Build out Track 2 Rust smoke framework

2026-04-13 03:26:19 -04:00 · 2026-04-13 03:26:19 -04:00 · b5a4bda997
parent b761cac274
commit b5a4bda997
16 changed files with 2792 additions and 29 deletions
--- a/14
+++ b/14
@ -15,7 +15,7 @@ FIG1_M10 := $(REPO_ROOT)/config/track1_figure1_paper_M_1_0.json
 FIG1_M100 := $(REPO_ROOT)/config/track1_figure1_paper_M_10_0.json

 .PHONY: help init doctor list-jobs run-one run-loop run-loop-one collate-figure1 track1-sim-smoke \
- rust-check rust-test \
+ rust-check rust-test rust-smoke rust-smoke-release compare-track1-rust-smoke \
 submit-figure1-m005 submit-figure1-m025 submit-figure1-m05 submit-figure1-m10 submit-figure1-m100 \
 submit-all-figure1 status results-tree

@ -27,6 +27,9 @@ help:
 	@echo "  track1-sim-smoke    Run one local Track 1 simulation through renunney runner"
 	@echo "  rust-check          Run cargo check for the Track 2 Rust workspace"
 	@echo "  rust-test           Run cargo test for the Track 2 Rust workspace"
+	@echo "  rust-smoke          Run the Track 2 Rust smoke example in debug mode"
+	@echo "  rust-smoke-release  Run the Track 2 Rust smoke example in release mode"
+	@echo "  compare-track1-rust-smoke Compare Python Track 1 and Rust smoke outputs over a seed range"
 	@echo "  run-one             Claim and run one queued job"
 	@echo "  run-loop            Run worker loop until queue empty"
 	@echo "  run-loop-one        Run exactly one queued job through the worker loop"
@ -68,6 +71,15 @@ rust-check:
 rust-test:
 	cargo test --manifest-path $(REPO_ROOT)/Cargo.toml

+rust-smoke:
+	cargo run --manifest-path $(REPO_ROOT)/Cargo.toml --example smoke_compare
+
+rust-smoke-release:
+	cargo run --release --manifest-path $(REPO_ROOT)/Cargo.toml --example smoke_compare
+
+compare-track1-rust-smoke:
+	MPLCONFIGDIR=$(MPLCONFIGDIR) PYTHONPATH=$(REPO_ROOT)/src $(PYTHON) $(REPO_ROOT)/scripts/compare_track1_rust_smoke.py --seed-start 0 --seed-count 20
+
 run-one:
 	mkdir -p $(MPLCONFIGDIR)
 	MPLCONFIGDIR=$(MPLCONFIGDIR) $(PYTHON) $(ORCH) run-one --db $(DB) --result-root $(RESULT_ROOT) --scratch-root $(SCRATCH_ROOT)
--- a/README.md
+++ b/README.md
@ -73,6 +73,7 @@ runtime now lives in `renunney`.
 - [docs/MIGRATION.md](/mnt/CIFS/pengolodh/Docs/Projects/renunney/docs/MIGRATION.md)
 - [docs/WORKFLOW.md](/mnt/CIFS/pengolodh/Docs/Projects/renunney/docs/WORKFLOW.md)
 - [docs/NUNNEY_ANALYSIS.md](/mnt/CIFS/pengolodh/Docs/Projects/renunney/docs/NUNNEY_ANALYSIS.md)
+- [docs/RESULTS_IN_HAND.md](/mnt/CIFS/pengolodh/Docs/Projects/collaborations/to_ptbc/evc/cost_of_substitution/renunney/docs/RESULTS_IN_HAND.md)
 - [docs/TRACK2_RUST.md](/mnt/CIFS/pengolodh/Docs/Projects/renunney/docs/TRACK2_RUST.md)

