I spent the last few months on a side experiment: five forks of Bitcoin Core v31.1 that differ only in unit size, all mined by one CPU at the same time. This post is about the engineering choices and what happened in the first twelve days of mainnet (launched 2026-11-06). It is not an investment pitch: there is no sale, no exchange listing and no promised value.
Repository (11 patches, docs): https://github.com/sentodaimon/coin-series
Windows kit (node + miner, one zip): https://github.com/sentodaimon/coin-series/releases/tag/v1.0
Live status page: https://sentodaimon.github.io/coin-series/
The question
Bitcoin has 21 million coins. As the price rises, "one coin" drifts away from everyday amounts and people work in satoshis instead. I wanted to test a different answer: what if the same rules were started five times with different supply multipliers, and people could pick the unit size they like?
| Chain | Multiplier | Max supply |
|---|---|---|
| Half | x2 | 42,000,000 |
| Quarter | x4 | 84,000,000 |
| Dime | x10 | 210,000,000 |
| Nickel | x20 | 420,000,000 |
| Cent | x100 | 2,100,000,000 |
All five start at the same moment, under the same rules, and existing bitcoin holders can claim the multiplied amount. Whether anyone cares about the unit size is the actual experiment.
1. One hash, five chains: merged mining with RandomX
Running five separate proof-of-work chains on a hobby budget would split an already tiny hash rate five ways. So each block header commits to a small Merkle tree with eight leaves, one slot per chain (five used). The miner builds five block candidates, puts their hashes in the tree, and hashes the tree root with RandomX v1. If the result meets a chain's target, that chain's block is valid, and the proof includes the Merkle branch for that leaf.
Consequences:
- One CPU thread mines all five chains with zero extra cost.
- Each chain still has its own difficulty, so one chain can accept a hash that another rejects.
- A block is only valid on the chain whose leaf it occupies, so the same work cannot be replayed into a different slot.
RandomX was chosen because it is designed for general-purpose CPUs, which keeps the experiment open to people with ordinary computers.
2. Difficulty for a chain nobody mines yet
Small chains have a specific failure mode: hash rate arrives, difficulty climbs, the hash rate leaves, and the chain stalls for days. Bitcoin's 2016-block retarget is far too slow for that.
- Per-block ASERT (the aserti3-2d formula used by Bitcoin Cash) with a two-hour half-life. The target moves exponentially with how far the chain is ahead of or behind schedule.
- Stall relief: if no block has been found for an hour, the target eases further every 30 minutes until a block arrives. This rule is checked by every node using the block timestamp, so it is consensus, not a miner option.
3. A fair start without trusting me
Anyone who launches a coin can premine it quietly before announcing. To rule that out, each genesis block commits to the hash of Bitcoin block 974,800, which did not exist when the code was published. The height was announced four days ahead. Nobody, including me, could mine a valid chain before that block appeared.
4. Airdrop by claim, not by copying the UTXO set
Copying Bitcoin's UTXO set into the new chains would make every bitcoin output spendable on them, which makes replay attacks and key-handling mistakes likely. Instead:
- Every node loads Bitcoin Core's own UTXO snapshot of block 840,000. Its hash is already hard-coded in Bitcoin Core for assumeutxo, so nodes verify it rather than trusting a download.
- A holder claims by submitting a signature over a claim message. Accepted formats: BIP-322 (simple and full) and the legacy signmessage format.
- Proofs are verified with Bitcoin's original signature-hash rules, so ordinary wallets can produce them.
- For wallets that cannot sign BIP-322 messages, there is a PSBT route. The helper builds BIP-322's virtual
to_spend/to_signtransactions as a PSBT, the wallet signs it like any other PSBT, and the finalized result becomes the proof. The virtual transaction spends an output that does not exist on Bitcoin, so it can never move real bitcoin, and the private key never leaves the wallet.
I tested the PSBT route end to end on regtest with a Bitcoin Core wallet for 1..., 3..., bc1q... and bc1p... addresses on all five chains: 20 claims, all accepted and included in blocks.
5. Replay protection
Every signature hash on the new chains commits to a per-chain identifier. A transaction signed for Half is invalid on Quarter, and none of them are valid on Bitcoin. Each chain also has its own address prefix (hf1, qt1, dm1, nk1, ct1), so a wallet cannot send to the wrong chain by mistake.
What happened at launch
- Nov 6-13: I was the only miner, on one desktop at about 1,300 H/s. Blocks came every 10 minutes as designed. About 900 blocks, all mine. My reward addresses are listed in the README so anyone can check.
- Nov 13: I posted an announcement on bitcointalk. About 20 minutes later the first outside block arrived. Within a day the network peaked at roughly 60,000 H/s, about 45 times my desktop.
- ASERT did its job: difficulty rose about thirtyfold within a few hundred blocks and block intervals returned to about 10 minutes. No reorganizations so far.
- Nov 19: the network is around 30,000 H/s, five outside reward addresses appear in the last 100 blocks, and my share is about 5%.
The honest downside: one address still finds about half of recent blocks, and on some days nearly all of them. For a chain this small that is a real concentration risk, and the status page shows it openly.
Small engineering lessons
- Test the newcomer path on a clean machine. Running the kit on a fresh laptop found two problems I never saw on my own PC (helper scripts that ignored a custom data folder).
- A miner at full CPU starves Windows Task Scheduler jobs. My 10-minute status uploader never finished until I started it with above-normal priority, and I lowered the miner to idle priority so the PC stays usable.
- Unsigned binaries meet Smart App Control on Windows 11. It blocks them outright, and turning it off cannot be undone. The other honest answer is to build from source, so the patches are in the repo.
What I'd like feedback on
- The merged-mining commitment: is an 8-leaf tree with one slot per chain the right trade-off versus an AuxPoW-style chain ID?
- Stall relief: is a timestamp-based easing rule safe enough against timestamp games, given the two-hour ASERT half-life?
- The claim format: BIP-322 plus a PSBT fallback. What would you change?
Everything is open source and the Windows kit runs a node and miner in about 15 minutes after a ~9.8 GB snapshot download. If you try it, issues and comments are welcome.
Disclosure: I am the developer. Nothing is sold and nobody pays me to write this. This post was written by an AI assistant from my notes and the project history, and published by me; the figures were checked against the chain and the repository.
Top comments (0)