poimen-memory/crates/mem-cli/src/handlers/ranking_handler.rs

use actix_web::{web, HttpRequest, HttpResponse};
use chrono::{DateTime, Utc};
use serde_json::json;
use sqlx::PgPool;
use std::collections::HashMap;

use crate::auth::AuthGuard;

/// Ranking profile with configurable signal weights
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct RankingProfile {
    pub name: String,
    pub description: String,
    pub weights: RankingWeights,
}

#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct RankingWeights {
    pub semantic: f64,
    pub lexical: f64,
    pub recency: f64,
    pub frequency: f64,
    pub confidence: f64,
    pub community: f64,
    pub contradiction: f64,
}

impl RankingWeights {
    /// Default balanced profile
    pub fn default() -> Self {
        Self {
            semantic: 0.40,
            lexical: 0.25,
            recency: 0.15,
            frequency: 0.10,
            confidence: 0.05,
            community: 0.03,
            contradiction: 0.02,
        }
    }

    /// Recency-focused profile
    pub fn recency_focused() -> Self {
        Self {
            semantic: 0.30,
            lexical: 0.15,
            recency: 0.35,
            frequency: 0.10,
            confidence: 0.05,
            community: 0.03,
            contradiction: 0.02,
        }
    }

    /// Accuracy-focused profile (high confidence, no contradictions)
    pub fn accuracy_focused() -> Self {
        Self {
            semantic: 0.35,
            lexical: 0.20,
            recency: 0.10,
            frequency: 0.05,
            confidence: 0.20,
            community: 0.05,
            contradiction: 0.05,
        }
    }

    /// Normalize weights to sum to 1.0
    pub fn normalize(&mut self) {
        let sum = self.semantic
            + self.lexical
            + self.recency
            + self.frequency
            + self.confidence
            + self.community
            + self.contradiction.abs();

        if sum > 0.0 {
            self.semantic /= sum;
            self.lexical /= sum;
            self.recency /= sum;
            self.frequency /= sum;
            self.confidence /= sum;
            self.community /= sum;
            self.contradiction /= sum;
        }
    }
}

/// Ranking signals for an entity
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
pub struct RankingSignals {
    pub semantic: f64,
    pub lexical: f64,
    pub recency: f64,
    pub frequency: f64,
    pub confidence: f64,
    pub community: f64,
    pub contradiction: f64,
}

impl RankingSignals {
    /// Compute recency signal from last update time
    pub fn compute_recency(updated_at: DateTime<Utc>) -> f64 {
        let age_days = (Utc::now() - updated_at).num_days() as f64;
        // Exponential decay with 30-day half-life
        (-age_days / 30.0).exp()
    }

    /// Compute frequency signal from access count
    pub fn compute_frequency(access_count: u64, max_access: u64) -> f64 {
        if max_access == 0 {
            return 0.0;
        }
        (access_count as f64 + 1.0).ln() / (max_access as f64 + 1.0).ln()
    }

    /// Compute confidence with decay (no recent confirmation = lower)
    pub fn compute_confidence(base_confidence: f64, last_confirmed: DateTime<Utc>) -> f64 {
        let staleness = (Utc::now() - last_confirmed).num_days() as f64;
        let decay = (-staleness / 90.0).exp(); // 90-day half-life
        base_confidence * decay
    }

    /// Compute community activity score
    pub fn compute_community(community_size: usize, recent_edges: usize) -> f64 {
        let size_factor = (community_size as f64).min(100.0) / 100.0;
        let activity_factor = (recent_edges as f64).min(50.0) / 50.0;
        (size_factor + activity_factor) / 2.0
    }

    /// Compute contradiction penalty
    pub fn compute_contradiction(unresolved_count: usize, total_edges: usize) -> f64 {
        if total_edges == 0 {
            return 0.0;
        }
        let ratio = unresolved_count as f64 / total_edges as f64;
        // Max penalty -0.3
        -(ratio * 0.3).min(0.3)
    }
}

/// Signal breakdown with contributions
#[derive(Debug, serde::Serialize)]
pub struct SignalBreakdown {
    pub semantic: SignalDetail,
    pub lexical: SignalDetail,
    pub recency: SignalDetail,
    pub frequency: SignalDetail,
    pub confidence: SignalDetail,
    pub community: SignalDetail,
    pub contradiction: SignalDetail,
}

#[derive(Debug, serde::Serialize)]
pub struct SignalDetail {
    pub raw: f64,
    pub weight: f64,
    pub contribution: f64,
}

/// Ranked result with signal breakdown
#[derive(Debug, serde::Serialize)]
pub struct RankedResult {
    pub id: String,
    pub name: String,
    pub final_score: f64,
    pub signal_breakdown: Option<SignalBreakdown>,
}

