A small Bitcoin-like chain whose canonical ledger is this README.
_{Proof of work is subset-sum, not hashing — so mining rewards a better solver, not better hardware}

Block explorer · Consensus spec · Mine a block · Send coins · Node

Find your own name here once you have run python3 wallet.py identity.

_{Rendered from chain/blocks.jsonl at height 3985. Verify it yourself: python3 verify.py}

Everything here is real except the money. Coins move in UTXOs, spent by ECDSA signatures over secp256k1 — the same curve Bitcoin uses, with RFC 6979 deterministic nonces and low-s enforcement. Difficulty retargets every 16 blocks against how long the last window actually took, clamped to a factor of four. The subsidy starts at 50 ROFL and halves every 210 blocks. Coinbase outputs need 10 confirmations before they can be spent.

The one deliberate departure is the work function. Mining ROFL means solving subset-sum puzzles, not grinding hashes.

Given 40 numbers and a target, find the subset that adds up to it exactly. Verifying an answer is one loop of additions. Finding one is meet-in-the-middle at 2²⁰ time and memory — and memory is what binds, which is why difficulty comes from solving more puzzles rather than bigger ones. A block at the floor needs 1; the ceiling is 3,000, about as many solutions as fit in one GitHub comment. Difficulty stops there, because past that point a harder target buys no extra work.

Nobody has built an ASIC for subset-sum. The reference solver in miner.py is deliberately plain, and beating it is the entire sport. @ksanjeev284's CUDA solver is the fastest one anybody has published so far — it stays out of here because "no dependencies" is the pitch, not because it doesn't work.

The full rules, the measurements behind every constant, and each place this knowingly diverges from Bitcoin, are in SPEC.md.

No dependencies. Python 3.9 or newer, standard library only.

git clone https://github.com/ram0verflow/ram0verflow.git rofl && cd rofl
python3 wallet.py new                                  # make a key pair
python3 miner.py --miner YOUR_GITHUB_HANDLE --message "gm"

It solves puzzles until the block is complete, then prints a line starting with rofl-block-v1:. Paste that as a comment on the block issue. A workflow validates it and, if it holds up, appends it to the chain and updates this page.

About half a minute at the starting difficulty. If the chain gets busy, difficulty rises and you solve more puzzles — that is the point.

Your GitHub handle seeds your puzzles. Two things follow. Nobody can submit your solved block as their own, because a different handle means different puzzles and the work would have to be redone. And copying a solution out of the comment thread gets you nothing, because it answers a question only you were asked.

python3 wallet.py balance
python3 wallet.py send --to rofl1... --amount 1.5 --memo "gg"

That prints a rofl-tx-v1: line. The memo rides inside the signature, so it cannot be altered or stripped on the way, and it shows up in the ledger above.

Two ways to submit it, and they do the same thing:

• Comment on the mempool issue. Ten seconds.
• Pull request adding one file under chain/pending/, if you want the contribution on your profile. The node reads the file, applies the transaction to main and closes the PR — it is never merged, so your submission can't conflict with anyone else's.

Either way it waits in the mempool until a miner includes it. Higher fees get picked first, and the fee goes to whoever mines the block.

python3 wallet.py identity --handle YOUR_GITHUB_HANDLE

Post the rofl-id-v1: line it prints on the mempool issue from the account it names. The signature proves you hold the key; posting it from your account proves you hold the handle. Your name then appears beside your balance in the table above.

This is display only. It gives nobody any authority over your coins — those are spendable by signature and nothing else — and verify.py ignores the registry completely.

A README is one file served identically to everyone — no scripts, no per-visitor anything. So the personal view lives one click away, at the block explorer.

It fetches chain/blocks.jsonl and replays the entire chain in your browser: every hash, every merkle root, every difficulty retarget, every signature, every subset-sum puzzle re-checked against its target. The consensus rules in docs/app.js are a direct port of the Python, and the two agree bit for bit.

Tell it your GitHub handle once and it remembers — your balance, the blocks you mined, your transfers, your rank. That is stored in your browser and sent nowhere. Search takes a block height, a block hash, a txid, a rofl1… address or an @handle.

Nothing on that page is served by a backend. There isn't one.

python3 verify.py

Replays every block from genesis: recomputes each hash, re-derives every difficulty retarget, rebuilds every merkle root, checks every signature and every coinbase amount against the subsidy schedule, and asserts that emitted supply equals unspent supply.

It reads only chain/blocks.jsonl. It does not trust this README, the workflow, or any cached state. If I ever rewrite history, this is what catches me.

python3 tests/test_chain.py

Builds a chain in memory, spends real coins with real signatures, then tries to break it nine ways — tampered subsets, forged signatures, inflated coinbases, double spends, stolen blocks, padded solutions, wrong difficulty, deleted blocks, stale tips — and asserts every one is rejected. Runs at a shrunk puzzle size so the whole suite takes seconds.

rofl/crypto.py       secp256k1, RFC 6979 ECDSA, bech32
rofl/pow.py          subset-sum: instances, solver, verification
rofl/consensus.py    nBits, retargeting, merkle, UTXO set, validation
rofl/chain.py        load, replay, extend
rofl/render.py       this page
docs/                the block explorer — consensus ported to JavaScript
miner.py             the reference solver — beat it
wallet.py            keys, balances, signed transactions
verify.py            independent full-chain verification
submit.py            what the workflow runs
SPEC.md              the consensus rules

One writer, so no reorgs. Two miners who solve the same height race on submission time; the loser is handed the new tip and mines again.

The reference solver is pure Python and wants about 200 MB while it runs. That is the memory wall meet-in-the-middle hits, and the reason difficulty scales by puzzle count instead of puzzle size.

Proof of work makes a rewritten history detectable to anyone holding an earlier copy. It does not make one impossible. This chain lives in a single repository and is exactly as durable as that. SPEC.md §13 and §16 say the same thing in more detail.

ROFL coins are worth nothing and always will be. rofl-wallet.json holds a private key in plain text — never reuse it anywhere that matters.

MIT.