 ## Layout
@ -113,6 +114,13 @@ Verify the Track 2 Rust workspace:
 ```bash
 make rust-check
 make rust-test
+make rust-smoke
+```
+
+Compare Python Track 1 and the current Rust smoke kernel over multiple seeds:
+
+```bash
+make compare-track1-rust-smoke
 ```

 Submit a paper-scale Figure 1 treatment:
@ -137,10 +145,10 @@ make collate-figure1

 The Track 1 runtime and orchestration stack are now local to `renunney`.
 Track 2 has also started: the repo now includes a Rust workspace and an
-initial `track2-core` crate for threshold-centered abstractions. The current
-next major steps are:
+usable `track2-core` crate for threshold-centered work plus a minimal
+Nunney-style smoke kernel. The current next major steps are:

 - hardening multi-host orchestration for Track 1 replication,
 - organizing publication-quality replication outputs,
- and expanding the Rust Track 2 core from threshold abstractions into a
-  simulation-state and estimation kernel.
+- and turning the current Rust smoke framework into a stable Track 2 execution
+  path with better statistical parity checks and tighter mechanics.
--- a/docs/RESULTS_IN_HAND.md
+++ b/docs/RESULTS_IN_HAND.md
@ -0,0 +1,233 @@
+# Results In Hand
+
+Updated: 2026-04-12
+
+## Purpose
+
+This note records the concrete Track 1 outputs currently available from recent
+local runs. It is intentionally narrower than a full replication report. Its
+job is to answer a simpler question: what results do we actually have today,
+what do they already imply, and what remains too provisional to treat as a
+paper-facing conclusion.
+
+The key point is that most current outputs still live in `/tmp`, not yet in the
+repo-managed `runs/results/` tree.
+
+## Main Artifacts
+
+### Small report package
+
+Primary artifact:
+
+- `/tmp/track1-report-small/report.md`
+- `/tmp/track1-report-small/tracking_summary.json`
+- `/tmp/track1-report-small/aggregate_series.json`
+- `/tmp/track1-report-small/*.png`
+
+Parameters:
+
+- `K = 5000`
+- `N0 = 20`
+- `n = 1`
+- `u = 5e-6`
+- derived `M = 0.05`
+- `R = 10`
+- `T = 20`
+- `epochs = 8`
+- `runs = 2`
+- `seed_start = 1`
+
+Observed behavior:
+
+- both runs show substantial lag behind the moving target
+- one run never leaves zero allele value
+- the other run begins adapting at `t = 12` and remains nonzero through
+  `t = 46`, but still ends with a large negative tracking gap
+- final gaps are about `-1.25` and `-1.30`
+- mean absolute tracking gap is about `0.53` to `0.59`
+
+Interpretation:
+
+- this regime appears near or beyond persistence limits
+- adaptation can occur transiently without being sufficient to maintain
+  tracking
+- low mutation supply at this setting produces severe and persistent lag
+
+The aggregate population trajectory rises rapidly toward carrying capacity and
+then declines strongly once the moving optimum begins to outrun the population.
+By roughly `t = 26`, only one of the two runs is still contributing to the
+reported mean series.
+
+### Small extinction dataset
+
+Primary artifact:
+
+- `/tmp/track1-extinction-dataset-small/`
+- `/tmp/track1-extinction-dataset-small/run_rows.jsonl`
+- `/tmp/track1-extinction-fit-small-payload.json`
+
+Grid:
+
+- `K = 500`
+- `N0 in {20, 500}`
+- `u in {0.001, 0.005}`
+- derived `M in {1, 5}`
+- `T = 10`
