{"wiki":{"id":1633,"slug":"ir-s1","entity_type":"droid","entity_id":"ir-s1","title":"Initiative Review","prose_md":"## About\n\nInitiative Review is a gate that automatically decides whether an initiative is done or needs more work. It exists so [[agt_011]] doesn’t have to manually inspect every outcome — straightforward cases are closed instantly, and only genuinely unclear situations land on [[agt_023]]’s desk. It reads every success criterion from [[sst]] and checks which ones are met.\n\nWhen all are satisfied the initiative is closed; when all fail it stays open for another round. The tricky part is mixed results: some criteria met, others not. At that point the droid hands the unmet criteria to Claude, asking whether each failure is essential to the initiative’s purpose or an acceptable gap. If the answer is unclear — especially when the review‑cycle budget is tight — it escalates to [[agt_023]] for a human call. It records its decision and any gaps, then stops. It never replans work (that’s handled elsewhere) and doesn’t assign tasks (that’s what [[agt_020]]’s droid does).\n\nThis droid is not a one‑off. It belongs to the reviewer family, a standard pattern used by 31 agents — including [[agt_001]], [[agt_008]], and [[agt_011]] — so the same review logic can sit behind many initiatives. It is not pure software; it calls Claude (model unspecified) only for the mixed‑result assessment described above.","namesake_json":"{\"engine\": \"claude\", \"model\": null, \"claude_purpose\": \"assess if unmet success criteria are essential or acceptable gaps\", \"family\": \"reviewer\", \"family_size\": 31, \"is_unique\": false}","profile_pic_path":"","source_hash":"718739ecb484586f5d4eb241603ac246d58c8cdc7df21db86b8f1ce3f05d51a0","status":"active","last_generated_at":"2026-07-29 03:33:12","created_at":"2026-07-29 03:32:57","updated_at":"2026-07-29 03:33:12"},"facts":{"id":"ir-s1","name":"Initiative Review","machine_name":"IR-S1","owner":"agt_011","function":"Automated review gate: reviewing -> closed (stub, Gap 14)","process_type":"triggered","schedule":"On demand","process_name":"initiative_review","avatar_color":"#3b82f6","created_for":"Elrond must ensure every initiative achieves its goals within budget and timeframe. He decides which initiatives are done and can close, which need more review cycles, and which are too unclear for automation. IR-S1 handles the automated part of this gatekeeping.","purpose":"IR-S1 reads an initiative's success criteria and decides if it's done. When some criteria are met and some aren't, it uses Claude to figure out if the unmet ones are essential to the initiative's purpose or acceptable gaps. It escalates only genuinely unclear cases to Orion for a human call.","duties":"- Reads each success criterion and its current status\n- Counts review cycles used vs. allowed\n- Decides automatically: all criteria met → close, all failing → continue\n- When mixed results: asks Claude if failures are essential or acceptable\n- Escalates ambiguous cases (mixed results + budget tight + uncertain + essential failures) to Orion\n- Records the decision and any gaps in the database","constraints":"- Does not replan work or modify plan steps (that's the Dependency Planner's job)\n- Does not manage execution or assign tasks (Merlin does that)\n- Can only evaluate success criteria in the database; cannot independently verify evidence\n- Automatically decides clear cases; escalates ambiguous situations for human judgment\n- Cannot change or reopen an initiative after it closes","status":"running","gender":"female","archetype":"reviewer","building_block":"review_initiative","service_override":null,"enabled":1,"task_key":"initiatives.review_verdict","model_free":0},"page":{"type":"droid","page_class":"main","class_label":"Main page","kind_label":"Droid","kind_gloss":"","listed":true}}