The cron's home page. One daily job keeps ai.fedm8.com sounding like a person, not a demo bot.
What it does, every day at 10:07 UTC: fetch live pages → hunt AI-sounding copy against the corpus → judge against the Human Framework → recommend plain rewrites → verify yesterday's fixes are still fixed → learn new tells → write every change to a ledger → republish the framework's living section.
| You want | Go to |
|---|---|
| Today's report — findings, rewrites, what it learned | https://ngjmqdzpnhwpybtssykz.supabase.co/functions/v1/voice-audit?report=1 |
| Every change ever made (drift / curating / English review) | Human Framework → Change ledger — machine-managed section at the bottom |
| Schedule health — did it run, is it overdue | Portal ops board (#/ops), row FedM8 voice audit — Human Framework |
| Raw data | voice_audit schema on the ops [Database]: runs, findings, patterns, changes |
POST https://ngjmqdzpnhwpybtssykz.supabase.co/functions/v1/voice-audit with {"actor":"<you>","force":true} — force needs a live fedm8 claim. Throttled to ~one run per 20 hours otherwise.
_Status note: the deep judgement pass is waiting on one credential — LITELLM_VIRTUAL_KEY in the Edge Function secrets. Until then runs are pattern-only and say so in the report._