Four items today sat on the gap between claimed and tested
Victoria Shinder2 min readFour of today's items are about the same uncomfortable question: how much of what a system promises is actually tested, and how much is assumed.
Bitcoin's PSBT specification gained a test vector because a bug escaped, and with it an explicit instruction that the combiner vectors be run in both orders. The specification had already decided that combining is not commutative; the tests had not noticed. For as long as that gap existed, two wallets could disagree and every vector would still pass.
Ethereum researchers spent the day sorting hash functions by how hard people have tried to break them, not by speed. BLAKE2 has no published attack and fewer than ten cryptanalysis papers; the post declines to call the ZK-friendly families mature while analysis against them is still moving. Untested is not the same as sound.
A Polkadot forum post proposed replacing re-execution with proof verification - a direct trade of first-hand checking for cheap second-hand checking - and asked for collaborators. The architecture is plausible. What it has not got is the economics of who pays the provers.
And the Solana ecosystem read an SEC order as opening tokenised equities for five years. The account is written by an advocacy organisation; the order itself is where the conditions live.
The thread
Every one of these is about the distance between a claim and the thing that would catch the claim being wrong. A test vector that runs one order. A hash with a short literature. An architecture with no cost model. A summary standing in for an order. None of them is a failure yet, and each is the place a failure would come from.
The practical reading: when something says it has been verified, ask what the verification would have missed. In specifications that means asking which orders and cases the vectors exercise; in cryptography it means counting papers, not attacks; in architecture it means asking who pays; and in regulation it means reading the instrument rather than the summary of it.
⚠️ None of these four is a result. Two are proposals, one is a test-suite change, and one is a second-hand description of a legal instrument we have not opened.