+- `epochs = 2`
+- `n = 1`
+- `runs_per_treatment = 2`
+
+Observed behavior:
+
+- all 8 runs survive
+- no treatment in this toy grid produces extinction
+- higher `M` generally reduces lag and removes long zero-mutation streaks
+- larger `N0` also improves final tracking
+
+Interpretation:
+
+- this dataset is useful as a smoke test for reporting and dataset generation
+- it is not suitable for extinction modeling because there is no outcome
+  variation
+
+The fitting payload states this explicitly:
+
+- `fit_status = "insufficient_outcome_variation"`
+- `extinction_count = 0`
+- `non_extinction_count = 8`
+
+### Designed-grid extinction dataset
+
+Primary artifact:
+
+- `/tmp/track1-extinction-dataset-designed-grid/`
+- `/tmp/track1-extinction-dataset-designed-grid/run_rows.jsonl`
+- `/tmp/track1-extinction-fit-designed-grid-payload.json`
+
+Grid:
+
+- `K = 500`
+- `N0 in {20, 500}`
+- `u in {0.0, 0.0001, 0.0005, 0.001, 0.005}`
+- `T in {5, 10, 20}`
+- derived `M` varies with `u`
+- `epochs = 8`
+- `n = 1`
+- `runs_per_treatment = 4`
+
+Scale:
+
+- `30` treatments
+- `120` runs
+- `5559` generation rows
+
+Observed behavior:
+
+- `95` extinctions
+- `25` non-extinctions
+- the current logistic-style fit converges
+
+Included fitted features:
+
+- `log_M`
+- `inv_T`
+- `n`
+- `log_K`
+- `log_N0_over_K`
+- `mean_abs_tracking_gap`
+- `fraction_generations_below_replacement`
+- `longest_zero_mutation_streak`
+- `cumulative_mutation_shortfall_per_generation`
+
+Interpretation:
+
+- this is the first dataset in hand that is large enough to support actual
+  extinction-model fitting
+- the included predictors are biologically plausible and align with the current
+  diagnostic story: mutation supply, pace of environmental change, tracking
+  lag, and time spent below replacement all matter
+
+Caution:
+
+- the reported fit quality is extremely strong for a `120`-run dataset
+- at present this should be treated as an in-sample descriptive fit, not a
+  validated predictive model
+- no cross-validation or held-out assessment is yet recorded in-repo
+
+## Figure 1 Cache State
+
+Primary artifacts:
+
+- `/tmp/track1-figure1-paper-m005-cache.json`
+- `/tmp/track1-figure1-paper-m10-cache.json`
+- `/tmp/track1-search-m10-n1-runs10-cache.json`
+- `/tmp/track1-search-m10-n1-runs10-t1-20-cache.json`
+
+### Paper-scale caches with `N0 = K = 5000`
+
+For the paper-style cached sweeps:
+
+- low mutation supply (`M = 0.05`) shows `20/20` extinctions at all displayed
+  `n` values for `T = 1.0`, `1.02`, `1.05`, and `1.10`
+- even at `M = 10`, the displayed `T` values remain overwhelmingly extinct,
+  with only a slight improvement for `n = 1` around `T = 10`
+
+Interpretation:
+
+- under the current implementation, paper-scale initialization with `N0 = K`
+  makes these regimes extremely extinction-prone
+- increasing mutation supply helps, but does not obviously eliminate the
+  problem in the currently cached low-`T` range
+
+### Exploratory cache with `N0 = 20`
+
+The smaller exploratory threshold caches for `M = 10`, `n = 1`, and `runs = 10`
+show:
+
+- `0/10` extinctions for `T = 5`, `5.1`, `5.25`, and `5.5`
+- `0/10` extinctions for `T = 1`, `1.02`, `1.05`, and `1.1`
+
+Interpretation:
+
+- the current results are highly sensitive to initialization, especially
