TraceCore Guide
Reference expectations for bundled agents and tasks.
Quick Start
Use the success list below to pick a dependable agent for each task.
Deterministic
All tasks listed here are deterministic; match the agent to the expected win path.
Notes
Failures are expected outside the listed success tasks or when budgets drift.
Tutorial Mode
Use the dashboard and CLI in a simple loop: run an episode, inspect the trace, verify the latest artifact, then seal or audit bundles once behavior is stable.
Suggested sequence
- Run an agent/task/seed combination from the dashboard or CLI.
- Inspect Output JSON, Trace Dump, and IO Audit to find the earliest divergence.
- Verify the latest artifact and compare against known-good behavior.
- Seal or audit bundles only after the run is stable and reproducible.
Supporting views
- Dashboard for fast trace/result inspection
- Metrics for success-rate and budget trend checks
- Ledger for bundle and trust-pipeline evidence
- CLI verify/bundle for replayable, auditable validation
Tip: treat this page as the UI-side companion to the FAQ, debugging playbook, and contributor onboarding docs.
| Agent | Expected successes | Expect failures | Notes |
|---|---|---|---|
| {{ entry.agent }} | {% if entry.success %} {% for task in entry.success %} {{ task }} {% endfor %} {% else %} None {% endif %} | {% if entry.success %} All other tasks {% else %} All tasks {% endif %} | {{ entry.notes }} |