trainingsites/campus-ai-os/results-aggregator

results-aggregator — read the back socket

Kernel preamble §5 applies: references/kernel-preamble.md.

Internal. Scans the handoff location for every team's results.json, parses them, and groups by run_id so one topic's whole journey across teams is one row. It NEVER rewrites another team's ledger — append-only, per-team (Team Interop Spec v1 §3).

CRITICAL DIRECTIVES

• Tool-agnostic; zero hardcoded paths. Read-only over the ledgers.
• Tolerate pre-spec ledgers: missing run_id → mint from slug; missing schema_version → assume "1.0". Never fail on an older shape.
• The envelope is ONE file: schemas/results.schema.json (v1.1 = assets[] + run-level completion + optional bindings[]). Classify, don't guess: python3 tools/validate_results.py --campus-root . --json returns each ledger as v1.1-valid / v1.1-invalid / legacy-1.0 / ledger / foreign / malformed. Aggregate v1.1-* and legacy-1.0; skip ledger/foreign (job logs, not envelopes); report malformed and v1.1-invalid counts in one line, never silently.

Steps

1. Resolve the handoff location: Connected → outputs/{YYYY-MM}/runs/**/results.json; standalone/peer → ./teams-shared/runs/**/results.json. Read both if both exist.
2. Parse each ledger: run_id, team, slug, assets[] (type, status, tier, url, optional performance), run-level completion (three booleans; absent on asset-only 1.0 ledgers — derive nothing, report "no completion"), and bindings[] when present (who ran each stage — carry employee/employee_type into the roll-up so outcomes attach to the employee that produced them).
3. Group by run_id. For each topic: list teams_touched, count assets_total / assets_published, and roll up any reported outcomes (opens, clicks, attendance, replies, completion, sales).
4. Pull live performance where a connector is present and capability-gated (sales_stats, list_campaigns, course_analytics…); otherwise leave outcomes at user-reported or zero and flag no-data.
5. Cross-reference brief score if the matching brief.json is on disk (adds brief_score).
6. Hand the grouped structure to outcome-reporter to verdict and write the roll-up.