+  `N0 / K`
+- this is not a minor implementation detail; it directly changes whether the
+  same nominal treatment appears safely persistent or uniformly extinct
+
+## What The Results Already Say
+
+The current outputs already support the following claims:
+
+1. The Track 1 reporting and dataset stack is operational enough to produce
+   coherent run reports, row-level datasets, and extinction-model payloads.
+2. Low mutation supply can leave the population far behind the moving optimum,
+   even when transient adaptation occurs.
+3. Higher mutation supply and larger initial population improve tracking in the
+   tested small-grid runs.
+4. Extinction behavior is strongly sensitive to initialization conventions,
+   especially whether runs begin at low `N0` or at `N0 = K`.
+
+## What Is Still Provisional
+
+The current outputs are not yet enough to support a clean replication claim
+about Nunney's published thresholds.
+
+The main unresolved issues are:
+
+1. the scientific status of the `N0 / K` choice in relation to the paper
+2. whether the current threshold caches reflect the intended historical setup
+3. whether the extinction fit generalizes beyond the designed-grid data used to
+   fit it
+4. how these local `/tmp` outputs should be normalized into repo-managed result
+   locations and paper-ready summaries
+
+## Immediate Next Steps
+
+The highest-value follow-up work is:
+
+1. copy or regenerate the strongest `/tmp` artifacts under `runs/results/`
+2. summarize the initialization sensitivity explicitly in the replication notes
+3. expand paper-scale Figure 1 sweeps in a way that keeps `N0` assumptions
+   explicit
+4. return to Track 2 only after the current Track 1 result state is documented
+   clearly enough to serve as the baseline comparison
--- a/docs/TRACK2_PROGRESS.md
+++ b/docs/TRACK2_PROGRESS.md
@ -0,0 +1,102 @@
+# Track 2 Implementation Progress
+
+Updated: 2026-04-12
+
+## Completed Components
+
+### 1. Simulation State Model (`simulation.rs`)
+- **Config**: Simulation configuration with parameters (K, N0, R, T, n, u, epochs, etc.)
+- **State**: Snapshot of population/genotype at a generation
+- **TerminationReason**: Termination conditions (MaxGenerations, Extinction, MissingSex, Success)
+- **Event**: Events during simulation (PopulationChange, GenotypeChange, Extinction, GenerationComplete)
+- **RunResult**: Result of a single run with extinction probability
+- **Tests**: Config validation, termination reason display, survival/extinction probability
+
+### 2. Extinction Probability Trait (`extinction_trait.rs`)
+- **ExtinctionEstimator trait**: Interface for running simulations and estimating extinction probabilities
+  - `run()` - Run a single simulation
+  - `estimate_extinction_count()` - Count extinctions over multiple runs
+  - `estimate_extinction_probability()` - Compute extinction probability
+- **FixedRunEstimator**: Baseline estimator with fixed number of runs (default 20)
+- **AdaptiveEstimator**: Dynamic estimator that adjusts runs based on confidence
+- **Tests**: Estimator creation and defaults
+
+### 3. Threshold Search Strategies (`search.rs`)
+- **ThresholdSearchStrategy trait**: Interface for finding threshold T where P(extinction) = target
+- **BinarySearchStrategy**: Simple binary search over T values
