diff --git a/apexchainx_calculator/src/lib.rs b/apexchainx_calculator/src/lib.rs
index e139144..dddfb2b 100644
--- a/apexchainx_calculator/src/lib.rs
+++ b/apexchainx_calculator/src/lib.rs
@@ -24,11 +24,8 @@ mod tests;
 #[cfg(test)]
 mod fuzz_tests;
 
-/// Parity checker: compares current `compute_result` against the locked-in
-/// canonical golden vectors in `test_snapshots/tests/parity_baseline.json`.
-/// Run with `cargo test --lib parity_tests::` or `just parity-check`.
 #[cfg(test)]
-mod parity_tests;
+mod prune_benchmark;
 
 pub mod audit_state;
 pub mod calculation;
diff --git a/apexchainx_calculator/src/prune_benchmark.rs b/apexchainx_calculator/src/prune_benchmark.rs
new file mode 100644
index 0000000..82587ba
--- /dev/null
+++ b/apexchainx_calculator/src/prune_benchmark.rs
@@ -0,0 +1,312 @@
+//! #204 – Reproducible benchmark artifact for history prune operations.
+//!
+//! This module provides a deterministic, reproducible benchmark suite for
+//! the `prune_history` and `prune_history_by_age` operations at 1k, 10k,
+//! and 100k history entry sizes.
+//!
+//! # Usage
+//!
+//! ```bash
+//! cargo test --package apexchainx_calculator -- prune_benchmark --nocapture --ignored
+//! ```
+//!
+//! The `--ignored` flag is required because 100k-entry tests are expensive
+//! and should only run in CI or explicitly requested local runs.
+//!
+//! # Output Artifact
+//!
+//! The test prints a machine-readable JSON benchmark artifact to stdout,
+//! suitable for CI ingestion and trend tracking.
+
+#[cfg(test)]
+mod prune_benchmark {
+    use crate::SLACalculatorContract;
+    use alloc::format;
+    use soroban_sdk::{
+        symbol_short,
+        testutils::{Address as _, Ledger},
+        Address, Env, Symbol,
+    };
+
+    /// CPU instruction budget ceiling per entry size tier.
+    /// Values are generous to allow for test infrastructure overhead
+    /// while still catching regressions.
+    const BUDGET_1K: u64 = 5_000_000;   // 5M instructions
+    const BUDGET_10K: u64 = 50_000_000;  // 50M instructions
+    const BUDGET_100K: u64 = 500_000_000; // 500M instructions
+
+    struct PruneBenchEntry {
+        size: u32,
+        prune_kept: u32,
+        cpu_instructions: u64,
+        budget: u64,
+        passed: bool,
+    }
+
+    fn run_prune_bench(env: &Env, size: u32, prune_kept: u32, budget: u64) -> PruneBenchEntry {
+        let admin = Address::generate(env);
+        let op = Address::generate(env);
+
+        let contract_id = env.register_contract(None, SLACalculatorContract);
+        let client = crate::SLACalculatorContractClient::new(env, &contract_id);
+        client.initialize(&admin, &op);
+
+        // Populate history with `size` entries
+        for i in 0..size {
+            let outage_id = Symbol::new(env, &format!("BENCH_{}", i));
+            client.calculate_sla(
+                &op,
+                &outage_id,
+                &symbol_short!("low"),
+                &(10u32 + (i % 5)),
+            );
+        }
+
+        let history_before = client.get_history();
+        assert_eq!(history_before.len(), size, "History population failed");
+
+        // Measure prune performance
+        env.budget().reset_default();
+        let before = env.budget().cpu_instruction_cost();
+        client.prune_history(&admin, &prune_kept);
+        let after = env.budget().cpu_instruction_cost();
+        let cpu_instructions = after.saturating_sub(before);
+
+        let history_after = client.get_history();
+        assert_eq!(
+            history_after.len(),
+            prune_kept,
+            "Prune did not produce expected count"
+        );
+
+        PruneBenchEntry {
+            size,
+            prune_kept,
+            cpu_instructions,
+            budget,
+            passed: cpu_instructions < budget,
+        }
+    }
+
+    fn run_prune_by_age_bench(env: &Env, size: u32, keep_ratio: f64, budget: u64) -> PruneBenchEntry {
+        let admin = Address::generate(env);
+        let op = Address::generate(env);
+
+        let contract_id = env.register_contract(None, SLACalculatorContract);
+        let client = crate::SLACalculatorContractClient::new(env, &contract_id);
+        client.initialize(&admin, &op);
+
+        // Populate history with staggered timestamps
+        let base_timestamp = 1000u64;
+        for i in 0..size {
+            let ts = base_timestamp + (i as u64 * 10);
+            env.ledger().set_timestamp(ts);
+            let outage_id = Symbol::new(env, &format!("BENCH_A_{}", i));
+            client.calculate_sla(
+                &op,
+                &outage_id,
+                &symbol_short!("low"),
+                &(10u32 + (i % 5)),
+            );
+        }
+
+        // Set timestamp far in the future so prune by age works
+        let final_ts = base_timestamp + (size as u64 * 10) + 1000;
+        env.ledger().set_timestamp(final_ts);
+
+        // Calculate age window to keep `keep_ratio` fraction of entries
+        let total_span = final_ts - base_timestamp;
+        let age_window = (total_span as f64 * keep_ratio) as u64;
+
+        let history_before = client.get_history();
+        assert_eq!(history_before.len(), size);
+
+        env.budget().reset_default();
+        let before = env.budget().cpu_instruction_cost();
+        client.prune_history_by_age(&admin, &age_window);
+        let after = env.budget().cpu_instruction_cost();
+        let cpu_instructions = after.saturating_sub(before);
+
+        let history_after = client.get_history();
+        assert!(
