switchyard_libsy/algorithms/util/
escalation.rs

1// SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2// SPDX-License-Identifier: Apache-2.0
3
4//! Trajectory-judge components for the escalation router — the judge, its verdict policy, and
5//! the transcript condenser they read.
6//!
7//! [`build_judge`] is the whole surface; the confirmation policy that consumes its verdicts
8//! lives with the assembled algorithm in [`crate::algorithms::escalation`].
9
10use serde::Deserialize;
11use serde_json::Value;
12use switchyard_protocol::{Category, InstructionBlock, Message, Role};
13
14use super::classifier_contract::{ClassifierContract, ClassifierContractConfig, validate_prompt};
15use super::llm_judge::{
16    ClassifierInput, JudgeClassifier, JudgePolicy, JudgeRuntimeConfig, SerdeDecoder,
17    StructuredJudge,
18};
19use super::transcript::{collect_text, message_text};
20use crate::core::algorithm::Driver;
21use crate::core::classifier::{Classification, Score};
22use crate::core::state::State;
23use crate::{LibsyError, Result};
24use switchyard_protocol::Request;
25
26const PROMPT_TEMPLATE: &str = include_str!("../../prompts/escalation/prompt.md");
27const DEESCALATION_PROMPT: &str = include_str!("../../prompts/escalation/deescalation.md");
28const SCHEMA_TEMPLATE: &str = include_str!("../../prompts/escalation/schema.json");
29
30/// Separator marking where [`truncate_middle`] dropped a message's interior.
31const TRIM_MARKER: &str = " ...[trimmed] ";
32
33/// Suffix marking a transcript cut off by [`MAX_REQUEST_CHARS`].
34const TRUNCATION_SUFFIX: &str = "...<truncated>";
35
36/// Per-message cap for system and developer anchors, which carry no trajectory signal but
37/// which coding-agent harnesses make very large.
38const SYSTEM_CHARS: usize = 1_000;
39
40/// Per-message cap for task-framing user messages — every user message that precedes the first
41/// assistant reply. Coding-agent harnesses often send environment boilerplate as the first user
42/// message and the task itself as the second, so anchoring only the first would pin the
43/// boilerplate and let the task scroll out of the window. Feature specifications run to several
44/// thousand characters, so this gets the widest anchor budget.
45const TASK_CHARS: usize = 4_000;
46
47/// Backstop on the assembled transcript; the per-message caps normally bind first.
48const MAX_REQUEST_CHARS: usize = 18_000;
49
50/// Optional policy for returning an escalated session to the efficient tier.
51#[derive(Clone, Copy, Debug, Deserialize)]
52#[serde(deny_unknown_fields)]
53pub struct DeescalationConfig {
54    /// Minimum number of capable-tier turns before the judge may release the session.
55    pub strong_min_calls: u32,
56    /// Optional hard limit on capable-tier turns before a forced return.
57    #[serde(default)]
58    pub strong_max_calls: Option<u32>,
59    /// Consecutive judge declines required to return to the efficient tier.
60    pub confirmations: u32,
61    /// Efficient calls served without judging after a hard-limit return.
62    #[serde(default)]
63    pub weak_cooldown_calls: u32,
64}
65
66impl DeescalationConfig {
67    fn validate(&self) -> Result<()> {
68        let reject = |message: &str| {
69            Err(LibsyError::AlgorithmError {
70                message: message.to_string(),
71            })
72        };
73        if self.strong_min_calls == 0 {
74            return reject("deescalation.strong_min_calls must be at least 1");
75        }
76        if self.confirmations == 0 {
77            return reject("deescalation.confirmations must be at least 1");
78        }
79        if self
80            .strong_max_calls
81            .is_some_and(|strong_max_calls| strong_max_calls < self.strong_min_calls)
82        {
83            return reject(
84                "deescalation.strong_max_calls must be at least deescalation.strong_min_calls",
85            );
86        }
87        Ok(())
88    }
89}
90
91/// The tuning surface for the trajectory judge.
92///
93/// The routing settings retain their benchmarked defaults. Everything else is a fixed invariant
94/// (the constants above).
95#[derive(Clone, Debug, Deserialize)]
96#[serde(default, deny_unknown_fields)]
97pub struct EscalationJudgeConfig {
98    /// Consecutive fresh-evidence verdicts in the same category required before a turn moves to
99    /// the capable tier, which is also the turn that latches the session. Any decline or stale
100    /// evidence clears the streak.