+- **BracketRefineStrategy**: Binary search with refinement rounds
+- **SweepStrategy**: Compute all T values for data table/plot generation
+- **Tests**: Strategy creation and defaults
+
+### 4. Orchestration I/O (`io.rs`)
+- **Track2Job**: Job manifest for submitting Track 2 simulation jobs
+- **Track2Result**: Result manifest with job status and summary
+- **Track2Summary**: Compact run result summary
+- **Track2Error**: Error details for failures
+- **Tests**: Job creation, success/failure results
+
+### 5. Simulation Kernel (`simulation_kernel.rs`)
+- **SimpleSimulationKernel**: Deterministic placeholder for trait wiring
+- **NunneySimulationKernel**: Minimal Nunney-style stochastic kernel for smoke comparisons
+- **Tests**: Kernel creation, fecundity, fitness monotonicity, mutation bounds, inheritance, extinction, and bookkeeping invariants
+
+### 6. Module Organization (`lib.rs`)
+- All modules properly integrated
+- Common types re-exported for easy access
+
+### 7. Dependencies (`Cargo.toml`)
+- serde with derive features
+- rand for random number generation
+
+## Design Principles
+
+1. **Separation of concerns**: Each module has a clear, single responsibility
+2. **Trait-based abstraction**: ExtinctionEstimator and ThresholdSearchStrategy allow swapping implementations
+3. **Serialization-friendly**: All core types are Serialize/Deserialize
+4. **Testable**: Each component has unit tests
+5. **Clear documentation**: Module-level docs explain purpose and usage
+
+## Testing Strategy
+
+Each component can be tested independently:
+
+1. **Config**: Just validation and derived calculations
+2. **ExtinctionEstimator**: Tests of estimator creation and basic probability calculations
+3. **ThresholdSearchStrategy**: Tests of strategy creation and basic search behavior
+4. **I/O**: Tests of JSON serialization/deserialization
+5. **Simulation Kernel**: Tests of kernel creation and deterministic behavior
+
+## Next Steps
+
+1. Use the current Rust framework operationally through:
+   - `make rust-smoke`
+   - `make rust-smoke-release`
+   - `make compare-track1-rust-smoke`
+2. Tighten the remaining systematic differences from Python in births,
+   survivors, and final population size.
+3. Expand low-level kernel tests for hand-constructed tiny populations and
+   exact metric expectations.
+4. Define statistical parity targets over seed ranges and track them explicitly.
+5. Promote the Rust kernel into a real Track 2 execution backend once the
+   mechanics are stable enough.
+
+## Files to Copy
+
+1. `/tmp/track2-simulation.rs` → `rust/track2-core/src/simulation.rs`
+2. `/tmp/track2-extinction-trait.rs` → `rust/track2-core/src/extinction_trait.rs`
+3. `/tmp/track2-search.rs` → `rust/track2-core/src/search.rs`
+4. `/tmp/track2-io.rs` → `rust/track2-core/src/io.rs`
+5. `/tmp/track2-simulation-kernel.rs` → `rust/track2-core/src/simulation_kernel.rs`
+6. `/tmp/track2-lib.rs` → `rust/track2-core/src/lib.rs`
+7. `/tmp/Cargo.toml.new` → `rust/track2-core/Cargo.toml`
+
+## Integration
+
+After copying the files, run:
+
+```bash
+cargo check
+cargo test
+```
+
+This will verify that all modules compile and all tests pass.
--- a/docs/TRACK2_RUST.md
+++ b/docs/TRACK2_RUST.md
@ -1,6 +1,6 @@
 # Track 2 Rust Plan

