Run first recurring Research Radar trial week

[t3chn]

Goal: run the first recurring Research Radar trial week before selecting v0.8.

This is an operational recurring research process, not repo automation.

Scope:

• no new task family;
• no automation scripts in the public repo;
• no public scheduled jobs;
• no scraping or collectors in the repo;
• no private eval material;
• no customer data;
• no raw feeds committed to public repo;
• no automatic GitHub issues unless explicitly approved.

Cadence:

• weekday morning Daily Research Radar brief;
• weekly synthesis checkpoint after the daily briefs;
• monthly watchlist pruning later, only if the trial is useful.

Daily brief rules:

• use v0.7 Research Radar watchlist, source map, repos, people, and query sets;
• focus on benchmark mechanics, not generic AI news;
• output brief only, not repo changes;
• recommended actions max 5;
• answer: Should Agent Bench Lab change direction today?

Weekly synthesis rules:

• review daily briefs;
• decide keep / modify / defer / stop;
• propose issues or decision-log entries;
• do not choose v0.8 from inertia.

Success criteria:

• 3 daily briefs completed;
• 1 weekly synthesis completed;
• weekly synthesis explicitly recommends keep / modify / defer / stop for each v0.8 candidate;
• no repo writes from daily briefs;
• no auto-created issues without approval;
• no private eval material or customer data.

Possible v0.8 candidates:

• SEC-01 decision-grade / security overlay;
• Private Eval Bundle manifest validator;
• Report schema runtime v1;
• lightweight Research Radar automation;
• browser/replay standard;
• MCP-local extension;
• nothing yet.

Decision rule:
Do not pick v0.8 until weekly synthesis is complete.