101    /// `1` escalates on the first verdict; the router's main cost dial.
102    /// `2` or higher needs a session id, since the streak is retained per session.
103    pub confirmations: u32,
104    /// Trailing messages shown on top of the anchors. A loop longer than this is invisible.
105    pub recent_turn_window: usize,
106    /// Per-message cap inside the trailing window.
107    pub window_message_chars: usize,
108    /// De-escalation policy. `None` preserves permanent latching to the capable tier.
109    pub deescalation: Option<DeescalationConfig>,
110}
111
112impl EscalationJudgeConfig {
113    /// Rejects settings that would leave the judge with nothing useful to read.
114    fn validate(&self) -> Result<()> {
115        let reject = |message: String| Err(LibsyError::AlgorithmError { message });
116        if self.confirmations == 0 {
117            return reject("confirmations must be at least 1".to_string());
118        }
119        if self.recent_turn_window == 0 {
120            return reject("recent_turn_window must be at least 1".to_string());
121        }
122        if self.window_message_chars < 50 {
123            return reject(format!(
124                "window_message_chars must be at least 50, got {}",
125                self.window_message_chars
126            ));
127        }
128        if let Some(deescalation) = self.deescalation {
129            deescalation.validate()?;
130        }
131        Ok(())
132    }
133}
134
135impl Default for EscalationJudgeConfig {
136    fn default() -> Self {
137        Self {
138            confirmations: 2,
139            recent_turn_window: 28,
140            window_message_chars: 500,
141            deescalation: None,
142        }
143    }
144}
145
146/// Router-controlled phase attached to judge input when de-escalation is enabled.
147#[derive(Clone, Copy, Debug, Eq, PartialEq)]
148pub(crate) enum EvaluationPhase {
149    Efficient,
150    Strong,
151}
152
153impl EvaluationPhase {
154    fn marker(self) -> &'static str {
155        match self {
156            Self::Efficient => "EFFICIENT_EVALUATION",
157            Self::Strong => "STRONG_EVALUATION",
158        }
159    }
160}
161
162/// Bounded trouble pattern used to correlate escalation confirmations across turns.
163#[derive(Clone, Copy, Debug, Deserialize, Eq, PartialEq)]
164#[serde(rename_all = "snake_case")]
165pub(crate) enum EscalationCategory {
166    None,
167    Repetition,
168    FalseProgress,
169    Drift,
170    Desperation,
171    CapabilityGap,
172}
173
174impl EscalationCategory {
175    /// Stable state and telemetry label.
176    pub(crate) const fn label(self) -> &'static str {
177        match self {
178            Self::None => "none",
179            Self::Repetition => "repetition",
180            Self::FalseProgress => "false_progress",
181            Self::Drift => "drift",
182            Self::Desperation => "desperation",
183            Self::CapabilityGap => "capability_gap",
184        }
185    }
186}
187
188/// The judge's typed verdict, including the evidence needed to confirm a stable pattern.
189#[derive(Deserialize)]
190pub(crate) struct EscalationVerdict {
191    pub(crate) escalate: bool,
192    pub(crate) category: EscalationCategory,
193    pub(crate) new_evidence: bool,
194    pub(crate) reason: String,
195}
196
197/// Builds the condensed trajectory presented to the escalation judge.
198pub(crate) struct EscalationInput {
199    config: EscalationJudgeConfig,
200    phase: Option<EvaluationPhase>,
201}
202
203impl ClassifierInput for EscalationInput {
204    fn build_messages(&self, _state: &State, request: &Request) -> Vec<Message> {
205        let summary = summarize_for_judge(
206            &request.llm_request.instructions,
207            &request.llm_request.messages,
208            conversation_turn(request),
209            self.phase,
210            &self.config,
211        );
212        vec![Message::text(Role::User, summary)]
213    }
214}
215
216/// Structured trajectory judge with a typed escalation verdict.
217pub(crate) type EscalationJudge = StructuredJudge<EscalationInput, SerdeDecoder<EscalationVerdict>>;
218
219/// Maps the judge's verdict to a classification. A verdict names the tier to serve — capable
220/// on escalate, efficient on decline — so the caller reads it straight off the winning score.
221/// [`Classification::Ambiguous`] carries the unavailable case, which names no tier: both a
222/// decline and an outage stay efficient, but only a decline is evidence, so only a decline
223/// clears the streak.