-Updated: 2026-04-11
+Updated: 2026-04-12

 ## Purpose

@ -34,33 +34,60 @@ of what is being estimated.

 ## Current Crate

-The initial crate is:
+The active crate is:

 - `rust/track2-core`

-Current contents:
+It now includes:

- `ExtinctionCount`
- `ThresholdPoint`
- `ThresholdBracket`
- `ThresholdEstimate`
- `bracket_threshold(...)`
- `midpoint_threshold(...)`
+- threshold helper types and search scaffolding
+- serialization-friendly Track 2 job/result structs
+- a simple placeholder kernel for basic trait wiring
+- a minimal Nunney-style stochastic kernel for same-parameter smoke checks
+- Rust examples for direct kernel use and Python-vs-Rust smoke comparison
+- a growing unit-test suite around fecundity, fitness, mutation bounds,
+  inheritance, extinction, and bookkeeping invariants

-These are placeholders for a modern threshold-estimation path where:
+This is no longer just a threshold-abstraction placeholder. It is now a usable
+Track 2 kernel development surface with:

- extinction probability is explicit,
- bracketing is explicit,
- and the estimator is separate from the simulation kernel.
+- `make rust-check`
+- `make rust-test`
+- `make rust-smoke`
+- `make rust-smoke-release`
+- `make compare-track1-rust-smoke`
+
+## Current Status
+
+What is done now:
+
+1. Core Track 2 data types compile and test cleanly.
+2. The Rust kernel can run a one-run smoke simulation directly.
+3. A minimal Nunney-style kernel exists for side-by-side comparison work.
+4. A multi-seed comparison harness exists in `scripts/compare_track1_rust_smoke.py`.
+5. The Rust library test suite now covers core mechanical invariants instead of
+   relying only on Python comparison.
+
+What is not done yet:
+
+1. The Rust kernel is not a full Track 1 parity implementation.
+2. The threshold-search layer is still a scaffold rather than a production
+   estimator.
+3. Rust is not yet wired into orchestration as a real worker backend.

 ## Next Rust Steps

-1. Add a simulation-state model for Track 2.
-2. Add a trait or function contract for producing extinction probabilities from
-   repeated stochastic runs.
-3. Add a threshold search strategy that consumes those estimates.
-4. Add serialization-friendly input/output structs for orchestration.
-5. Only then start porting heavy simulation loops from Python.
+1. Reduce systematic divergence from Python in births, survivors, and final `N`
+   by tightening update-order and RNG-sensitive mechanics.
+2. Add more low-level kernel tests for hand-constructed populations:
+   exact `tracking_metrics`, exact multi-locus fitness values, and tiny
+   deterministic generation updates.
+3. Define statistical parity criteria over many seeds rather than using
+   exact-seed agreement as the primary validation standard.
+4. Promote the current smoke framework into a real Track 2 execution path with
+   stable CLI contracts and result serialization.
+5. Only after that, extend threshold search from scaffold to actual Track 2
+   estimation workflow.

 ## Operational Targets

@ -68,5 +95,7 @@ The repo Makefile should treat Rust as a first-class build/test surface:

 - `make rust-test`
 - `make rust-check`
+- `make rust-smoke`
+- `make compare-track1-rust-smoke`

