These accompanied the BDD subsystem buildout: how a single scenario maps to a cluster of sessions, how the orchestrator dispatches verbs across multiple personas, how scenario-outline column values are substituted at the right boundary, and the full porting plan from the existing Walks framework. Self-contained HTML with hand-drawn flow diagrams, sequence rails, and inline rationale — the format we reach for when a prose ADR can't carry the geometry of the decision.
The umbrella plan. Maps every concept in the legacy Walks framework — actors, screens, transitions, verbs, asserts — onto its BDD equivalent. Includes a delta table, migration phases, and the suspicion-first reading of which patterns survive the move and which need rethinking from the ground.
Open visualization →The cluster implementation plan. Each scenario is a cluster of sessions across personas; this is how they spawn, run, observe, and tear down. Includes the agent-tui driver inventory and the journey diagram from cluster start → per-persona checkpoints → cluster done.
Open visualization →Session-vs-persona model and the verb-routing semantics. How a single Gherkin step like "Alice raises her hand" gets dispatched to the right persona's BA (Business Action) handler in the right format, with the right shared state. Lists the scenarios this unblocks.
Open visualization →
Substitution semantics — where in the pipeline the
<column> placeholders in a Gherkin
outline get replaced with the actual values from the
Examples table. Existing-machinery check, location
decision, format-to-BA dispatch flow. The "subtle correctness"
plan — getting this wrong silently produces tests that pass
the wrong way.