# Reproduce: Turn source leads into research tasks without promoting unsupported claims

These records, people and .invalid links are synthetic. No live account is connected.

1. Put sources.json and prompt.txt in a new dedicated trial folder. Keep the supplied accepted result.json elsewhere for comparison.
2. Replace <trial-folder> in prompt.txt with the absolute path of that folder.
3. Use your configured Hermes model and 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. Read the new result.json, compare every retained source identity against sources.json and check the fixture-specific acceptance checks in receipt.json. Hash the source file before and after to ensure it was not modified.
5. The receipt records the actual provider/model, observed tool calls, raw trace hash, input hash and accepted output hash. Safe mode and file tools bounded this local exercise. The provider allowance for a new run belongs to your existing configured account; site builds do not rerun these trials.

Scope: Real Hermes synthesis of five synthetic research leads into an unpublished editorial queue.

No live integration, external writes or broad production reliability claim follows from this exercise.