Which actions and state transitions could a simulated environment expose?

MirrorFirm could generate an environment with separate obligation, case and communication state. An initial snapshot could contain `obligation_id`, `case_id`, `case_version`, `assigned_role`, `follow_up_stage` and `next_action_at`, linked to policy and communication records.

Tool actions could include `get_case`, `list_contact_history`, `draft_request`, `submit_follow_up`, `record_response` and `refer_for_review`. Each action needs a defined input schema, authorisation check and state effect. Drafting a request should not implicitly record a sent communication. Submitting it could create an event and advance the case only when the applicable preconditions hold.

A `reset_episode` operation could restore the starting snapshot and clear episode-created events. A `step_result` response could expose the action outcome, updated record versions and any policy rejection. Evaluation hooks could capture attempted actions as well as successful transitions, making rejected behaviour visible.

A reseller can request a JSON tool contract alongside Parquet snapshot tables and an event schema. The licence discussion should distinguish distribution of static assets from access to executable environment behaviour. These are different product surfaces, even when they represent the same generated workflow.