ATIF-v1.8 projection of ordered native Imp.Run.Event observations.

to_atif/2 projects one run's ordered events into a JSON-encodable ATIF document. The initial model context keeps its actual message roles and is marked as copied context; every model request is also listed in extra.model_requests. A model response step carries the observed typed output. Lifecycle events and capture gaps become entries in extra.diagnostics rather than dialogue steps. extra.terminal_event is the kind of the event that ended the run ("run_finished", "run_failed", "run_cancelled"), or "unknown" when the events hold none.

A tool result's extra.outcome, on the result and on its call step, is the Imp.Tool.outcome/0 the loop recorded on the :tool_result event (metadata.outcome); an event that carries none, and a call whose result was not captured, is "unknown".

Stored event maps are read by their "kind". A kind that is not one of Imp.Run.Event.kinds/0, such as one a host emitted itself, keeps its stored string and becomes a diagnostic entry.

Semantic tool dispatches have llm_call_count: 0; a model response leaves the count null, because a request may have been served from a cache. Inference counts, metrics, reasoning and tool results are never synthesized from missing evidence. A tool observation attaches to its call step as ATIF requires, retaining the native sequence and time; step tool call IDs are scoped by run and event sequence. A provider tool call ID may be reused once its result has arrived, but an overlapping reuse raises ArgumentError.

The document is a projection, not a replay and not a durable effect ledger. Streaming paths that bypass Imp.LM.request/2 do not produce complete model episodes. Redaction runs again on export, including caller metadata; prompts and results are still private application data.