weavatrix-memory 0.3.2

Event-sourced bitemporal evidence memory and budgeted context retrieval for coding agents
Documentation 
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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146

use super::model::{Document, PreparedBenchmark, PreparedCase};
use blazingly_json::Value;
use std::collections::{BTreeMap, BTreeSet};

pub fn prepare(bytes: &[u8]) -> Result<PreparedBenchmark, String> {
    let samples =
        blazingly_json::from_slice::<Vec<Value>>(bytes).map_err(|error| error.to_string())?;
    let mut groups = BTreeMap::new();
    let mut cases = Vec::new();
    for (sample_index, sample) in samples.iter().enumerate() {
        let sample_id = sample
            .get("sample_id")
            .and_then(Value::as_str)
            .map_or_else(|| format!("sample-{sample_index}"), str::to_owned);
        let documents = conversation_documents(sample, &sample_id)?;
        let document_ids = documents
            .iter()
            .map(|document| document.id.as_str())
            .collect::<BTreeSet<_>>();
        let qa = sample
            .get("qa")
            .and_then(Value::as_array)
            .ok_or_else(|| format!("{sample_id} has no qa array"))?;
        for (question_index, question) in qa.iter().enumerate() {
            let query = required_string(question, "question")?;
            let evidence = question
                .get("evidence")
                .and_then(Value::as_array)
                .into_iter()
                .flatten()
                .filter_map(Value::as_str)
                .flat_map(split_evidence)
                .map(|id| document_id(&sample_id, id))
                .filter(|id| document_ids.contains(id.as_str()))
                .collect::<BTreeSet<_>>();
            if evidence.is_empty() {
                continue;
            }
            let category = question
                .get("category")
                .map_or_else(|| "unknown".to_owned(), value_text);
            cases.push(PreparedCase {
                id: format!("{sample_id}::qa::{question_index}"),
                group_id: sample_id.clone(),
                category: format!("locomo_{category}"),
                query,
                relevant_ids: evidence,
            });
        }
        groups.insert(sample_id, documents);
    }
    Ok(PreparedBenchmark {
        name: "locomo".to_owned(),
        groups,
        cases,
    })
}

fn conversation_documents(sample: &Value, sample_id: &str) -> Result<Vec<Document>, String> {
    let conversation = sample
        .get("conversation")
        .and_then(Value::as_object)
        .ok_or_else(|| format!("{sample_id} has no conversation object"))?;
    let mut documents = Vec::new();
    for (key, turns) in conversation {
        if !key.starts_with("session_") || key.ends_with("_date_time") {
            continue;
        }
        let Some(turns) = turns.as_array() else {
            continue;
        };
        for turn in turns {
            let Some(id) = turn.get("dia_id").and_then(Value::as_str) else {
                continue;
            };
            let Some(text) = turn.get("text").and_then(Value::as_str) else {
                continue;
            };
            let speaker = turn
                .get("speaker")
                .and_then(Value::as_str)
                .unwrap_or("speaker");
            documents.push(Document {
                id: document_id(sample_id, id),
                text: format!("{speaker}: {text}"),
            });
        }
    }
    documents.sort_by(|left, right| left.id.cmp(&right.id));
    Ok(documents)
}

fn split_evidence(value: &str) -> impl Iterator<Item = &str> {
    value
        .split(';')
        .map(str::trim)
        .filter(|value| !value.is_empty())
}

fn document_id(sample: &str, dialog: &str) -> String {
    format!("{sample}::dialog::{dialog}")
}

fn required_string(value: &Value, field: &str) -> Result<String, String> {
    value
        .get(field)
        .and_then(Value::as_str)
        .map(str::to_owned)
        .ok_or_else(|| format!("missing string field {field}"))
}

fn value_text(value: &Value) -> String {
    value
        .as_str()
        .map_or_else(|| value.to_string(), str::to_owned)
}

#[cfg(test)]
mod tests {
    #[test]
    fn adapter_splits_compound_dialog_evidence() {
        let input = br#"[
          {
            "sample_id":"conv",
            "qa":[
              {"question":"What happened?","category":2,"evidence":["D1:1; D1:2"]},
              {"question":"Unknown?","category":3,"evidence":[]}
            ],
            "conversation":{
              "speaker_a":"A",
              "speaker_b":"B",
              "session_1_date_time":"today",
              "session_1":[
                {"speaker":"A","dia_id":"D1:1","text":"first"},
                {"speaker":"B","dia_id":"D1:2","text":"second"}
              ]
            }
          }
        ]"#;
        let benchmark = super::prepare(input).unwrap();

        assert_eq!(benchmark.groups["conv"].len(), 2);
        assert_eq!(benchmark.cases.len(), 1);
        assert_eq!(benchmark.cases[0].relevant_ids.len(), 2);
    }
}