+            history_after.len() < size,
+            "Prune by age should have removed some entries"
+        );
+
+        PruneBenchEntry {
+            size,
+            prune_kept: history_after.len(),
+            cpu_instructions,
+            budget,
+            passed: cpu_instructions < budget,
+        }
+    }
+
+    // ================================================================
+    // Individual benchmark tests (grouped by size)
+    // ================================================================
+
+    #[test]
+    fn bench_prune_1k() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_bench(&env, 1_000, 100, BUDGET_1K);
+        assert!(result.passed, "1k prune exceeded budget: {} > {}", result.cpu_instructions, result.budget);
+        println!("  prune(1k → 100): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    #[test]
+    fn bench_prune_1k_by_age() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_by_age_bench(&env, 1_000, 0.5, BUDGET_1K);
+        assert!(result.passed, "1k prune_by_age exceeded budget");
+        println!("  prune_by_age(1k, 50%): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    #[test]
+    fn bench_prune_10k() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_bench(&env, 10_000, 100, BUDGET_10K);
+        assert!(result.passed, "10k prune exceeded budget: {} > {}", result.cpu_instructions, result.budget);
+        println!("  prune(10k → 100): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    #[test]
+    fn bench_prune_10k_by_age() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_by_age_bench(&env, 10_000, 0.5, BUDGET_10K);
+        assert!(result.passed, "10k prune_by_age exceeded budget");
+        println!("  prune_by_age(10k, 50%): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    #[test]
+    #[ignore = "expensive: 100k entries; run with --ignored"]
+    fn bench_prune_100k() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_bench(&env, 100_000, 100, BUDGET_100K);
+        assert!(result.passed, "100k prune exceeded budget: {} > {}", result.cpu_instructions, result.budget);
+        println!("  prune(100k → 100): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    #[test]
+    #[ignore = "expensive: 100k entries; run with --ignored"]
+    fn bench_prune_100k_by_age() {
+        let env = Env::default();
+        env.mock_all_auths();
+        env.budget().reset_unlimited();
+
+        let result = run_prune_by_age_bench(&env, 100_000, 0.5, BUDGET_100K);
+        assert!(result.passed, "100k prune_by_age exceeded budget");
+        println!("  prune_by_age(100k, 50%): {} instructions [PASS]", result.cpu_instructions);
+    }
+
+    // ================================================================
+    // Aggregate benchmark reporter (produces the reproducible artifact)
+    // ================================================================
+
+    #[test]
+    #[ignore = "aggregate: runs all benchmarks including 100k; run with --ignored"]
+    fn bench_prune_full_report() {
+        println!("\n=== PRUNE BENCHMARK REPORT (#204) ===\n");
+
+        let mut results: Vec<PruneBenchEntry> = Vec::new();
+
+        // 1k benchmarks
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_bench(&env, 1_000, 100, BUDGET_1K));
+        }
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_by_age_bench(&env, 1_000, 0.5, BUDGET_1K));
+        }
+
+        // 10k benchmarks
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_bench(&env, 10_000, 100, BUDGET_10K));
+        }
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_by_age_bench(&env, 10_000, 0.5, BUDGET_10K));
+        }
+
+        // 100k benchmarks
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_bench(&env, 100_000, 100, BUDGET_100K));
+        }
+        {
+            let env = Env::default();
+            env.mock_all_auths();
+            env.budget().reset_unlimited();
+            results.push(run_prune_by_age_bench(&env, 100_000, 0.5, BUDGET_100K));
+        }
+
+        // Print results table
+        println!("{:<12} {:<14} {:<18} {:<14} {:<8}", "Size", "Operation", "CPU Instructions", "Budget", "Status");
+        println!("{}", "-".repeat(70));
+
+        let mut all_passed = true;
+        for r in &results {
+            let op = if r.prune_kept == 100 { "prune_history" } else { "prune_by_age" };
+            let status = if r.passed { "PASS" } else { "FAIL" };
+            println!(
+                "{:<12} {:<14} {:<18} {:<14} {:<8}",
+                format!("{}", r.size),
+                op,
+                format!("{}", r.cpu_instructions),
+                format!("{}", r.budget),
+                status,
+            );
+            if !r.passed {
+                all_passed = false;
+            }
+        }
+
+        // Print JSON artifact for CI ingestion
+        println!("\n--- BENCHMARK ARTIFACT (JSON) ---");
+        print!("[");
+        for (i, r) in results.iter().enumerate() {
+            if i > 0 {
+                print!(",");
+            }
+            print!(
+                "{{\"size\":{},\"operation\":\"{}\",\"cpu_instructions\":{},\"budget\":{},\"passed\":{}}}",
+                r.size,
+                if r.prune_kept == 100 { "prune_history" } else { "prune_by_age" },
+                r.cpu_instructions,
+                r.budget,
+                r.passed,
+            );
+        }
+        println!("]");
+        println!("--- END BENCHMARK ARTIFACT ---\n");
+
+        assert!(all_passed, "One or more prune benchmarks exceeded their budget");
+    }
+}