Three protocol write-ups this week were about work done before anyone checks who pays or what is real
Three technical items from the past week, on Ethereum, Polkadot and the XRP Ledger, look unrelated. Read together, they describe the same weakness in protocol software: a component does work, or trusts a value, before anything has confirmed it should.
On Ethereum, a Nethermind benchmark of EIP-8141 frame transactions measures the cost of validation that nobody pays for. A node must run a transaction's VERIFY code before it knows whether the transaction is valid, and a failing transaction pays nothing. The researchers found that raising the cap from 100,000 to 300,000 gas, which privacy and post-quantum accounts need, is only safe if nodes stop validating gossiped transactions while they import a block; even then, honest transactions get crowded out of the mempool under attack.
On Polkadot, a forum post traces how dotUSD's recorded supply fell to 153,884 while balances held 4,213,885. A credit object dropped outside a rolled-back block lowered supply without moving any balance. Nobody had treated supply as a price input, yet a safety cap in the assets pallet meant the trading pool read its reserves through that number. A bug rated medium in June, before the pool existed, became a mispricing in October.
On the XRP Ledger, rippled pull request 7827 adds a null check to two RPC handlers that already had siblings with the same guard. Its author says the path is unreachable without a corrupted ledger, and the lines are excluded from test coverage. It is the cheapest kind of fix, made because consistency across similar code is easier to review than an argument that a case can never happen.
The common thread is that the risk sat between components. Validation cost lands on block processing; a supply figure lands on a price; an unchecked entry lands on a public RPC server. Each fix moves a check closer to the place where the assumption is used.

What it means
For teams building on these chains, the practical lessons are to list everything that reads a value before trusting its accuracy, to make expensive unpaid work yield to critical work, and to apply the same defensive checks across sibling code paths even when a failure looks impossible.