# Reproduce the bounded briefing exercise

These are archived official release excerpts with deliberately duplicated, stale and unavailable entries. They are not today's news. Upstream release material: MIT, Copyright (c) 2025 Nous Research.

1. Save sources.json and prompt.txt in a dedicated trial folder. Keep the accepted result.json separately for comparison.
2. Replace <trial-folder> in prompt.txt with your actual folder. Use your configured model and provider; the receipt records what this run used.
3. Run: hermes chat --safe-mode -t file --max-turns 8 --run-budget 120 --source tool --format stream-json --query-file prompt.txt --in <absolute-trial-folder>
4. Inspect the written result.json: exactly v0.22.2 and v0.22.1 once each, two source identities for v0.22.2, v0.21.7 excluded, missing-source in gaps. This historical output is not installation advice.
5. To reproduce the outage, save the accepted output as last-good.json in a separate trial folder. Create sources.json with the same reporting-window fields but only an unavailable record (sourceId missing-source, access unavailable, excerpt null), set the window to October 17?19, 2026, and use outage-prompt.txt. Verify no new findings, old window retained separately and identical last-good bytes.
6. Compare file hashes before and after. Actual receipts and accepted outputs are included. Live retrieval, recurring execution and external delivery are not tested here.