1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
/*
* Hanzo Cloud API
*
* The Hanzo Cloud API as a customer calls it: every operation under /v1/ except the operator's admin product, relay routes, legacy spellings and capabilities still reached by flag. Tagged by product: the first path segment after /v1/.
*
* The version of the OpenAPI document: v1
*
* Generated by: https://openapi-generator.tech
*/
use crate::models;
use serde::{Deserialize, Serialize};
#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
pub struct LeaderRow {
/// CIHigh is the upper bound of that interval. Wilson rather than the normal approximation because the normal one produces bounds past 100 exactly where benchmark scores live — at 194/198 that is the top of the board, not a corner case.
#[serde(rename = "ciHigh", skip_serializing_if = "Option::is_none")]
pub ci_high: Option<f64>,
/// CILow and CIHigh are the 95% Wilson interval on Measured, in percent. They are what makes the score comparable: at n=198 a 98% carries roughly ±2 points, so most differences at the top of a board are not distinguishable and a bare number implies a precision it does not have. Absent when there is no measurement.
#[serde(rename = "ciLow", skip_serializing_if = "Option::is_none")]
pub ci_low: Option<f64>,
/// Claims is how many independent claims exist for this model on this benchmark. More than one means several sources reported it.
#[serde(rename = "claims", skip_serializing_if = "Option::is_none")]
pub claims: Option<i32>,
/// published − measured (the arena signal)
#[serde(rename = "gap", skip_serializing_if = "Option::is_none")]
pub gap: Option<f64>,
/// Mean is the unweighted average of every claim, which answers a different question from Published: what the field says on average, rather than what the vendor says about itself. With one claim the two are equal.
#[serde(rename = "mean", skip_serializing_if = "Option::is_none")]
pub mean: Option<f64>,
/// hanzo-measured accuracy % (nil if unrun)
#[serde(rename = "measured", skip_serializing_if = "Option::is_none")]
pub measured: Option<f64>,
/// MeasuredAt is when the run behind Measured was recorded.
#[serde(rename = "measuredAt", skip_serializing_if = "Option::is_none")]
pub measured_at: Option<String>,
/// the model this row scores
#[serde(rename = "model", skip_serializing_if = "Option::is_none")]
pub model: Option<String>,
/// coverage — NEVER compare across different n
#[serde(rename = "n", skip_serializing_if = "Option::is_none")]
pub n: Option<i32>,
/// how the vendor scored their claim: single-attempt, pass@k or agentic
#[serde(rename = "protocol", skip_serializing_if = "Option::is_none")]
pub protocol: Option<String>,
/// provider-claimed % (nil if none)
#[serde(rename = "published", skip_serializing_if = "Option::is_none")]
pub published: Option<f64>,
/// Run names the measurement Measured came from, and MeasuredAt is when it ran. A score with no date is not a fact about a model, it is a fact about a model on a day — and models change, so the date is what makes the number checkable rather than merely quoted.
#[serde(rename = "run", skip_serializing_if = "Option::is_none")]
pub run: Option<String>,
/// Spread is the distance between the highest and lowest of them, nil when there is only one. It is the disagreement AMONG sources, which a single Published number cannot show — signal in the same way the published-minus-measured gap is.
#[serde(rename = "spread", skip_serializing_if = "Option::is_none")]
pub spread: Option<f64>,
}
impl LeaderRow {
pub fn new() -> LeaderRow {
LeaderRow {
ci_high: None,
ci_low: None,
claims: None,
gap: None,
mean: None,
measured: None,
measured_at: None,
model: None,
n: None,
protocol: None,
published: None,
run: None,
spread: None,
}
}
}