Datasets
Four files worth paying for.
Raw Bitcoin data is free. Amazon publishes the entire chain as analytics-ready files; so do others. We don't sell it, because it isn't worth anything.
What takes a month is the work sitting on top of it — knowing which blocks belong to which era, which measurements are traps, and which coins are genuinely untouched rather than merely appearing so. These four carry that work. Anything we could not defend on that basis, we cut from this page rather than pad it out.
Quantum Exposure — weekly census
How much Bitcoin sits in coins whose public key is already exposed on chain, measured every week and accumulating from now on. The live 2026 risk question, as a series instead of an argument.
Why it can't be copied: nobody can backfill a time series they didn't start. The value is the derivative over months, and that clock has already started here.
Honest limit: the series begins in 2026. It is not a history.
The Sleeping Fleet
Every Satoshi-era mining reward — 196,479 outputs — classified coin by coin as asleep or awoken, with 157,766 departures dated to the block.
Why it can't be copied: it requires classifying at the outpoint level. Script-level analysis reports Satoshi's block-1 coins as active, and is confidently, silently wrong.
Honest limit: P2PK coinbase outputs only. 70 reused keys are recorded as deferred, never guessed.
The Scaling War
Bitcoin's entire governance history extracted from one four-byte field: every version-bit signal from Bitcoin XT and Classic through CSV, SegWit, Taproot, and into the version-rolling era.
Why it can't be copied: era bounds. Bit 28 means "I vote for Bitcoin Classic" in 2016 and "this is mining entropy" from 2018. Our own first query got this wrong by a factor of thirty — and the wrong answer looked perfectly reasonable.
Honest limit: signalling records intent, not hashrate identity.
Nonce Archaeology
The per-block winning-nonce series, plus the pool-conditioned anomaly panels: the early collapse toward zero when miners scanned upward from the bottom of the range, and the modern hardware fingerprints that appear and fade.
Why it can't be copied: the statistics are conditioned per pool and per window, and the obvious way to bin the data rounds — which manufactured a false result here once already, since retracted.
Honest limit: pool attribution comes from coinbase tags, which proxy pools blur.
Included free with every dataset
What ships in the box
- The data itself, as Parquet and CSV.
- A manifest listing row counts, per-column hashes, and the exact query that produced every derived table.
- A root hash and its OpenTimestamps proof, committing the file to the Bitcoin blockchain.
verify.py— recomputes all of the above from the file alone, so you never have to take our word for anything. How that works →- The full schema and a sample, published before you pay, so you can judge the thing rather than the description of it.
Deliberately absent
What we will not sell you
- Exchange labels and flow metrics. Labelling addresses as belonging to an exchange means clustering, and the published accuracy of those techniques against legally-verified ground truth is roughly 0.36 precision and 0.44 recall. We will not dress a coin flip as a fact.
- Trading signals or price predictions. Not as a product, not as a chart, not as a hint.
- Raw block and transaction dumps. Free elsewhere. Charging for them would be charging for bandwidth.
- An uptime guarantee. This runs on self-hosted hardware on residential power, which has gone dark before and will again. That's exactly why the product is files rather than a live service — a file doesn't care whether we're awake.
The shop is not open. Prices are not set, and there is no way to buy a dataset today. These pages exist so the work can be judged on its merits before any of that. There is no form here, no newsletter, and nothing that collects your details.