the.great_game / monitor console

Re-analyze rescoring runs the current rubric over the most recent 75 stored articles (3 batched Gemini calls) — including ones scored under an older rubric or never analyzed — and sends alerts only for newly-worthy stories that were never sent. Stored articles auto-delete after 30 days.

Calls Claude once, on demand only. Uses the.great_game style: dark luxury esports, Hinglish, lowercase titles/hashtags, "Yogah Karmasu Kaushalam" sign-off. Tip: the → script button on any article prefills this page.

Every generated script is saved automatically — click one to load it into the viewer above.

Once a day (configurable in settings), combines the day's worth-sharing stories — first seen before the cutoff time — into ONE 5-6 minute conversational video script, like a conversation between you and the audience. Language/tone above default to your settings but only apply to a manual run — zero AI cost on a day with nothing worth covering.

Keys are stored server-side (Firestore, no public access) and never shown again in full — only whether they're set and the last 4 characters. Leave a field blank to keep its current value. Changes apply from the next run, no redeploy.

The primary model scores every batch; the fallback takes over automatically if the primary fails (missing key, outage, quota, retired model). Each provider also auto-recovers from retired model ids via -latest aliases. Applies from the next run after save — no redeploy.

Used by the script generator page. Same fallback behavior as analysis models — the primary writes every script; the fallback takes over on failure. Separate from the analysis models above, so you can e.g. use Claude for scoring and Gemini for writing.

Stories first seen before the cutoff time and scoring at or above min score are combined into ONE 5-6 minute video script at run time daily (checked every 15 min, fires once per day). Uses the script generation model above. Both times are 24h IST, e.g. "16:00".

Reads Google's "Trending Now" search feed for the region above and flags any trending search matching a keyword — often catches a launch or big moment before any RSS feed publishes. Unofficial Google endpoint, not a documented API — it can change or go down without notice; a failure here never blocks the hourly run. Zero extra AI calls on quiet days: matches are merged into the same single batched analysis call already used for RSS, and a still-trending term is only ever analyzed once per day.