light client
A light client is software that checks the blockchain without downloading or executing the whole thing. A full node re-runs every transaction in history; a light client instead downloads only the small block headers — the chain's table of contents — and then asks well-connected full nodes for the specific pieces of data it cares about, together with a cryptographic proof that those pieces really belong to the chain. The trick is that headers are tiny (Bitcoin headers are 80 bytes each; the whole header chain is only tens of megabytes) yet they commit, via Merkle roots, to all the heavy data underneath.
Bitcoin's original light-client design is called SPV (Simplified Payment Verification), described in the very first Bitcoin paper. To confirm a payment landed, an SPV wallet does not need the full ledger; it checks that the transaction's hash sits under a block header's Merkle root (a short Merkle proof) and that the header is buried under enough proof-of-work. The wallet trusts the economics of mining: producing a fake header chain with valid proof-of-work is prohibitively expensive, so a header backed by real work is very likely honest.
Proof-of-stake light clients work differently because there is no proof-of-work to anchor on. Ethereum's beacon-chain light client instead follows a rotating 'sync committee' of 512 validators that signs each header; a light client verifies the committee's aggregated BLS signature and tracks committee handoffs every period, so it can follow the canonical head while downloading almost nothing. This is what makes trust-minimized bridges and in-browser or on-phone verification possible.
The honest limitation is that a light client gives weaker guarantees than a full node. It verifies that data was committed to and (in PoW) backed by work, but it generally cannot check that every state transition was valid or that no valid transaction was censored — it must assume at least one of its peers is honest and will hand over a proof if something is wrong. Light clients also have a privacy cost: by asking specific nodes about specific addresses, they can leak which addresses they care about unless they use techniques to obscure their queries.
A mobile wallet wants to confirm you received 0.5 BTC. It holds only the 80-byte headers. It asks a full node for the transaction plus the list of sibling hashes up the Merkle tree, hashes them together, and checks the result equals the Merkle root already inside the header it trusts — confirming inclusion without ever downloading the 600+ GB chain.
SPV proves a transaction was included and buried under work; it does not by itself prove the rest of the chain followed every consensus rule — that is the gap full nodes close.