OpenCode 1.18.34 default-title scenario
On 2026-10-05, the official CLI at source revision aec0b9a6d8898f68f923aaf08b7306d931fd9d76 called a real local CPU model in isolated rootless Podman with networking disabled and fresh HOME. Tokens are native; the container record describes the procedure and source. Extraction retains the original three-table DDL and permitted usage fields. Bodies are removed; IDs, paths and titles are anonymized. Timestamps share one offset into 2026-10-05 UTC; tokens are unchanged. Message and session cumulative totals are comparison values and are never added to individual usage.
| Measurement | Calls / usage |
|---|---|
| Native main-loop API usage | input 298 (including cache read 3), output 1, total 299 |
| CLI / part / assistant message / session | uncached input 295, cache read 3, write 0, output 1, reasoning 0, total 299 |
| Application | one event; input_total 298, output_total 1, total 299; repeat scan changes no events, revisions or summaries |
| Default-title API usage | another call: input 539, output 10, total 549; absent from step-finish and session cumulative totals |
The two default API calls total 848. A matched comparison between main-loop events and cumulative totals does not establish complete client coverage. Do not invent a title event from the difference or infer a cloud model or price from the local model name. The client wrote cost=0; no actual bill or positive cost was verified. Version handling remains latest_fallback. Per-record version verification and old-cursor upgrade testing are unfinished, so this sample cannot verify other sessions in the same database.