224pub(crate) struct EscalationPolicy {
225    phase: Option<EvaluationPhase>,
226}
227
228impl JudgePolicy for EscalationPolicy {
229    type Verdict = EscalationVerdict;
230
231    fn to_classification(
232        &self,
233        verdict: Option<&EscalationVerdict>,
234        driver: &Driver,
235    ) -> Result<Classification> {
236        if let Some(verdict) = verdict {
237            tracing::debug!(
238                escalate = verdict.escalate,
239                reason = %verdict.reason,
240                "escalation judge verdict"
241            );
242        }
243        match verdict {
244            Some(verdict) if verdict.escalate => Ok(Classification::Scores(vec![Score {
245                target: driver.first_model_for(&Category::Capable)?.clone(),
246                confidence: 1.0,
247                category: Some(Category::Capable),
248            }])),
249            Some(_) => Ok(Classification::Scores(vec![Score {
250                target: driver.first_model_for(&Category::Efficient)?.clone(),
251                confidence: 1.0,
252                category: Some(Category::Efficient),
253            }])),
254            None => Ok(Classification::Ambiguous(Vec::new())),
255        }
256    }
257}
258
259/// Maps present verdicts to phase-aware evidence; absent verdicts add nothing.
260fn escalation_evidence(
261    policy: &EscalationPolicy,
262    verdict: Option<&EscalationVerdict>,
263) -> Option<Value> {
264    verdict.map(|verdict| {
265        let verdict = match (policy.phase, verdict.escalate) {
266            (Some(EvaluationPhase::Strong), true) => "retain",
267            (Some(EvaluationPhase::Strong), false) => "deescalate",
268            (_, true) => "escalate",
269            (_, false) => "continue",
270        };
271        serde_json::json!({
272            "source": "escalation",
273            "verdict": verdict,
274        })
275    })
276}
277
278/// Builds the trajectory judge, scoring the runtime capable category when it escalates.
279///
280/// Loads the packaged prompt and schema, so an unusable asset or an unusable `config` value
281/// fails here rather than on the first request.
282pub(crate) fn build_judge(
283    contract_config: &ClassifierContractConfig,
284    config: EscalationJudgeConfig,
285    phase: Option<EvaluationPhase>,
286    max_output_tokens: u64,
287    judge_deadline_ms: Option<u64>,
288) -> Result<JudgeClassifier<EscalationJudge, EscalationPolicy>> {
289    config.validate()?;
290    let contract = build_contract(contract_config, phase.is_some())?;
291    Ok(JudgeClassifier::new(
292        StructuredJudge::new(
293            EscalationInput { config, phase },
294            contract,
295            SerdeDecoder::new(),
296            JudgeRuntimeConfig::new(max_output_tokens)?.with_deadline_ms(judge_deadline_ms)?,
297        ),
298        EscalationPolicy { phase },
299    )
300    .with_evidence(escalation_evidence))
301}
302
303fn build_contract(
304    contract_config: &ClassifierContractConfig,
305    phase_aware: bool,
306) -> Result<ClassifierContract> {
307    let prompt = contract_config.prompt().unwrap_or(PROMPT_TEMPLATE);
308    validate_prompt(prompt)?;
309    let phase_aware_config = phase_aware.then(|| {
310        contract_config.clone().with_prompt(format!(
311            "{}\n\n{}",
312            prompt.trim_end(),
313            DEESCALATION_PROMPT.trim()
314        ))
315    });
316    let contract_config = phase_aware_config.as_ref().unwrap_or(contract_config);
317    ClassifierContract::from_config(contract_config, PROMPT_TEMPLATE, SCHEMA_TEMPLATE)
318}
319
320/// The 1-indexed model invocation the transcript ends on: one per assistant reply.
321///
322/// The judge reads the turn *including* the reply it is judging, so the newest assistant
323/// message is this turn — no `+ 1`. Counting the caller's request instead would report the
324/// turn after the one under judgement.
325///
326/// Messages are already normalized by `switchyard-protocol`, so this needs no
327/// per-format branching.
328pub(crate) fn conversation_turn(request: &Request) -> usize {
329    request
330        .llm_request
331        .messages
332        .iter()
333        .filter(|message| message.role == Role::Assistant)
334        .count()
335}
336
337/// Keeps the head and tail of `text` within `limit` characters.
338///
339/// The head gets two thirds of the surviving budget: for a trajectory judge the command or
340/// error signature that opens a message carries more signal than its trailing output.
