Lemmings With GPUs
Why Miners Trust Closed-Source Miners Without Looking

Chapter I: A coin with no pool, no stratum, and no fuss

Proxima is a genuinely odd and rather lovely little coin. It has no blocks and no mempool. Transactions form a DAG, token holders reach consensus cooperatively, and the plumbing is built from sequencers and delegations: holders delegate their stake to sequencers that keep the ledger moving. If you have spent the last decade with Bitcoin forks, as we have, it feels a bit like walking into a pub and finding the bar on the ceiling. Different, but once you get the hang of it, perfectly sensible.

When I stumbled across it in October there was no pool, no stratum protocol, and no third-party miner. Solo mining against your own node existed, it worked well, and it still does. In fact Proxima supports solo mining natively and recommends it, so pools are a convenience here, not a necessity.

So I did what we have done since 2013: I wrote a miner. ProxiChorizo (yes, the name is a sausage joke; no, we are not sorry) went up on GitHub with its complete source: kernels, field arithmetic, autotuning, pool code, the lot. It costs nothing. It has no dev fee. It mines solo against your own node as happily as it mines on a pool. Then we opened a test pool at proxi.suprnova.cc with a 1% fee, and published the stratum protocol next to it so anyone could build a compatible miner or pool.

The pool ran well. People mined. The kettle was on. All was right with the world, for about a week.

Chapter II: Enter the competition, stage left, carrying a megaphone

Shortly after our pool opened, another pool appeared. Let’s call it Pool Fanfare, because its main feature was the announcement. It came with its own miner: closed source, apparently AI-generated, speaking a different stratum protocol of its own, and advertised as “much faster than all other miners”.

A small detail: at that point there was only one other miner. Ours. So “faster than all other miners” meant “faster than ProxiChorizo”, a claim one could test in about an hour with a GPU and a cup of tea. Few people did.

Those who poked at the binary found something rather familiar inside. The kernels were ours, lightly rewritten, as if someone had copied a classmate’s homework and changed a few variable names so the teacher wouldn’t notice. The give-away was charming in its laziness: for a while, the error messages matched ours word for word. Including the ones we wrote at two in the morning and are not especially proud of.

Meanwhile, we kept doing the boring work. ProxiChorizo got several updates and optimisations. We pointed, repeatedly and with a smile, to the open stratum protocol on our pool and suggested everyone adopt it: one shared, open base miner that every pool could use, instead of three miners all doing the same maths. That is how healthy mining ecosystems work. You compete on service, uptime and fees, not on who can hide the kernel best.

No one took us up on it.

Chapter III: The great migration

Here is where a reasonable person would expect the story to end well. On one side:

Add it up: roughly 8% against 1%, for the same hashes from the same kernels on the same cards. And yet, to our genuine bewilderment, miners started leaving the open pool and marching, in a tidy single file, towards the one that charged them eight times as much.

We have a word for this in the trade. Several, actually, but the polite one is lemmings. (We know: real lemmings don’t actually jump off cliffs; that myth came from a nature film in which the cliff was, shall we say, assisted. Which, now we think about it, makes the comparison even better.)

Chapter IV: And then it got worse

Shortly afterwards a second pool popped up. We’ll call it Pool Mystery Meat, because nobody knows what’s in it. Its miner was, again, closed source, but this time it was also heavily encrypted and obfuscated against reverse engineering. Software that goes to great lengths to stop you seeing what it does is, as a general rule, not doing so because it’s shy.

When I dug into it, I found what looks like an “auto-update” feature. I’ll be careful here, because the binary is built so that nobody can be entirely certain. But as far as I can tell, it downloads a binary from a web server, executes it and installs it on the local machine, on Windows and Linux alike. That might be a perfectly innocent updater. It might also be a way to put any file onto your PC, at any time, chosen by whoever controls that web server. From the outside you cannot tell the difference, and that is exactly the problem.

And the performance? Despite the claims, basically the same as ProxiChorizo. Which is not surprising when you are running more or less the same kernels.

Chapter V: The great relaunch race that wasn’t

On 10 November 2026 Proxima switched its proof of work to BLAKE2b. A hard switch: old miners stop, new miners start, everybody scrambles. We updated the pool first, and ProxiChorizo 2.0 was ready many hours before the relaunch, so our miners simply carried on.

A few hours later, one of our closed-source neighbours announced that it had been “the first pool to reopen after the restart”. This is a charming claim, and also one that can be checked by anybody with a web browser: the block explorer shows Suprnova’s transactions on the new chain well before theirs. The blockchain is, after all, a public timestamped record. That is rather the point of it.

Did anyone check? You can guess. A fair number of miners read the announcement, nodded, and moved over. Again.

Chapter VI: So, why? A short field guide to the herd

We have run pools since 2013 and we like miners, honestly we do. But every so often the community does something that makes us stare into our tea. So let’s ask the question as kindly as we can: why would someone who spent thousands on GPUs, and who keeps wallets on the same machine, install an obfuscated binary from a stranger, pay eight times the fee, and never once compare it with the free miner sitting right next to it?

