scale.
In practice
Anyone can verify the balance of a given address by querying a node, but knowing that the address belongs to a particular exchange comes from labeling, not from the chain.
The common misunderstanding
On-chain data is verifiable but not self-explanatory: the labels that make it readable, such as which address belongs to which entity, are estimates that can be wrong.