fetchium-core 1.0.0

Fetchium core library — search, extract, rank, validate, cache
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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
//! Progressive Detail Streaming (PDS) — PRD §27.
//!
//! 4-tier content delivery system that adapts output depth to the
//! caller's token budget. Agents can request exactly the level of
//! detail they need without over-fetching.
//!
//! Tier 1 (key_facts):  ~200  tokens — top 3-5 facts, highest-relevance segments only
//! Tier 2 (summary):    ~1000 tokens — overview + key points, moderate context
//! Tier 3 (detailed):   ~5000 tokens — full content with all context
//! Tier 4 (complete):   unlimited    — raw extracted text, no truncation

use crate::token::counter::{count_tokens, TokenBudget};
use crate::types::{PdsTier, Segment, SegmentType};
use serde::{Deserialize, Serialize};
use tracing::debug;

/// Token targets for each PDS tier.
pub const KEY_FACTS_BUDGET: u32 = 200;
pub const SUMMARY_BUDGET: u32 = 1_000;
pub const DETAILED_BUDGET: u32 = 5_000;

/// Result of a PDS operation.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct PdsResult {
    /// Tier applied.
    pub tier: PdsTier,
    /// Content adapted to the tier budget.
    pub content: String,
    /// Segments selected for this tier (for agent output).
    pub segments: Vec<Segment>,
    /// Tokens in the result.
    pub tokens_used: u32,
    /// Whether the content was truncated from the original.
    pub truncated: bool,
}

/// Apply PDS to extracted text, producing a tier-appropriate view.
///
/// # Arguments
/// * `text` - Full extracted text from CEP pipeline
/// * `segments` - Pre-scored QATBE segments (sorted by relevance, highest first)
/// * `tier` - Requested PDS tier
///
/// # Returns
/// A `PdsResult` containing content adapted to the tier budget.
pub fn apply_tier(text: &str, segments: &[Segment], tier: PdsTier) -> PdsResult {
    match tier {
        PdsTier::KeyFacts => apply_key_facts(text, segments),
        PdsTier::Summary => apply_summary(text, segments),
        PdsTier::Detailed => apply_detailed(text, segments),
        PdsTier::Complete => apply_complete(text, segments),
    }
}

/// Tier 1: key_facts — highest-relevance segments only, ~200 tokens.
///
/// Selects the top 3-5 segments by relevance score. If no pre-scored
/// segments are available, returns the first 200 tokens of text.
fn apply_key_facts(text: &str, segments: &[Segment]) -> PdsResult {
    let budget = KEY_FACTS_BUDGET;

    if segments.is_empty() {
        // Fallback: take first N tokens of raw text
        let content = truncate_to_budget(text, budget);
        let tokens_used = count_tokens(&content);
        let truncated = tokens_used < count_tokens(text);
        debug!("PDS key_facts (fallback): {} tokens", tokens_used);
        return PdsResult {
            tier: PdsTier::KeyFacts,
            content,
            segments: Vec::new(),
            tokens_used,
            truncated,
        };
    }

    // Take top-relevance segments within budget, max 5
    let mut selected = Vec::new();
    let mut tracker = TokenBudget::new(budget);
    let mut content_parts = Vec::new();

    // Segments should already be sorted by relevance desc
    for seg in segments.iter().take(10) {
        if tracker.is_exhausted() || selected.len() >= 5 {
            break;
        }
        if tracker.try_consume(seg.tokens) {
            let text_val = segment_to_text(seg);
            content_parts.push(text_val);
            selected.push(seg.clone());
        }
    }

    let content = content_parts.join("\n\n");
    let tokens_used = tracker.used;
    let truncated = tokens_used < count_tokens(text);

    debug!(
        "PDS key_facts: {} segments, {} tokens",
        selected.len(),
        tokens_used
    );

    PdsResult {
        tier: PdsTier::KeyFacts,
        content,
        segments: selected,
        tokens_used,
        truncated,
    }
}

/// Tier 2: summary — overview + key points, ~1000 tokens.
///
/// Takes top segments by relevance up to the summary budget.
/// Adds a brief heading structure if the content permits.
fn apply_summary(text: &str, segments: &[Segment]) -> PdsResult {
    let budget = SUMMARY_BUDGET;

    if segments.is_empty() {
        let content = truncate_to_budget(text, budget);
        let tokens_used = count_tokens(&content);
        let truncated = tokens_used < count_tokens(text);
        debug!("PDS summary (fallback): {} tokens", tokens_used);
        return PdsResult {
            tier: PdsTier::Summary,
            content,
            segments: Vec::new(),
            tokens_used,
            truncated,
        };
    }

    let mut selected = Vec::new();
    let mut tracker = TokenBudget::new(budget);
    let mut content_parts = Vec::new();

