Pump.fun bot mechanics
What a launchpad bot is at the program level: one subscription, one decode, one instruction and one signature, and the four places that chain breaks.
IX-01 Open the noteProgram Notes is written for people who would rather read a transaction than a feature list. Every claim about what a Pump.fun bot does resolves, eventually, into a program invocation: a named instruction, a fixed set of accounts, a few bytes of arguments, and a state change the runtime either accepts or rejects. That layer is public, and it is where the disagreements get settled.
Nothing here reports profit, win rate or speed. Those are claims about outcomes; this desk deals with mechanism, which is the part you can check yourself by opening a signature and reading what the program was actually asked to do.
Four working assumptions sit under every note. They are the reason the pages look like reference material rather than commentary, and the reason none of them ends in a recommendation to buy anything.
A tool description is a marketing artefact. An instruction is a byte string with a name, an account list and a program that will reject it if the accounts are wrong. When the two disagree, the note follows the instruction.
Which accounts are writable, which sign, and which are derived from seeds tells you more about a program than its documentation does. A slot that must be writable is a slot the program intends to change.
Launchpad programs are deployed behind upgrade authorities and their layouts move. Every note here describes a version, states that it describes a version, and expects to be corrected when a deployment changes the shape.
The shortest path through the material: what the whole loop looks like, what one buy contains, how different classes of automation give themselves away, and what the chain records at the moment a curve finishes.
What a launchpad bot is at the program level: one subscription, one decode, one instruction and one signature, and the four places that chain breaks.
IX-01 Open the noteA single buy taken apart: discriminator, arguments, the compute budget preamble, the token account that may have to be created first, and the balance deltas it leaves behind.
IX-02 Open the noteSnipers, bundlers, volume engines and copy traders leave different marks. What separates them is fan-out, funding shape, timing and the instructions they bother to send.
BH-02 Open the noteThe moment a curve completes: the flag that flips, the reserves that move, the pool that gets initialised, and how to verify the handover without trusting a dashboard.
BH-03 Open the noteInstructions is about what a program is asked to do. Behaviour is about what the chain shows once it has done it. Terms exists because half the arguments about bots are arguments about what a word means.
A confirmed transaction is a structured receipt. The outer instructions say what the sender asked for, the inner instructions say what the programs did on the way, the logs say what the program wanted you to know, and the balance deltas say what actually moved. Most confident claims about bot activity collapse when you check which of those four the claim came from.
Snipers, bundlers, volume engines and copy traders leave different marks. What separates them is fan-out, funding shape, timing and the instructions they bother to send.
BH-02 Open the noteThe events that reveal a launch: mint initialisation, curve account creation, the program event log, and how much of the state exists at the moment each one fires.
BH-01 Open the noteThe moment a curve completes: the flag that flips, the reserves that move, the pool that gets initialised, and how to verify the handover without trusting a dashboard.
BH-03 Open the noteSix notes and a glossary. Read in order they build up from a single instruction to a full migration; read alone each one still answers its own question.
Writing your own listener, decoder, account builder and retry loop is worth doing when the mechanism is the thing you want to own. Plenty of operators do not want to own it. They want flow through a market on a schedule, with wallet funding, routing and reporting already solved, and they are willing to be a user rather than an author.
That is a different product with different questions attached, and the useful ones are about evidence rather than speed. Does it hand back signatures and slots? Does it separate what it attempted from what confirmed? A console that answers those can be checked against the same trace data described on this site; one that answers with a single headline figure cannot be checked at all.
Narrow scope, stated version, no numbers that cannot be sourced. The longer version, including what happens when a note turns out to be wrong, is on the page about this desk.
A note begins from something readable in public documentation or in a confirmed transaction, and only then draws a conclusion. If a claim cannot be traced to an account, an instruction or a log line, it does not get published as a fact.
Figures appear here as protocol constants, as arithmetic with inputs you supply, or not at all. No page publishes a fee schedule, a graduation threshold or an account count as a permanent value, because a program upgrade can move any of them.
Nothing here is a runnable bot. The notes describe what an instruction contains and what a trace shows, which is what you need to audit somebody else's activity, and deliberately stop short of assembling anything you could point at a market.