Iterate the evicted target ids in arbitrary order.

Gets an immutable typed extension by Rust type.

Inserts a typed extension, returning the previous value of the same type.

Returns whether the extension map is empty.

Appends one observation.

Return *request* in the target wire format.

Build an Anthropic Messages request.

Return the builder method for *preset_id*. Raises: ValueError: If *preset_id* is not in :attr:`PRESETS`. The err

Locks the context unless an async Rust component currently owns it.

Build an OpenAI Chat Completions request.

Return a cached value and mark it as most recently used.

Parse *raw* dict and build a :class:`RouteTable`. Thin entrypoint that the YAML loader and any other dict-driven caller uses. Parses + env-ex

Build a buffered OpenAI Chat Completions response.

Creates a one-shot server returning JSON.

Iterate ``(model_id, chain, metadata)`` triples in registration order. Used when a caller needs to merge one table into another — e.g. a

Process a request to completion, returning the final [`Response`] and the trace of [`Decision`]s the algorithm made along the way.

Translates a request body from source format to target format.

Return the concrete backend format for ``tier``. Explicit formats are already resolved. ``AUTO`` needs a real endpoint probe, so miss

Returns the stable lowercase tier label.

Build an OpenAI Responses request.

Push a fire-and-forget [`DriverStep::Info`] payload; there is no promise to await for a response. Awaits channel capacity (the consumer pacing the str

Returns the number of stored extension values.

Build a streaming OpenAI Chat Completions response.

Returns a computed snapshot suitable for JSON serialization.

Apply one chunk. Later `MessageStart`/`Usage`/`MessageStop` fields overwrite earlier ones; text, reasoning, and tool-call arguments append.

The prompt configured for `target`, if any.

Return whether a Rust-backed response has the requested wire shape.

Builds a target/backend pair for `MultiLlmBackend`.

Create a FastAPI app serving *switchyard* over all inbound formats. Registers three LLM endpoints plus a liveness probe: - ``POST /v1/chat/c

Appends a classifier to the cascade.

Returns whether a value of type `T` is present.

Return whether a request has the requested wire format.

Builds a contract from user settings and packaged defaults. The response format must contain `json_schema.schema` and is retained separately for the

Register *switchyard* as the exact-match chain for *model*.

Waits for and returns the captured request.

Records one successful routed backend call.

Returns a serializable point-in-time snapshot.

Creates a warning diagnostic.

Calls the backend for `model_name` (or the request's own model), over the wire format the request pins in its metadata (else the model's default backe

Whether any target has a prompt, so a caller can skip wiring the processor when none does.

Return the chain for *model*. Records *model* as the last successfully resolved id (see :attr:`last_looked_up`). Raises:

Returns a cloned copy of all observations.

Returns the wrapped JSON request body.

Remembers that `target` overflowed for `identity`, tracking at most [`MAX_EVICTION_IDENTITIES`] identities.

Records usage and terminal latency after a successful routed call.

Removes a typed extension and returns it when present.

Builds a single-turn request: one user message carrying `prompt`, for `model`. This deliberately creates only the common text shape. Construct [`LlmR

Translates one streaming source event into zero or more target events.

Translates a response body from source format to target format.

Build the profile-backed runtime used by these tests.

Retain asynchronous lifecycle compatibility without owning a transport.

Build a buffered Anthropic response.

Records request-start metadata and returns the request unchanged.

Return a copy with serving-level stats and processor hooks applied. Profile configs stay user-facing and parseable; shared serving resources

Reduce to the buffered aggregate: return an `Agg` unchanged, or drive a `Stream` to completion, folding its normalized chunks into an [`AggLlmResponse

Take the single consumer stream of [`DriverStep`]s. Callable once: a second call yields a one-item stream carrying an `Err`, since the receiver is gon

Builds a target with explicit endpoint credentials for native backend tests.

Chat request with enough prior assistant turns to pass the judge gate.

Stable lowercase label used in stats.

Binds the configured address and prepares the HTTP router.

A one-model config list: "gpt" served over OpenAI Chat at base_url.

Fetches one model stats block with an explicit test error on absence.