341fn truncate_middle(text: &str, limit: usize) -> String {
342    let chars: Vec<char> = text.chars().collect();
343    if chars.len() <= limit {
344        return text.to_string();
345    }
346    let keep = limit
347        .saturating_sub(TRIM_MARKER.chars().count())
348        .max(20)
349        .min(chars.len());
350    let head = keep * 2 / 3;
351    let tail = keep - head;
352    let mut out: String = chars[..head].iter().collect();
353    out.push_str(TRIM_MARKER);
354    out.extend(chars[chars.len() - tail..].iter());
355    out
356}
357
358/// Renders a compact role-labelled transcript for the judge.
359///
360/// Task-framing user messages are capped individually. System/developer instructions share
361/// the remaining budget after reserving space for task framing and the newest window entry.
362/// The trailing window carries recent activity. A coverage header states how much history is
363/// not shown, so the judge can reason about pace rather than assuming it sees everything.
364///
365/// When the assembled text still exceeds `max_request_chars`, the oldest window lines go
366/// first: for a trajectory judge the newest evidence is strictly the most valuable.
367fn summarize_for_judge(
368    instructions: &[InstructionBlock],
369    messages: &[Message],
370    turn: usize,
371    phase: Option<EvaluationPhase>,
372    config: &EscalationJudgeConfig,
373) -> String {
374    let mut instruction_anchors: Vec<String> = Vec::new();
375    let mut anchors: Vec<String> = Vec::new();
376    let mut window: Vec<String> = Vec::new();
377    let mut assistant_seen = false;
378
379    for instruction in instructions {
380        let mut parts = Vec::new();
381        collect_text(&instruction.content, &mut parts, &[]);
382        instruction_anchors.push(format!(
383            "[{}] {}",
384            role_label(instruction.role),
385            truncate_middle(&parts.join(" "), SYSTEM_CHARS)
386        ));
387    }
388
389    for message in messages {
390        let text = message_text(message);
391        match message.role {
392            Role::System | Role::Developer => instruction_anchors.push(format!(
393                "[{}] {}",
394                role_label(message.role),
395                truncate_middle(&text, SYSTEM_CHARS)
396            )),
397            // Everything the user said before the agent first replied is task framing.
398            Role::User if !assistant_seen => {
399                anchors.push(format!(
400                    "[user (task)] {}",
401                    truncate_middle(&text, TASK_CHARS)
402                ));
403            }
404            role => {
405                if role == Role::Assistant {
406                    assistant_seen = true;
407                }
408                window.push(format!(
409                    "[{}] {}",
410                    role_label(role),
411                    truncate_middle(&text, config.window_message_chars)
412                ));
413            }
414        }
415    }
416
417    if window.len() > config.recent_turn_window {
418        window.drain(..window.len() - config.recent_turn_window);
419    }
420
421    let assemble = |instructions: Option<&str>, window: &[String]| {
422        let header = format!(
423            "Conversation turn {turn}; showing the last {} of {} messages after the task framing.",
424            window.len(),
425            messages.len(),
426        );
427        phase
428            .map(|phase| format!("Routing phase: {}", phase.marker()))
429            .into_iter()
430            .chain(std::iter::once(header))
431            .chain(instructions.map(str::to_owned))
432            .chain(anchors.iter().cloned())
433            .chain(window.iter().cloned())
434            .collect::<Vec<_>>()
435            .join("\n")
436    };
437
438    let reserved = assemble(None, &window[window.len().saturating_sub(1)..])
439        .chars()
440        .count();
441    let instruction_budget = MAX_REQUEST_CHARS.saturating_sub(reserved + 1);
442    // The remaining budget may be smaller than truncate_middle's minimum retained span.
443    let instruction_text = truncate_middle(&instruction_anchors.join("\n"), instruction_budget)
444        .chars()
445        .take(instruction_budget)
446        .collect::<String>();
447    let instructions = (!instruction_text.is_empty()).then_some(instruction_text.as_str());
448    let mut text = assemble(instructions, &window);
449    while text.chars().count() > MAX_REQUEST_CHARS && !window.is_empty() {
450        window.remove(0);
451        text = assemble(instructions, &window);
452    }
453    if text.chars().count() > MAX_REQUEST_CHARS {
454        let keep = MAX_REQUEST_CHARS.saturating_sub(TRUNCATION_SUFFIX.chars().count() + 1);
455        text = text.chars().take(keep).collect::<String>() + TRUNCATION_SUFFIX;
456    }
457    text
458}
459
460/// The transcript label for a role.
