Adds two new fields to the unified completion Response:
- cost_usd: Option<f64> — USD cost for the completion
- cost_source: Option<CostSource> — Authoritative | Estimated
A new fabro_llm::cost::estimate_cost_usd helper centralizes the catalog
math (tokens × catalog price → USD micros → f64). Every adapter response
path (Anthropic, OpenAI, Gemini, openai_chat) calls the helper when a
catalog is available, producing Estimated costs. Adapters with
authoritative provider-side pricing (e.g. OpenRouter, in a later commit)
will bypass the helper and set Authoritative.
The openai_chat streaming path now plumbs catalog + speed through
StreamState so finish_events can call the helper before assembling the
final Response.
Public API surface: CompletionResponse in fabro-api.yaml is extended
with cost_usd and cost_source. Progenitor + the TS client are
regenerated; apps/fabro-web typechecks clean since both fields are
optional.
42 Response constructor sites updated:
- 5 live adapter sites call estimate_cost_usd
- ~32 synthesized/fallback/test sites set (None, None) explicitly
- 5 test fixtures in types.rs set (None, None)
397 lib tests pass (was 392, +5 new helper unit tests).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
480ff56aba