pub enum Rationale {
Legacy(String),
Thinking(Vec<Thought>),
}Expand description
Why a seat said what it said.
Until 0.64 the seats wrote their reasoning into a reasoning field of
the decision tool, which is not something a model should be asked to
do: Anthropic’s reasoning_extraction safeguard declines exactly that
(docs). The seats now think before they answer, and the record keeps
what the API returns for that thinking.
Variants§
Legacy(String)
What the seat wrote in the decision tool’s reasoning field
Thinking(Vec<Thought>)
The thinking blocks of the seat’s turn, in order and as the API
returned them. On Anthropic models each is a summary of the model’s
reasoning, written by Anthropic, with a signature over the full
reasoning; the SeatModels entry says which model to replay it
against. Local models’ blocks are their own reasoning, unsigned.
Implementations§
Trait Implementations§
Source§impl<'de> Deserialize<'de> for Rationale
impl<'de> Deserialize<'de> for Rationale
Source§fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>where
__D: Deserializer<'de>,
fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>where
__D: Deserializer<'de>,
Source§impl From<&Content> for Rationale
The thinking blocks of content; anything else in it is not thinking
impl From<&Content> for Rationale
The thinking blocks of content; anything else in it is not thinking