Theory 1: Marketing works. Depressingly well.

“Much faster!” “First to reopen!” “Next-gen AI-optimised kernels!” A confident sentence in a Discord announcement beats a GitHub repository every time, because reading a sentence takes three seconds and reading a repository takes effort. We write code; they write adverts. It turns out the adverts get more miners. Lesson noted, with a heavy sigh.

Theory 2: Benchmarks are believed, not measured

A screenshot of a big number is not a benchmark. A benchmark is both miners, on your own card, for at least an hour each, with the hashrate checked at the pool, not on the miner’s own screen. A miner can print any number it likes; only the pool sees what actually arrives. Yet almost nobody does this. One would think that people who have spent the price of a small car on graphics cards might want to know which software actually makes those cards earn more. One would be wrong, apparently.

Theory 3: “Free” has become suspicious, and “closed” looks professional

There seems to be a creeping belief that something free must be worse, and something hidden must be clever. It is precisely the other way round. Open code is the version that can be checked. Closed code with a fee is the version where you are paying for the privilege of not knowing what you’re running. Thirty years of free software built the internet you’re reading this on. It deserves rather more credit than a binary with a padlock on it.

Theory 4: Nobody wants to be last

When half a Discord channel moves, the other half follows, because surely all those people must know something. They usually don’t. They are simply following the person in front of them, who is following the person in front of them. That is not research. That is a queue.

Theory 5: Well…

We were asked to be polite, and we shall be. Let us simply say that if you would not let a stranger into your house to “update your boiler” without checking who he is, you should probably not let a stranger’s obfuscated binary download whatever it fancies onto the PC that holds your seed phrase. That is not paranoia. That is the bare minimum of common sense, and it is a little worrying how many people with expensive hardware appear to have left theirs in the other trousers.

Chapter VII: Why this matters beyond one coin

You might think this is a pool operator being grumpy about losing hashrate. Partly, yes, we’re human. But the bigger problem is what it teaches the next coin’s developers and pool operators:

• Copying open work and closing it pays. If the person who wraps someone else’s kernels in an obfuscated binary wins the crowd, the next author will think twice before publishing anything.
• Marketing beats engineering. If the loudest claim wins, the next project will spend its budget on claims, not code.
• Remote-controlled miners become normal. Once miners accept a binary that fetches and runs code from a server, it is only a matter of time until someone uses that channel for something truly nasty. We have seen it before: see our trojanised SRBMiner advisory, where a “miner” turned out to be a rootkit dropper.
• Fees creep up. Every miner who happily pays 8% tells the market that 8% is fine. Mining margins are thin; that difference is often your entire profit.

Crypto was supposed to be about verifying instead of trusting. “Don’t trust, verify” is practically printed on the box. Mining software is the one place where every miner can actually do that, and it is the place where so many simply… don’t.

Chapter VIII: A ten-minute due-diligence checklist

We don’t want to just grumble, so here is something useful. Before you point a single GPU at a new miner or pool, run through this. It takes about ten minutes, plus an hour of letting things run while you have a nice cup of something.

1. Is the source available, and can you build it yourself? Not “we’ll publish it later”. Now. If it is open, check that the release binary matches what the source builds, or simply build it yourself.
2. Is the binary obfuscated or packed? Legitimate miners have no reason to hide from reverse engineering. If it’s wrapped in three layers of encryption, ask yourself what it’s hiding.
3. Does it talk to anything other than your pool? Watch its network connections for a while (Resource Monitor on Windows, ss -tp or lsof -i on Linux). A miner needs exactly one destination: the pool, or your own node. Anything else needs a very good explanation.
4. Does it download or execute files? An “auto-updater” you cannot inspect is a remote-control channel. Turn it off, firewall it, or don’t run the miner.
5. What is the total fee? Miner fee plus pool fee. Check the miner’s own hashrate against what the pool credits over an hour or more; the gap is the real fee, whatever the README says.
6. Benchmark yourself. Same card, same clocks, an hour per miner, measured at the pool. Ignore screenshots, including ours.
7. Check claims against the chain. “First to find a block”, “first to relaunch”: the block explorer will tell you in seconds.
8. Keep wallets off the mining rig. Whatever you run, the PC that mines should not be the PC that holds your seed phrase. A cheap hardware wallet costs less than one month of a 7% fee difference.
9. Consider solo mining. Where a coin supports it natively, as Proxima does, solo mining against your own node removes the pool from the trust equation entirely.

Chapter IX: A closing word, with the kettle still on

We are not asking anyone to be loyal to Suprnova. Pool hop to your heart’s content: that is your right, and competition keeps us all honest. We are asking for something much smaller: before you hand your machine to a stranger’s software, spend ten minutes finding out what it does. Compare it with the free alternative. Measure instead of believing. Check the chain instead of the advert.

If, after all that, you still prefer the closed binary with the 8% fee and the mystery downloader, then you have made an informed decision, and we shall wish you well with a perfectly straight face. But if you never looked… well. The cliff is right there, and it was never a very good film.