Posts
Posts
Arguments grounded in systems I have built, operated, measured, or contributed to.
Changing a specification is mostly a migration problem
The Nostr wiki format change was agreed before I touched it. What kept it from merging was that nobody had written the converter, which is a different kind of work from being right.
A duplicate can be an observation
BigBrotr stores one Nostr event once and records every relay that served it, because repeated content is redundant while repeated observation is the measurement.
An index must preserve the specification's distinctions
A Nostr storage backend returned the wrong rows because its index kept tag values but discarded tag keys, collapsing a distinction required by the protocol.
A failed crawl is not an empty result
A measurement pipeline has to preserve three different states: observed, absent, and unavailable. Collapsing them produces confident findings from data the collector never saw.