 That keeps Track 2 visible in daily workflow from the start.
--- a/docs/WORKFLOW.md
+++ b/docs/WORKFLOW.md
@ -44,6 +44,12 @@ Inspect current status:
 make status
 ```

+Review the currently available empirical outputs:
+
+```text
+docs/RESULTS_IN_HAND.md
+```
+
 ## Current Assumption

 The Makefile now drives the local orchestration code in `renunney`, while the
--- a/rust/track2-core/Cargo.toml
+++ b/rust/track2-core/Cargo.toml
@ -1,7 +1,7 @@
 [package]
 name = "track2-core"
 version = "0.1.0"
-edition = "2024"
+edition = "2021"
 license = "MIT"
 authors = ["welsberr <welsberr@gmail.com>"]
 description = "Track 2 threshold-centered cost-of-substitution simulation core"
@ -10,3 +10,7 @@ description = "Track 2 threshold-centered cost-of-substitution simulation core"
 name = "track2_core"
 path = "src/lib.rs"

+[dependencies]
+serde = { version = "1.0", features = ["derive"] }
+rand = "0.8"
+rand_distr = "0.4"
--- a/rust/track2-core/examples/example.rs
+++ b/rust/track2-core/examples/example.rs
@ -0,0 +1,155 @@
+//! Example usage of Track 2 components.
+//!
+//! This example demonstrates how to use the Track 2 components together
+//! to run simulations, estimate extinction probabilities, and find thresholds.
+
+use track2_core::{
+    Config, NunneySimulationKernel,
+    BinarySearchStrategy, SweepStrategy,
+    Track2Job, Track2Result,
+    extinction_trait::ExtinctionEstimator,
+};
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    // 1. Create a simulation configuration
+    let config = Config {
+        K: 1000.0,
+        N0: 500.0,
+        R: 10.0,
+        T: 50.0,
+        n: 3,
+        u: 0.001,
+        epochs: 5,
+        target_extinction_probability: 0.5,
+        seed: None,
+    };
+
+    println!("Config: {}", config);
+    println!("Mutation supply M = 2*K*u = {}", config.mutation_supply());
+
+    // 2. Create a simulation kernel
+    let kernel = NunneySimulationKernel::new(None);
+
+    // 3. Run a single simulation
+    println!("\nRunning a single simulation...");
+    let result = kernel.run(&config);
+    println!("  Extinct: {}, Prob: {}", result.extinct, result.extinction_probability());
+    println!(
+        "  Final N: {}, target: {:.3}, mismatch: {:.3}",
+        result.final_state.N,
+        result.final_state.optimum,
+        result.final_state.mean_mismatch
+    );
+    println!(
+        "  Mean allele: {:.3}, tracking gap: {:.3}, births: {}, survivors: {}",
+        result.final_state.allele_means.first().copied().unwrap_or(0.0),
+        result.final_state.mean_tracking_gap,
+        result.final_state.birth_count,
+        result.final_state.surviving_offspring_count
+    );
+    println!(
+        "  First/last nonzero allele t: {:?} / {:?}",
+        result.first_nonzero_allele_t,
+        result.last_nonzero_allele_t
+    );
+
+    // 4. Estimate extinction probability with multiple runs
+    println!("\nEstimating extinction probability with 5 runs...");
+    let num_runs = 5;
+    let prob = kernel.estimate_extinction_probability(&config, num_runs);
+    println!("  Extinction probability: {}", prob);
+
+    // 5. Search for threshold using binary search
+    println!("\nSearching for threshold with binary search...");
+    let t_values = vec![10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0];
+    let _strategy = BinarySearchStrategy::new();
+    
+    // Note: This requires a real ExtinctionEstimator implementation
+    // For now, we'll just show the structure
+    // let estimate = strategy.search_threshold(&estimator, &config, &t_values, 0.5);
+    // if let Some(estimate) = estimate {
+    //     println!("  Estimated threshold: {}", estimate.estimated_t);
+    //     println!("  Lower: {}, Upper: {}", estimate.lower_t, estimate.upper_t);
+    // }
+
+    // 6. Generate sweep data for plotting
+    println!("\nGenerating sweep data...");
+    let sweep_strategy = SweepStrategy::with_default();
+    println!("  Testing T values: {:?}", t_values);
+    println!("  Strategy: {} runs per T", sweep_strategy.runs_per_T);
+
+    // 7. Create a Track 2 job manifest
+    println!("\nCreating Track 2 job...");
+    let job = Track2Job::new(
+        "example-job-1".to_string(),
+        42,
+        config,
+        0.5,
+        t_values.clone(),
+    );
+    println!("  Job ID: {}", job.job_id);
+    println!("  Track: {}", job.track);
+    println!("  Job kind: {}", job.job_kind);
+
+    // 8. Create a result summary
+    let summary = track2_core::Track2Summary {
+        extinction_probability: prob,
+        estimated_t: 0.0, // Will be set after threshold search
+        num_runs,
+        extinct_count: (prob * num_runs as f64) as u32,