461fn role_label(role: Role) -> &'static str {
462    match role {
463        Role::System => "system",
464        Role::Developer => "developer",
465        Role::User => "user",
466        Role::Assistant => "assistant",
467        Role::Tool => "tool",
468    }
469}
470
471/// A request whose conversation sits at `turn`: `turn - 1` prior assistant replies, each
472/// answered by a further user message.
473///
474/// Shared with the assembled router's tests, which drive the same conversation shape.
475#[cfg(test)]
476pub(crate) fn request_at_turn(session_id: Option<&str>, turn: usize) -> Request {
477    use switchyard_protocol::{LlmRequest, Metadata};
478
479    let mut messages = vec![Message::text(Role::User, "What is 2+2?")];
480    for attempt in 1..turn {
481        messages.push(Message::text(Role::Assistant, format!("attempt {attempt}")));
482        messages.push(Message::text(Role::User, format!("still wrong {attempt}")));
483    }
484    Request {
485        llm_request: LlmRequest {
486            model: Some("auto".to_string()),
487            messages,
488            ..LlmRequest::default()
489        },
490        raw_request: None,
491        metadata: session_id.map(|id| Metadata {
492            session_id: Some(id.to_string()),
493            ..Metadata::default()
494        }),
495    }
496}
497
498#[cfg(test)]
499mod tests {
500    use switchyard_protocol::{ContentBlock, Message, Role};
501
502    use super::*;
503    use crate::algorithms::util::llm_judge::Judge;
504
505    fn escalation_judge(
506        max_output_tokens: u64,
507        phase: Option<EvaluationPhase>,
508        contract_config: &ClassifierContractConfig,
509    ) -> Result<EscalationJudge> {
510        Ok(StructuredJudge::new(
511            EscalationInput {
512                config: EscalationJudgeConfig::default(),
513                phase,
514            },
515            build_contract(contract_config, phase.is_some())?,
516            SerdeDecoder::new(),
517            JudgeRuntimeConfig::new(max_output_tokens)?,
518        ))
519    }
520
521    #[test]
522    fn judge_request_is_rubric_plus_summary_under_a_completion_cap() -> Result<()> {
523        let judge = escalation_judge(
524            super::super::DEFAULT_JUDGE_MAX_OUTPUT_TOKENS,
525            None,
526            &ClassifierContractConfig::default(),
527        )?;
528
529        // As the classifier calls it: the turn's reply is already on the transcript.
530        let mut judged = request_at_turn(None, 4);
531        judged.llm_request.instructions = [
532            (Role::System, "system constraint"),
533            (Role::Developer, "developer constraint"),
534        ]
535        .into_iter()
536        .map(|(role, text)| InstructionBlock {
537            role,
538            content: Message::text(role, text).content,
539        })
540        .collect();
541        judged
542            .llm_request
543            .messages
544            .push(Message::text(Role::Assistant, "this turn's reply"));
545        let built = judge.build_request(&State::default(), &judged);
546
547        // Rubric in instructions, condensed trajectory as the sole user message.
548        assert_eq!(built.llm_request.instructions.len(), 1);
549        assert_eq!(built.llm_request.instructions[0].role, Role::System);
550        assert_eq!(
551            built.llm_request.instructions[0].content.as_slice(),
552            &[ContentBlock::Text {
553                text: PROMPT_TEMPLATE.to_string()
554            }]
555        );
556        assert_eq!(built.llm_request.messages.len(), 1);
557        assert_eq!(built.llm_request.messages[0].role, Role::User);
558        let summary = built.llm_request.messages[0]
559            .text_content("")
560            .expect("summary");
561        assert!(summary.contains("Conversation turn 4"));
562        assert!(summary.contains(
563            "[system] system constraint\n[developer] developer constraint\n[user (task)] What is 2+2?"
564        ));
565        assert!(summary.contains("[assistant] this turn's reply"));
566        assert!(!summary.contains("Routing phase:"));
567        // Bounded output, so a reasoning judge cannot run away mid-verdict.
