zeroclaw_eval/
record.rs

1//! The artifact produced by running one eval case — what graders score.
2
3use zeroclaw_api::model_provider::ConversationMessage;
4
5use crate::observer::RecordedCall;
6
7/// Everything captured from a single agent run, ready for grading.
8pub struct RunRecord {
9    /// The agent's final text response for the case.
10    pub final_response: String,
11    /// The full conversation trajectory (messages + tool calls + tool results).
12    pub history: Vec<ConversationMessage>,
13    /// Every dispatched tool call with its arguments and result, in call order.
14    /// This creates the canonical dispatch fact: names, success, arguments, and
15    /// results are derived from this list rather than copied into parallel fields.
16    pub tool_calls: Vec<RecordedCall>,
17    /// Accumulated input tokens reported by the provider.
18    pub input_tokens: u64,
19    /// Accumulated output tokens reported by the provider.
20    pub output_tokens: u64,
21}
22
23impl RunRecord {
24    /// Names of tools actually dispatched, in call order.
25    pub fn tool_names(&self) -> Vec<&str> {
26        self.tool_calls
27            .iter()
28            .map(|call| call.name.as_str())
29            .collect()
30    }
31
32    /// Whether every dispatched tool call succeeded (vacuously true if none).
33    pub fn all_tools_succeeded(&self) -> bool {
34        self.tool_calls.iter().all(|call| call.success)
35    }
36}
zeroclaw_eval/record.rs

zeroclaw_eval/
record.rs
