lean_ctx/core/stats/
model.rs

1use serde::{Deserialize, Serialize};
2use std::collections::HashMap;
3
4/// Persistent store for all-time token savings, command stats, and daily history.
5#[derive(Serialize, Deserialize, Default, Clone)]
6pub struct StatsStore {
7    pub total_commands: u64,
8    pub total_input_tokens: u64,
9    pub total_output_tokens: u64,
10    pub first_use: Option<String>,
11    pub last_use: Option<String>,
12    pub commands: HashMap<String, CommandStats>,
13    pub daily: Vec<DayStats>,
14    #[serde(default)]
15    pub cep: CepStats,
16}
17
18/// Aggregated CEP (Cognitive Efficiency Protocol) metrics across sessions.
19#[derive(Serialize, Deserialize, Clone, Default)]
20pub struct CepStats {
21    pub sessions: u64,
22    pub total_cache_hits: u64,
23    pub total_cache_reads: u64,
24    pub total_tokens_original: u64,
25    pub total_tokens_compressed: u64,
26    pub modes: HashMap<String, u64>,
27    pub scores: Vec<CepSessionSnapshot>,
28    #[serde(default)]
29    pub last_session_pid: Option<u32>,
30    #[serde(default)]
31    pub last_session_original: Option<u64>,
32    #[serde(default)]
33    pub last_session_compressed: Option<u64>,
34}
35
36/// Point-in-time snapshot of CEP scores for a single session.
37#[derive(Serialize, Deserialize, Clone)]
38pub struct CepSessionSnapshot {
39    pub timestamp: String,
40    pub score: u32,
41    pub cache_hit_rate: u32,
42    pub mode_diversity: u32,
43    pub compression_rate: u32,
44    pub tool_calls: u64,
45    pub tokens_saved: u64,
46    pub complexity: String,
47}
48
49/// Per-command token statistics: invocation count and input/output totals.
50#[derive(Serialize, Deserialize, Clone, Default, Debug)]
51pub struct CommandStats {
52    pub count: u64,
53    pub input_tokens: u64,
54    pub output_tokens: u64,
55}
56
57/// Daily aggregate: command count and token totals for one calendar day.
58#[derive(Serialize, Deserialize, Clone, Default)]
59pub struct DayStats {
60    pub date: String,
61    pub commands: u64,
62    pub input_tokens: u64,
63    pub output_tokens: u64,
64    /// lean-ctx version active when this day's stats were last recorded.
65    /// Lets `lean-ctx gain` attribute per-day compression changes to a release
66    /// (#307). Empty for days recorded before this field existed.
67    #[serde(default)]
68    pub version: String,
69}
70
71/// High-level token savings summary for display.
72pub struct GainSummary {
73    pub total_saved: u64,
74    pub total_calls: u64,
75}
76
77/// Average LLM pricing per 1M tokens (blended across Claude, GPT, Gemini).
78pub const DEFAULT_INPUT_PRICE_PER_M: f64 = 2.50;
79pub const DEFAULT_OUTPUT_PRICE_PER_M: f64 = 10.0;
80
81/// LLM pricing model for estimating dollar savings from token compression.
82pub struct CostModel {
83    pub input_price_per_m: f64,
84    pub output_price_per_m: f64,
85    pub avg_verbose_output_per_call: u64,
86    pub avg_concise_output_per_call: u64,
87}
88
89impl Default for CostModel {
90    fn default() -> Self {
91        let env_model = std::env::var("LEAN_CTX_MODEL")
92            .or_else(|_| std::env::var("LCTX_MODEL"))
93            .ok();
94        let pricing = crate::core::gain::model_pricing::ModelPricing::load();
95        let quote = pricing.quote(env_model.as_deref());
96        Self {
97            input_price_per_m: quote.cost.input_per_m,
98            output_price_per_m: quote.cost.output_per_m,
99            avg_verbose_output_per_call: 180,
100            avg_concise_output_per_call: 120,
101        }
102    }
103}
104
105/// Detailed cost comparison: with vs. without lean-ctx compression.
106pub struct CostBreakdown {
107    pub input_cost_without: f64,
108    pub input_cost_with: f64,
109    pub output_cost_without: f64,
110    pub output_cost_with: f64,
111    pub total_cost_without: f64,
112    pub total_cost_with: f64,
113    pub total_saved: f64,
114    pub estimated_output_tokens_without: u64,
115    pub estimated_output_tokens_with: u64,
116    pub output_tokens_saved: u64,
117}
118
119impl CostModel {
120    /// Calculates the full cost breakdown from the stats store.
121    pub fn calculate(&self, store: &StatsStore) -> CostBreakdown {
122        let input_cost_without =
123            store.total_input_tokens as f64 / 1_000_000.0 * self.input_price_per_m;
124        let input_cost_with =
125            store.total_output_tokens as f64 / 1_000_000.0 * self.input_price_per_m;
126
127        let input_saved = store
128            .total_input_tokens
129            .saturating_sub(store.total_output_tokens);
130        let compression_rate = if store.total_input_tokens > 0 {
131            input_saved as f64 / store.total_input_tokens as f64
132        } else {
133            0.0
134        };
135        let est_output_without = store.total_commands * self.avg_verbose_output_per_call;
136        let est_output_with = if compression_rate > 0.01 {
137            store.total_commands * self.avg_concise_output_per_call
138        } else {
139            est_output_without
140        };
141        let output_saved = est_output_without.saturating_sub(est_output_with);
142
143        let output_cost_without = est_output_without as f64 / 1_000_000.0 * self.output_price_per_m;
144        let output_cost_with = est_output_with as f64 / 1_000_000.0 * self.output_price_per_m;
145
146        let total_without = input_cost_without + output_cost_without;
147        let total_with = input_cost_with + output_cost_with;
148
149        CostBreakdown {
150            input_cost_without,
151            input_cost_with,
152            output_cost_without,
153            output_cost_with,
154            total_cost_without: total_without,
155            total_cost_with: total_with,
156            total_saved: total_without - total_with,
157            estimated_output_tokens_without: est_output_without,
158            estimated_output_tokens_with: est_output_with,
159            output_tokens_saved: output_saved,
160        }
161    }
162}
lean_ctx/core/stats/model.rs

lean_ctx/core/stats/
model.rs