568        assert_eq!(
569            built.llm_request.output.max_output_tokens,
570            Some(super::super::DEFAULT_JUDGE_MAX_OUTPUT_TOKENS)
571        );
572        assert!(built.llm_request.output.response_format.is_some());
573
574        judged.llm_request.instructions.extend(vec![
575            InstructionBlock {
576                role: Role::Developer,
577                content: Message::text(Role::Developer, "x".repeat(SYSTEM_CHARS)).content,
578            };
579            MAX_REQUEST_CHARS / SYSTEM_CHARS + 1
580        ]);
581        let built = judge.build_request(&State::default(), &judged);
582        let summary = built.llm_request.messages[0]
583            .text_content("")
584            .expect("summary");
585        assert!(summary.chars().count() <= MAX_REQUEST_CHARS);
586        assert!(summary.contains(TRIM_MARKER));
587        assert!(summary.contains("[system] system constraint"));
588        assert!(summary.contains("[developer] developer constraint"));
589        assert!(summary.contains("[user (task)] What is 2+2?"));
590        assert!(summary.contains("[assistant] this turn's reply"));
591        Ok(())
592    }
593
594    #[test]
595    fn judge_request_uses_the_configured_completion_cap() -> Result<()> {
596        let judge = escalation_judge(512, None, &ClassifierContractConfig::default())?;
597
598        let built = judge.build_request(&State::default(), &request_at_turn(None, 1));
599
600        assert_eq!(built.llm_request.output.max_output_tokens, Some(512));
601        Ok(())
602    }
603
604    #[test]
605    fn conversation_turn_counts_assistant_replies() {
606        // The judge is handed the transcript with this turn's reply already appended, which
607        // is the shape asserted here: a request entering turn N plus its reply *is* turn N.
608        for turn in [1, 5] {
609            let mut judged = request_at_turn(None, turn);
610            judged
611                .llm_request
612                .messages
613                .push(Message::text(Role::Assistant, "this turn's reply"));
614            assert_eq!(conversation_turn(&judged), turn);
615        }
616    }
617
618    #[test]
619    fn truncate_middle_keeps_head_and_tail() {
620        let text = "a".repeat(40) + &"z".repeat(40);
621        let trimmed = truncate_middle(&text, 50);
622        assert!(trimmed.chars().count() <= 50, "{trimmed}");
623        assert!(trimmed.starts_with('a'));
624        assert!(trimmed.ends_with('z'));
625        assert!(trimmed.contains("[trimmed]"));
626
627        // Under the limit the text is returned untouched.
628        assert_eq!(truncate_middle("short", 50), "short");
629    }
630
631    #[test]
632    fn deescalation_settings_must_be_valid() {
633        let zero = EscalationJudgeConfig {
634            deescalation: Some(DeescalationConfig {
635                strong_min_calls: 0,
636                strong_max_calls: None,
637                confirmations: 2,
638                weak_cooldown_calls: 0,
639            }),
640            ..EscalationJudgeConfig::default()
641        };
642        assert!(
643            zero.validate()
644                .is_err_and(|error| error.to_string().contains("at least 1"))
645        );
646
647        let inverted = EscalationJudgeConfig {
648            deescalation: Some(DeescalationConfig {
649                strong_min_calls: 4,
650                strong_max_calls: Some(3),
651                confirmations: 2,
652                weak_cooldown_calls: 0,
653            }),
654            ..EscalationJudgeConfig::default()
655        };
656        assert!(
657            inverted
658                .validate()
659                .is_err_and(|error| error.to_string().contains("at least deescalation"))
660        );
661    }
662
663    #[test]
664    fn deescalation_contract_marks_both_routing_phases() -> Result<()> {
665        let contract = ClassifierContractConfig::default().with_prompt("Custom trajectory rubric.");
666        for (phase, marker) in [
667            (EvaluationPhase::Efficient, "EFFICIENT_EVALUATION"),
668            (EvaluationPhase::Strong, "STRONG_EVALUATION"),
669        ] {
670            let judge = escalation_judge(512, Some(phase), &contract)?;
671            let built = judge.build_request(&State::default(), &request_at_turn(None, 1));
672            let system_prompt = built.llm_request.instructions[0].content.iter().find_map(
673                |content| match content {
674                    ContentBlock::Text { text } => Some(text.as_str()),
675                    _ => None,
676                },
677            );
678            assert!(system_prompt.is_some_and(|prompt| {
679                prompt.starts_with("Custom trajectory rubric.")
