cupel 1.2.0

Context window management pipeline for LLM applications
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

use crate::CupelError;
use crate::model::{ContextBudget, ContextItem};

/// Result type for [`classify`]: `(pinned, scoreable, negative_token_items)`.
type ClassifyResult = Result<(Vec<ContextItem>, Vec<ContextItem>, Vec<ContextItem>), CupelError>;

/// Partitions input items into pinned and scoreable lists, excluding invalid items.
///
/// Items with `tokens < 0` are excluded (checked before pinned status) and
/// returned as the third tuple element so callers can record them.
/// Validates that pinned items fit within `maxTokens - outputReserve`.
pub(crate) fn classify(items: &[ContextItem], budget: &ContextBudget) -> ClassifyResult {
    let mut pinned = Vec::new();
    let mut scoreable = Vec::new();
    let mut neg_items = Vec::new();

    for item in items {
        // Negative-token check BEFORE pinned check
        if item.tokens() < 0 {
            neg_items.push(item.clone());
            continue;
        }
        if item.pinned() {
            pinned.push(item.clone());
        } else {
            scoreable.push(item.clone());
        }
    }

    // Validate pinned budget
    let pinned_tokens: i64 = pinned.iter().map(|i| i.tokens()).sum();
    let available = budget.max_tokens() - budget.output_reserve();

    if pinned_tokens > available {
        return Err(CupelError::PinnedExceedsBudget {
            pinned_tokens,
            available,
        });
    }

    Ok((pinned, scoreable, neg_items))
}