    // Include headings first (structural anchors), then body content
    let headings: Vec<&Segment> = segments
        .iter()
        .filter(|s| s.seg_type == SegmentType::Heading)
        .take(3)
        .collect();
    let body: Vec<&Segment> = segments
        .iter()
        .filter(|s| s.seg_type != SegmentType::Heading)
        .collect();

    for seg in headings.into_iter().chain(body) {
        if tracker.is_exhausted() {
            break;
        }
        if tracker.try_consume(seg.tokens) {
            content_parts.push(segment_to_text(seg));
            selected.push(seg.clone());
        } else {
            // Try partial inclusion for final segment
            let remaining = tracker.remaining();
            if remaining > 20 {
                let text_val = segment_to_text(seg);
                let truncated_text = truncate_to_budget(&text_val, remaining);
                let t = count_tokens(&truncated_text);
                tracker.consume(t);
                let mut partial = seg.clone();
                partial.tokens = t;
                partial.content = serde_json::Value::String(truncated_text.clone());
                content_parts.push(truncated_text);
                selected.push(partial);
            }
            break;
        }
    }

    let content = content_parts.join("\n\n");
    let tokens_used = tracker.used;
    let truncated = tokens_used < count_tokens(text);

    debug!(
        "PDS summary: {} segments, {} tokens",
        selected.len(),
        tokens_used
    );

    PdsResult {
        tier: PdsTier::Summary,
        content,
        segments: selected,
        tokens_used,
        truncated,
    }
}

/// Tier 3: detailed — full content with context, ~5000 tokens.
///
/// Includes all segments up to the detailed budget. For most pages
/// this captures the entire article.
fn apply_detailed(text: &str, segments: &[Segment]) -> PdsResult {
    let budget = DETAILED_BUDGET;

    if segments.is_empty() {
        let content = truncate_to_budget(text, budget);
        let tokens_used = count_tokens(&content);
        let truncated = tokens_used < count_tokens(text);
        debug!("PDS detailed (fallback): {} tokens", tokens_used);
        return PdsResult {
            tier: PdsTier::Detailed,
            content,
            segments: Vec::new(),
            tokens_used,
            truncated,
        };
    }

    let mut selected = Vec::new();
    let mut tracker = TokenBudget::new(budget);
    let mut content_parts = Vec::new();

    // For detailed, restore original document order (by source_ref position)
    let mut ordered: Vec<&Segment> = segments.iter().collect();
    ordered.sort_by_key(|s| s.source_ref.unwrap_or(u32::MAX));

    for seg in ordered {
        if tracker.is_exhausted() {
            break;
        }
        if tracker.try_consume(seg.tokens) {
            content_parts.push(segment_to_text(seg));
            selected.push(seg.clone());
        } else {
            let remaining = tracker.remaining();
            if remaining > 50 {
                let text_val = segment_to_text(seg);
                let truncated_text = truncate_to_budget(&text_val, remaining);
                let t = count_tokens(&truncated_text);
                tracker.consume(t);
                let mut partial = seg.clone();
                partial.tokens = t;
                partial.content = serde_json::Value::String(truncated_text.clone());
                content_parts.push(truncated_text);
                selected.push(partial);
            }
            break;
        }
    }

    let content = content_parts.join("\n\n");
    let tokens_used = tracker.used;
    let truncated = tokens_used < count_tokens(text);

    debug!(
        "PDS detailed: {} segments, {} tokens",
        selected.len(),
        tokens_used
    );

    PdsResult {
        tier: PdsTier::Detailed,
        content,
        segments: selected,
        tokens_used,
        truncated,
    }
}

/// Tier 4: complete — raw extracted text, no truncation.
///
/// Returns the entire extracted text as-is, plus all segments.
/// Use this when you want everything and handle truncation yourself.
fn apply_complete(text: &str, segments: &[Segment]) -> PdsResult {
    let tokens_used = count_tokens(text);
    debug!("PDS complete: {} tokens", tokens_used);

    PdsResult {
        tier: PdsTier::Complete,
        content: text.to_string(),
        segments: segments.to_vec(),
        tokens_used,
        truncated: false,
    }
}

/// Convert a segment's content to plain text for output.
fn segment_to_text(seg: &Segment) -> String {
    match &seg.content {
        serde_json::Value::String(s) => s.clone(),
        serde_json::Value::Object(_) | serde_json::Value::Array(_) => {
            serde_json::to_string_pretty(&seg.content).unwrap_or_default()
        }
        other => other.to_string(),
    }
}