680                    && prompt.contains("EFFICIENT_EVALUATION")
681                    && prompt.contains("STRONG_EVALUATION")
682            }));
683            assert!(
684                built.llm_request.messages[0]
685                    .text_content("")
686                    .is_some_and(|summary| summary.starts_with(&format!(
687                        "Routing phase: {marker}"
688                    )))
689            );
690        }
691        Ok(())
692    }
693
694    #[test]
695    fn summary_keeps_anchors_and_the_recent_window() {
696        let mut messages = vec![
697            Message::text(Role::System, "you are a coding agent"),
698            Message::text(Role::User, "fix the failing test"),
699        ];
700        for i in 0..10 {
701            messages.push(Message::text(Role::Assistant, format!("step {i}")));
702        }
703        let config = EscalationJudgeConfig {
704            recent_turn_window: 3,
705            ..EscalationJudgeConfig::default()
706        };
707
708        let summary = summarize_for_judge(&[], &messages, 11, None, &config);
709
710        assert!(
711            summary.contains("[system] you are a coding agent"),
712            "{summary}"
713        );
714        assert!(
715            summary.contains("[user (task)] fix the failing test"),
716            "{summary}"
717        );
718        assert!(summary.contains("Conversation turn 11; showing the last 3 of 12 messages"));
719        // Only the newest window entries survive.
720        assert!(summary.contains("step 9"), "{summary}");
721        assert!(summary.contains("step 7"), "{summary}");
722        assert!(!summary.contains("step 6"), "{summary}");
723    }
724
725    #[test]
726    fn summary_anchors_every_user_message_before_the_first_reply() {
727        // Codex sends environment boilerplate as the first user message and the task as the
728        // second. Both are framing; the task must stay visible after the window has moved on.
729        let mut messages = vec![
730            Message::text(
731                Role::Developer,
732                "<skills_instructions>...</skills_instructions>",
733            ),
734            Message::text(
735                Role::User,
736                "<environment_context><cwd>/app</cwd></environment_context>",
737            ),
738            Message::text(Role::User, "Implement RFC 5545 timezone interop in rrule."),
739        ];
740        for i in 0..40 {
741            messages.push(Message::text(Role::Assistant, format!("step {i}")));
742            messages.push(Message::text(Role::User, format!("later user note {i}")));
743        }
744        let config = EscalationJudgeConfig {
745            recent_turn_window: 3,
746            ..EscalationJudgeConfig::default()
747        };
748
749        let summary = summarize_for_judge(&[], &messages, 40, None, &config);
750
751        assert!(
752            summary.contains("[user (task)] <environment_context>"),
753            "{summary}"
754        );
755        assert!(
756            summary.contains("[user (task)] Implement RFC 5545 timezone interop in rrule."),
757            "{summary}"
758        );
759        // User messages after the first reply are ordinary window entries, not anchors.
760        assert!(
761            !summary.contains("[user (task)] later user note"),
762            "{summary}"
763        );
764        assert!(summary.contains("[user] later user note 39"), "{summary}");
765        assert!(!summary.contains("later user note 0\n"), "{summary}");
766    }
767
768    #[test]
769    fn summary_drops_oldest_window_lines_under_the_char_cap() {
770        // MAX_REQUEST_CHARS is a backstop, not a dial: at default settings the window caps
771        // bind first (28 x 500 plus anchors sits under it), so reaching it takes an unusually
772        // wide per-message cap. That is the point — it only fires on pathological input.
773        let mut messages = vec![
774            Message::text(Role::System, "framing"),
775            Message::text(Role::User, "task"),
776        ];
777        for i in 0..20 {
778            messages.push(Message::text(
779                Role::Assistant,
780                format!("{i} {}", "x".repeat(2_000)),
781            ));
782        }
783        let config = EscalationJudgeConfig {
784            window_message_chars: 2_000,
785            ..EscalationJudgeConfig::default()
786        };
787
788        let summary = summarize_for_judge(&[], &messages, 21, None, &config);
789
790        assert!(
791            summary.chars().count() <= MAX_REQUEST_CHARS,
792            "{}",
793            summary.chars().count()
794        );
795        // Anchors are never dropped, and the newest activity outlives the oldest.
796        assert!(summary.contains("[system] framing"), "{summary}");
797        assert!(summary.contains("[user (task)] task"), "{summary}");
798        assert!(summary.contains("19 xxx"), "{summary}");
799        assert!(!summary.contains("0 xxx"), "{summary}");
800    }
801}
switchyard_libsy/algorithms/util/escalation.rs

switchyard_libsy/algorithms/util/
escalation.rs
