Imported from srikanthbelwadi/atlas (skills/finance/conduct-outcome-monitor/SKILL.md). Install upstream with npx skills add srikanthbelwadi/atlas --skill conduct-outcome-monitor. Copyright stays with the author.

Conduct outcome monitor

When to use

"Did <tagged cohort> complaints about <product> get resolved with relief less often than untagged ones in <year>?", or a periodic outcomes check across products. Cohorts available in the public mirror: Older American, Servicemember (the closest public proxy for a vulnerability flag).

Procedure

1. For each product in scope (or all products), POST /ask: "Did <cohort>-tagged complaints about <product> get resolved with relief less often than untagged ones in <year>?" → expect ac.cfpb_outcome_gap_by_tag. Record relief_pct, explanation_only_pct, other_pct, relief_gap_pts per cohort row.
2. Optionally add timeliness: "<company>'s timely-response rate on <product> complaints by quarter since <date_from>, versus the top-10 banks by deposits." → ac.cfpb_timely_response_rate.
3. Build the outcomes table: one row per product, columns tagged relief %, untagged relief %, gap (pts), complaints in cohort. Flag any gap below −2 pts as "review", and state the cohort size beside every flag (a gap on 40 complaints is not a finding).
4. Write the definitions block verbatim from the receipts' citation_template lines (relief = response containing "relief"; tagged = tags containing the cohort), then the receipts list.

Output

conduct-outcomes-<cohort>-<year>.md with the table, flags, definitions and receipts. Never infer intent or causation; the skill reports gaps.

Guardrails

About $0.01 per product-year. Mirror ends 2023-03-23: <year> must be 2022 or earlier; say so if asked for later.