/// Truncate text to approximately fit within a token budget.
/// Cuts at a word boundary and appends "..." if truncated.
fn truncate_to_budget(text: &str, budget: u32) -> String {
    let current_tokens = count_tokens(text);
    if current_tokens <= budget {
        return text.to_string();
    }

    // Estimate max characters from budget
    let max_chars = (budget as f64 * 4.0) as usize;
    if max_chars >= text.len() {
        return text.to_string();
    }

    // Find safe char boundary at or before max_chars
    let safe_end = {
        let target = max_chars.min(text.len());
        let mut idx = target;
        while idx > 0 && !text.is_char_boundary(idx) {
            idx -= 1;
        }
        idx
    };
    let truncated = &text[..safe_end];
    match truncated.rfind([' ', '\n']) {
        Some(pos) if pos > safe_end / 2 => {
            format!("{}...", truncated[..pos].trim_end())
        }
        _ => format!("{truncated}..."),
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::types::SegmentType;

    fn make_segment(text: &str, seg_type: SegmentType, relevance: f64, position: u32) -> Segment {
        Segment {
            seg_type,
            relevance,
            tokens: count_tokens(text),
            content: serde_json::Value::String(text.to_string()),
            source_ref: Some(position),
        }
    }

    #[test]
    fn key_facts_respects_budget() {
        let text = "Long article content with many paragraphs. ".repeat(200);
        let segs = vec![
            make_segment("Key finding one about Rust.", SegmentType::Fact, 0.9, 0),
            make_segment(
                "Key finding two about performance.",
                SegmentType::Fact,
                0.8,
                1,
            ),
            make_segment(
                "Supporting detail with more context.",
                SegmentType::Paragraph,
                0.5,
                2,
            ),
        ];
        let result = apply_tier(&text, &segs, PdsTier::KeyFacts);
        assert_eq!(result.tier, PdsTier::KeyFacts);
        assert!(result.tokens_used <= KEY_FACTS_BUDGET + 10); // small slack for estimation
        assert!(!result.content.is_empty());
    }

    #[test]
    fn summary_tier_includes_more_than_key_facts() {
        let text = "Article with lots of content. ".repeat(200);
        let segs: Vec<Segment> = (0..20)
            .map(|i| {
                make_segment(
                    &format!("Segment {i} with some substantial content that fills tokens."),
                    SegmentType::Paragraph,
                    1.0 - (i as f64 * 0.05),
                    i as u32,
                )
            })
            .collect();

        let key_facts = apply_tier(&text, &segs, PdsTier::KeyFacts);
        let summary = apply_tier(&text, &segs, PdsTier::Summary);
        assert!(summary.tokens_used >= key_facts.tokens_used);
        assert!(summary.tokens_used <= SUMMARY_BUDGET + 20);
    }

    #[test]
    fn complete_tier_returns_everything() {
        let text = "Short content.";
        let result = apply_tier(text, &[], PdsTier::Complete);
        assert_eq!(result.tier, PdsTier::Complete);
        assert_eq!(result.content, text);
        assert!(!result.truncated);
    }

    #[test]
    fn truncate_at_word_boundary() {
        let text = "Hello world this is a long piece of text that should be truncated.";
        let truncated = truncate_to_budget(text, 5);
        assert!(truncated.ends_with("..."));
        assert!(count_tokens(&truncated) <= 10); // with some slack
    }

    #[test]
    fn key_facts_fallback_without_segments() {
        let text = "This is the article content. ".repeat(50);
        let result = apply_tier(&text, &[], PdsTier::KeyFacts);
        assert!(result.tokens_used <= KEY_FACTS_BUDGET + 10);
        assert!(!result.content.is_empty());
    }

    #[test]
    fn detailed_preserves_document_order() {
        let text = "Doc content. ".repeat(100);
        let segs = vec![
            make_segment("Third section.", SegmentType::Paragraph, 0.5, 2),
            make_segment("First section.", SegmentType::Paragraph, 0.9, 0),
            make_segment("Second section.", SegmentType::Paragraph, 0.7, 1),
        ];
        let result = apply_tier(&text, &segs, PdsTier::Detailed);
        // Should be reordered by source_ref: first, second, third
        let content = result.content;
        let first_pos = content.find("First").unwrap_or(usize::MAX);
        let second_pos = content.find("Second").unwrap_or(usize::MAX);
        let third_pos = content.find("Third").unwrap_or(usize::MAX);
        assert!(first_pos < second_pos, "First should come before Second");
        assert!(second_pos < third_pos, "Second should come before Third");
    }
}