Decoded

Eighteen years ago today, Satoshi published the Bitcoin whitepaper. It was nine pages. The first code in Bitcoin's history, from 2009, was about 11,500 lines. The software most nodes run today, Bitcoin Core, is 151,004 lines, and the consensus rules are about 7% of that. I measured all of it. Counting tests, Bitcoin Core 32.1 is about 320,000 lines of code, and roughly half of that is tests. The consensus code, the part every node has to agree on, is 10,122 lines. Another 13.5% of the software handles what happens after a block or transaction is accepted, like storing it and passing it along. The other 80% is the wallet, networking, RPC commands, and the graphical interface. Core also bundles the secp256k1 cryptography library. At 76,503 lines, it is more than seven times the size of the consensus code. Libraries pulled in at build time, like Qt, Boost, SQLite, and ZeroMQ, are not counted at all. Even validation.cpp, the file most people would call the consensus rules, is mostly something else. It is 4,247 lines, and only 562 of them are validity checks. The rest stores the chain and decides what enters the mempool. The consensus code is the best tested part. For every line of it there are about one and a half lines of tests. The rest of Core is closer to one for one. The consensus code is also the oldest part. Its typical line was last changed in 2017. The typical line in the code around it was last changed in 2023. Pieter Wuille was the last person to touch 44% of the consensus code still in place. A small group writes most of the code. Over the last three years, three people made half of the changes to the wallet, script, mempool, and interface code. Ava Chow made 31% of wallet changes and 29% of script changes. Pieter Wuille made 36% of mempool changes. Seven people made half of the validation changes, and seven made half of the peer-to-peer changes. The software also keeps growing. RPC commands went from 72 to 171. The peer-to-peer protocol, the messages nodes use to talk to each other, went from 21 message types to 36. The full breakdown is in the article. secsov.com/articles/the-rule…

At @LuganoPlanB showing people @DecodeBitcoin. It's an MCP server over roughly 900,000 passages of Bitcoin governance research, Core history, and the specification, so your own model can answer hard questions about consensus rules, past incidents, and who argued what, and it returns the passages it drew from. Access is paid by the day over Lightning. Find me here and I'll connect it to your model and let you ask it something difficult. #LuganoPlanB

AI is changing how Bitcoin software gets built. Bitcoin Commons (@BTCCommons) and Plebly (@JoinPlebly) are leaning into that rather than pretending it is not happening. The risk with AI-generated code is not obvious errors. It is subtle behavioral changes that pass code review but introduce vulnerabilities or alter consensus rules in ways no one catches until it is too late. Standard review is not equipped for that. So we built BTCDecoded Intelligence, a research and review tool trained specifically on Bitcoin's protocol history, governance record, and codebase. It gives us automated review tuned to the actual failure modes of Bitcoin software rather than software in general. btcdecoded.org

I built this for a specific set of people. If you work in or around Bitcoin, this tool makes a world of difference. Protocol reviewers who need to know if an argument has been made before, by whom, and what objection it received. Researchers and journalists who need the primary source, not a synthesis. Grant programs and PR review agents scoring a submission against a spec. Companies shipping Bitcoin software that cannot afford to invent a consensus rule. Anyone writing BIPs, security notes, or historical accounts. Don't accept second best.

I've spent a year documenting Bitcoin's governance record. Every argument, every stalled proposal, every funding tie. The consistent criticism I get is that it's too much to wade through. So I built a tool that makes the entire record searchable. Now you don't have to take my word for it. Go look yourself.