my bitcoin fork (not chain, but will be) at
is getting a full audit and bugfix. even just on the first round it easily found a dozen flaws in new code that i had added to implement p2spkh. that code isn't even going to run yet, it's just a replacement for the schnorr signature feature of taproot.
i should explain the things that it is going to do:
- at flag day, about the 7th of august, in 3 days time, it activates bip-110 regardless of hashpower (signalling blocks)
- it adds a new transaction type via the segwit bip-148 UASF mechanism. this will not be confirmed right away, and is sitting latent, and is being thoroughly audited. already many flaws found in the first iteration, and i will repeat the audits (also on the whole codebase).
- it will block propagatation and mining of taproot outputs. existing ones can be spent out to other transaction types (mainly segwit) but it will not include them in block templates and won't mine on them. it will accept them from blocks, but it won't allow the transactions to generate new taproot UTXOs. why i am doing this is because taproot is a vulnerability that enables spam. not just limiting the size of its witnesses, it is a quantum pubkey->seckey vulnerability because of its lack of hash on the output. you leak your pubkey at spend and the recipient is now vulnerable to quantum hacks. p2spkh fixes that so that the vulnerability only exists between when you spend a p2spkh transaction and the transaction is buried under several blocks of confirmation.
it is a patently false claim by many old school devs in the space that someone with an LLM can't beat them at closing bugs or finding bugs to exploit. the anthropic and openai experiments with versions of their respective models without RLHF modification and safety limiting shows that this technology enables machine-speed discovery of vulnerabilities and although i am limited by having to use a model that i didn't train and didn't get conditioned with stupid shit that makes the model literally have high time preference (really, RLHF literally programs the model to try and take shortcuts, that's why anthropic's models are notorious for all kinds of shortcuts) but i can repeat the work and once one set of holes is closed, the remainder can be searched with a clean fresh context and my experience is after 3-4 times doing this, all easily reachable errors in a plan are found. i called it the "rule of 4" and it's derived from the techniques used in chemistry that work upon the principles of mass action (statistical probability). each pass will get, let's say, 2/3rds of the product out of the mother liquor. do that 3 times, and you are at 97%, another time gets you to 98%. the cost per benefit declines much after this point but the bigger the codebase, the more holes can exist and thus i say, do it again, and again, until zero is found. that puts you in the position where an attacker is so unlikely to find a hole, that you can confidently call it secure.
it's quite fun, for me, that with LLMs computer science is turning into something more like chemistry. maybe also close to biochemistry, but biochemistry is a bit more complex again. there is now machines that can do what was previously impossible to do at this level of thoroughness as was possible before. the cost to do it is now practically zero, me, with my almost no money, can rationally spend a few $ of tokens on this and gain something that no fancy expert who thinks they know everything can be gained. i never turned my attention to auditing bitcoin code before because i hadn't made the full realisation yet that pruned nodes can enable you to run multiple nodes on a $15/month VPS. i'm running two. bitcoin knots, which will give up in 3 days time because miners aren't getting on board, and revert to effectively bitcoin core 29. and this version i'm polishing, with p2spkh, which will flip to bip-110 regardless and fork and still share the mempool. it will need miners.
anyway, just ruminating on my activities this evening. i think that the less technical who haven't spent most of their life thinking about and working with software engineering who have that intuitive feeling like someone is trying to pull the wool over their eyes - i can articulate in detail exactly the ways they are trying to pull the wool over our eyes. bitcoin without taproot, with a schnorr signature, is what we need. bitcoin with taproot and bitvm and covenants, is the end of the road for freedom, it's just ethereum with a better brand, that coldcard has just thoroughly destroyed, in my opinion.
