How a Bitcoin transaction works
There are no account balances. There are unspent outputs, and spending means consuming them whole and creating new ones — which explains the change, the fees and the privacy leaks.
Most people picture a balance being decremented and another incremented. Bitcoin does not work that way, and the difference explains several things that otherwise look like quirks.
Unspent outputs
The ledger is a set of unspent transaction outputs — UTXOs — each a specific amount locked to a specific address. Your wallet balance is the sum of the UTXOs your keys can unlock. There is no field anywhere recording it.
You hold one UTXO of 1.0 BTC and want to send 0.3 BTC. You cannot split it in place; you must consume the whole 1.0 as an input and create two outputs — 0.3 to the recipient and roughly 0.7 back to an address you control. That second output is the change. It is exactly like paying for a £3 item with a £10 note: the note is surrendered whole and change comes back.
Why fees depend on size, not amount
Block space is the scarce resource, so miners select transactions by fee per byte. A transaction consuming many small inputs is physically larger than one consuming a single large input, and therefore costs more to send — even if it moves less money. Moving $10m in one input can cost less than moving $100 assembled from forty small ones.
This is why wallets sometimes offer to consolidate inputs during quiet periods: combining many small UTXOs into one while fees are low reduces what future spends will cost.
The fee market
There is no fixed price. Senders attach what they choose, transactions wait in the mempool, and miners take the most profitable. When demand for block space is high, fees rise; when it is low, they fall to almost nothing. A transaction with too low a fee may wait indefinitely rather than fail, which surprises people expecting an error.
The privacy consequence
Combining several inputs in one transaction is strong evidence that the same party controls all of them, because signing requires every corresponding key. This single heuristic — common input ownership — is the backbone of blockchain analysis, and it means addresses are far less separable than new users assume. See on-chain analysis.
Scripts
Each output is locked by a small script specifying what unlocks it. Usually that is a single signature. It can also require several signatures out of a set, which is how multi-signature custody works, or impose a time lock preventing a spend before a given block height.
Bitcoin's scripting language is deliberately limited — no loops, no arbitrary computation. That restriction is a security decision, and it is the main line dividing Bitcoin from what Ethereum added.
Practical points
- An unconfirmed transaction is not settled. See consensus and finality.
- Most wallets support replacing a stuck transaction with a higher-fee version, if enabled when it was sent.
- Sending to an address on the wrong network is generally unrecoverable. Checksums catch typos, not wrong-chain mistakes.
- Every send creates a new change address by default. This is a privacy feature, and it is why your wallet shows addresses you do not remember creating.