Priority fees in practice
How the priority fee is actually computed, why the requested unit limit and not the consumed one sets the bill, and how to raise a fee without wasting it.
Read the entryLatency: How Long It Took and What It Cost
Once a transaction lands reliably, the remaining questions are how long it took and what it cost to get there. Those two are the same question asked in different units, because nearly every setting that buys speed is paid for in lamports and nearly every saving is paid for in delay.
The entries here treat fees, compute budgets, measurement and congestion as one subject, because tuning any of them without the other three produces a number that moved for reasons nobody recorded.
- Fee formula
- Requested compute units multiplied by the unit price in micro-lamports, divided by one million
- Unit ceiling
- A single transaction may not exceed 1,400,000 compute units, whatever it requests
- Slot cadence
- Slots advance on the cluster clock; block height advances only when a block is produced
- Leader rotation
- Leaders are assigned per epoch and hold four consecutive slots at a time
4 entries in this section
The cost and timing side of the same send path: priority fees, compute budgets, measured landing rate over time, and what the network does to all three when it is busy.
Compute units and limits
Default budgets, the per-transaction ceiling, per-block and per-account write ceilings, and how to set a limit from simulation instead of from a guess.
Read the entryMeasuring landing rate
A landing rate that means something: what counts as an attempt, what counts as a landing, and the four numbers that make the rate diagnosable rather than decorative.
Read the entryBehaviour under congestion
What changes when the network is busy: write-lock contention, skipped slots, queueing at the leader, and the settings that help against the ones that do not.
Read the entryEvery recommendation states its price
A tuning suggestion without a cost attached is a preference. Each entry in this section names what a change buys and what it charges, including the changes this desk recommends, so that a reader can decline one on their own terms rather than on ours. Where a trade has no good side, the entry says that too.
Fee Priority fee is charged on the units requested, not on the units consumed
Budget An over-generous limit costs lamports and makes the transaction harder to schedule
Measurement A landing rate is only comparable when its denominator and window are published
Congestion Four constraints can bind, and two of them do not respond to a higher fee at all
Pacing Reducing concurrency is the one correction that costs no lamports and is used least
Landing Did it get in Whether a signed transaction reaches a block at all: expiry windows, rejection classes, duplicate risk on retry, and the evidence that separates a dropped send from a failed one. How this log is written Method and limits What the desk checks before an entry is published, which claims it refuses to make, and how a correction changes a page that is already live.