17.8 s
The board
Every sampled mainnet slot, attributed to a delivery group and priced against that slot's reference. A landing shows up on the public board 17.8 s after confirmation (p95).
Solana · transaction delivery
What it really costs to land a Solana transaction — lane by lane.
Jito, Nozomi, bloXroute and plain RPC each charge for delivery in their own way. FairLane measures what each one actually took on mainnet, against the same slot, and puts the numbers side by side.
Open the live board
npm install @fairlanerouter/sdk
Live lane prices, last hour
- JITO
- NOZOMI
- BLOXROUTE
- PLAIN RPC
Reading the last hour…
Which lane to take for your transaction is a question with an intent — ask the API.
What it does
186 ms
Advice
Describe a transaction before it exists — program, compute units, cheap or fast — and get back a group, a tip, a priority fee and the full expected cost, with the age of the data behind it. 186 ms p95.
1.54 s
Keys without accounts
One request issues an access key: no sign-up, no e-mail, no personal data. 120 requests a minute with a key, 10 without.
How it works
-
Sample
Every 100th mainnet slot is read in full. Each successful non-vote transaction in it is a landing, with its base fee, priority fee and tip — a failed transaction did not land, and a vote pays nothing for delivery.
-
Attribute
A tip to a service's own accounts names its group. Services that share accounts cannot be told apart on-chain, so they form one group — the unit of every comparison.
-
Reference
Each slot gets one reference: the 10th percentile of full landing cost among its successful non-vote transactions, computed only when the slot has at least 50 of them. Vote transactions are always excluded — they would pull the reference to almost zero.
-
Overpay
Overpay is what a landing cost above its slot's reference. Comparing overpay, not tips, keeps a busy slot from looking like an expensive service.
Measured on the deployed product
| Criterion | Budget | Measured | Date |
|---|---|---|---|
| Landing visible on the board after confirmation (p95) | ≤ 60 s | 17.8 s | 2026-09-22 |
| First screen with data on a 3G connection | < 2 s | 1.76 s | 2026-09-22 |
| Answer to an advice request (p95) | < 300 ms | 186 ms | 2026-09-28 |
| Issuing an access key, no personal data | < 10 s | 1.54 s | 2026-09-24 |
Measured values are recorded in the release notes of each tag: v0.1.0, v0.2.0.
What FairLane does not do — yet, or ever
- It has not proven savings yet. Advice nobody followed saves nothing. The claim that the recommended lane is cheaper needs FairLane's own paired sends — that is the next milestone, and until then this page does not make it.
- It does not estimate landing probability yet. Passive collection sees what landed, not when it was sent, so the field is
nullrather than a number that only looks like a probability. - It does not name a brand inside a group. Services that share on-chain accounts are indistinguishable, and FairLane does not guess.
- It sees more lanes than it can send through. Some groups are observed only; when one of them is cheaper, the advice says so instead of hiding it.
- It does not send, sign or hold anything of yours. No custody, no user keys, no on-chain program.
- No MEV protection, no SLA. Only the price of delivery is measured, and advice is an estimate, not a guarantee.
- Not every landing is attributed. The unattributed share is counted and shown under the gantry, never folded into a group.
Questions
- Why groups and not services?
- Services that tip into the same on-chain accounts are one lane as far as the chain can tell. FairLane compares what it can prove, and never splits a group by guessing.
- Why the 10th percentile?
- It is what landing in that slot cost someone who did not overpay: a few freakishly cheap landings do not move it, and the overpay of the majority does not become the norm. Votes are left out — they are most of a block and pay only the base fee, so with them every lamport of delivery would count as overpay.
- Does it cost anything?
- The board, the API and the SDK are open. Advice is rate limited: 10 requests a minute without a key, 120 with one.
- How is it computed, exactly?
- The method view of the board describes how the reference and overpay are computed. Its worked examples are labelled synthetic.
- Is my transaction sent anywhere?
- No. The SDK gives advice only: it never builds, signs or sends a transaction, and depends on no Solana library.