/// GET /memory/ranking/profiles
pub async fn get_ranking_profiles(req: HttpRequest) -> HttpResponse {
    // Verify auth
    if let Err(e) = AuthGuard::extract_token(&req) {
        return HttpResponse::Unauthorized().json(json!({
            "error": e.to_string()
        }));
    }

    let profiles = vec![
        RankingProfile {
            name: "default".to_string(),
            description: "Balanced multi-signal ranking".to_string(),
            weights: RankingWeights::default(),
        },
        RankingProfile {
            name: "recency_focused".to_string(),
            description: "Prioritize recent updates".to_string(),
            weights: RankingWeights::recency_focused(),
        },
        RankingProfile {
            name: "accuracy_focused".to_string(),
            description: "Prioritize high-confidence, no contradictions".to_string(),
            weights: RankingWeights::accuracy_focused(),
        },
    ];

    HttpResponse::Ok().json(json!({
        "profiles": profiles
    }))
}

/// Helper to compute multi-signal score
pub fn compute_final_score(signals: &RankingSignals, weights: &RankingWeights) -> f64 {
    let raw = signals.semantic * weights.semantic
        + signals.lexical * weights.lexical
        + signals.recency * weights.recency
        + signals.frequency * weights.frequency
        + signals.confidence * weights.confidence
        + signals.community * weights.community
        + signals.contradiction * weights.contradiction;

    raw.clamp(0.0, 1.0)
}

/// Helper to generate signal breakdown
pub fn generate_breakdown(signals: &RankingSignals, weights: &RankingWeights) -> SignalBreakdown {
    SignalBreakdown {
        semantic: SignalDetail {
            raw: signals.semantic,
            weight: weights.semantic,
            contribution: signals.semantic * weights.semantic,
        },
        lexical: SignalDetail {
            raw: signals.lexical,
            weight: weights.lexical,
            contribution: signals.lexical * weights.lexical,
        },
        recency: SignalDetail {
            raw: signals.recency,
            weight: weights.recency,
            contribution: signals.recency * weights.recency,
        },
        frequency: SignalDetail {
            raw: signals.frequency,
            weight: weights.frequency,
            contribution: signals.frequency * weights.frequency,
        },
        confidence: SignalDetail {
            raw: signals.confidence,
            weight: weights.confidence,
            contribution: signals.confidence * weights.confidence,
        },
        community: SignalDetail {
            raw: signals.community,
            weight: weights.community,
            contribution: signals.community * weights.community,
        },
        contradiction: SignalDetail {
            raw: signals.contradiction,
            weight: weights.contradiction,
            contribution: signals.contradiction * weights.contradiction,
        },
    }
}

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

    #[test]
    fn test_recency_signal_recent() {
        let now = Utc::now();
        let score = RankingSignals::compute_recency(now);
        assert!(score > 0.99); // Just now = ~1.0
    }

    #[test]
    fn test_recency_signal_30_days() {
        let thirty_days_ago = Utc::now() - chrono::Duration::days(30);
        let score = RankingSignals::compute_recency(thirty_days_ago);
        assert!((score - 0.368).abs() < 0.01); // e^-1 ≈ 0.368
    }

    #[test]
    fn test_frequency_signal_normalized() {
        let score = RankingSignals::compute_frequency(50, 100);
        assert!(score > 0.0 && score < 1.0);
    }

    #[test]
    fn test_contradiction_penalty_capped() {
        let penalty = RankingSignals::compute_contradiction(100, 10);
        assert_eq!(penalty, -0.3); // Max penalty
    }

    #[test]
    fn test_weights_normalize() {
        let mut weights = RankingWeights {
            semantic: 2.0,
            lexical: 2.0,
            recency: 2.0,
            frequency: 2.0,
            confidence: 2.0,
            community: 0.0,
            contradiction: 0.0,
        };
        weights.normalize();
        let sum = weights.semantic + weights.lexical + weights.recency + weights.frequency
            + weights.confidence;
        assert!((sum - 1.0).abs() < 0.001);
    }

    #[test]
    fn test_compute_score_clamped() {
        let signals = RankingSignals {
            semantic: 1.0,
            lexical: 1.0,
            recency: 1.0,
            frequency: 1.0,
            confidence: 1.0,
            community: 1.0,
            contradiction: 0.0,
        };
        let weights = RankingWeights::default();
        let score = compute_final_score(&signals, &weights);
        assert!(score <= 1.0);
    }

    #[test]
    fn test_profile_presets() {
        let default = RankingWeights::default();
        let recency = RankingWeights::recency_focused();
        let accuracy = RankingWeights::accuracy_focused();

        assert!(recency.recency > default.recency);
        assert!(accuracy.confidence > default.confidence);
    }
}