Enum ReasoningPlan 

pub enum ReasoningPlan {
    Unset,
    Sent(Reasoning),
    Attempted(Reasoning),
    Unsupported(Reasoning),
}

What a configured Reasoning will do on the provider that was selected.

Reasoning tokens are billed, so “asked for, and this provider has nowhere to put it” has to be tellable from “never asked for”. Both used to resolve to the same None on SelectedLlm, which left korps doctor unable to report a setting the run would quietly discard.

Variants

Unset

Nothing was configured; the model’s own default stands.

Sent(Reasoning)

Configured, and this provider puts it on the wire.

Attempted(Reasoning)

Configured, and this provider sends it — but whether the model accepts it is not known until the first call.

OpenAI’s reasoning_effort and Gemini’s thinkingConfig are refused by the models that do not reason, and which models those are changes with every release. So the parameter is sent, a refusal is recognized from the 400, and the request is retried without it — meaning this setting may still end up dropped, and only the run knows.

Unsupported(Reasoning)

Configured, and this provider has no field that carries it. Dropped before any request is made.

Implementations

impl ReasoningPlan

pub fn carried(reasoning: Option<Reasoning>) -> Self

The plan for a provider that can carry whatever was configured.

pub fn attempted(reasoning: Option<Reasoning>) -> Self

The plan for a provider that sends it and finds out from the model.

pub fn dropped(reasoning: Option<Reasoning>) -> Self

The plan for a provider with no reasoning field.

pub fn sent(self) -> Option<Reasoning>

What reaches the wire, if anything. A setting dropped before the request reads as nothing here, because nothing is what gets sent; an attempted one reads as sent, because it is on the first request.

pub fn requested(self) -> Option<Reasoning>

What was configured, whether or not it will be sent.

pub fn unsupported(self) -> Option<Reasoning>

What this provider was asked for and will not send. Only the settings answered before a request is made — see Self::attempted for the ones the model gets the last word on.

pub fn attempting(self) -> Option<Reasoning>

What this provider will send and may still have refused. None when the answer is already known either way.

Trait Implementations

impl Clone for ReasoningPlan

fn clone(&self) -> ReasoningPlan

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for ReasoningPlan

impl Debug for ReasoningPlan

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ReasoningPlan

fn default() -> ReasoningPlan

Returns the “default value” for a type.

impl Display for ReasoningPlan

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for ReasoningPlan

impl PartialEq for ReasoningPlan

fn eq(&self, other: &ReasoningPlan) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for ReasoningPlan