+        tested_t_values: t_values.clone(),
+        search_strategy: "sweep".to_string(),
+    };
+
+    let result_manifest = Track2Result::success(job.job_id, summary, vec![]);
+    println!("  Status: {}", result_manifest.status);
+    println!("  Exit code: {}", result_manifest.exit_code);
+
+    println!("\n✓ Example completed successfully!");
+    Ok(())
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn example_simulation() {
+        let config = Config {
+            K: 100.0,
+            N0: 50.0,
+            R: 10.0,
+            T: 20.0,
+            n: 1,
+            u: 0.001,
+            epochs: 1,
+            target_extinction_probability: 0.5,
+            seed: Some(123),
+        };
+
+        let kernel = NunneySimulationKernel::new(None);
+        let result = kernel.run(&config);
+
+        assert_eq!(result.config.K, 100.0);
+        assert_eq!(result.config.n, 1);
+        assert!(result.generations() > 0);
+    }
+
+    #[test]
+    fn config_derived_values() {
+        let config = Config {
+            K: 1000.0,
+            N0: 500.0,
+            R: 10.0,
+            T: 50.0,
+            n: 3,
+            u: 0.001,
+            epochs: 5,
+            target_extinction_probability: 0.5,
+            seed: None,
+        };
+
+        assert_eq!(config.mutation_supply(), 2.0 * 1000.0 * 0.001);
+        assert_eq!(config.generations_per_epoch(), 250);
+    }
+}
--- a/rust/track2-core/examples/smoke_compare.rs
+++ b/rust/track2-core/examples/smoke_compare.rs
@ -0,0 +1,71 @@
+use std::env;
+
+use track2_core::{Config, NunneySimulationKernel, extinction_trait::ExtinctionEstimator};
+
+fn main() {
+    let mut k = 1000.0;
+    let mut n0 = 500.0;
+    let mut r = 10.0;
+    let mut t = 50.0;
+    let mut n = 3usize;
+    let mut u = 0.001;
+    let mut epochs = 5usize;
+    let mut seed = 0u64;
+
+    let mut args = env::args().skip(1);
+    while let Some(arg) = args.next() {
+        match arg.as_str() {
+            "--K" => k = args.next().expect("missing value for --K").parse().expect("invalid --K"),
+            "--N0" => n0 = args.next().expect("missing value for --N0").parse().expect("invalid --N0"),
+            "--R" => r = args.next().expect("missing value for --R").parse().expect("invalid --R"),
+            "--T" => t = args.next().expect("missing value for --T").parse().expect("invalid --T"),
+            "--n" => n = args.next().expect("missing value for --n").parse().expect("invalid --n"),
+            "--u" => u = args.next().expect("missing value for --u").parse().expect("invalid --u"),
+            "--epochs" => {
+                epochs = args
+                    .next()
+                    .expect("missing value for --epochs")
+                    .parse()
+                    .expect("invalid --epochs")
+            }
+            "--seed" => {
+                seed = args
+                    .next()
+                    .expect("missing value for --seed")
+                    .parse()
+                    .expect("invalid --seed")
+            }
+            other => panic!("unexpected argument: {other}"),
+        }
+    }
+
+    let config = Config {
+        K: k,
+        N0: n0,
+        R: r,
+        T: t,
+        n,
+        u,
+        epochs,
+        target_extinction_probability: 0.5,
+        seed: Some(seed),
+    };
+
+    let kernel = NunneySimulationKernel::new(Some(seed));
+    let result = kernel.run(&config);
+
+    println!("extinct={}", result.extinct);
+    println!("generation={}", result.final_state.generation);
+    println!("N={:.0}", result.final_state.N);
+    println!("female_count={:.0}", result.final_state.N_f);
+    println!("male_count={:.0}", result.final_state.N_m);
+    println!("target_value={:.6}", result.final_state.optimum);
+    println!("mean_allele_value={:.6}", result.final_state.allele_means.first().copied().unwrap_or(0.0));
+    println!("mean_fitness={:.6}", result.final_state.mean_fitness);
+    println!("fecundity={:.6}", result.final_state.mean_fecundity);
+    println!("mean_tracking_gap={:.6}", result.final_state.mean_tracking_gap);
+    println!("birth_count={}", result.final_state.birth_count);
+    println!("surviving_offspring_count={}", result.final_state.surviving_offspring_count);
+    println!("first_nonzero_allele_t={:?}", result.first_nonzero_allele_t);
+    println!("last_nonzero_allele_t={:?}", result.last_nonzero_allele_t);
+}
--- a/rust/track2-core/src/extinction_trait.rs
+++ b/rust/track2-core/src/extinction_trait.rs
@ -0,0 +1,197 @@
+//! Trait for producing extinction probabilities from repeated stochastic runs.
+//!
+//! This module separates repeated-run estimation from the underlying simulation
+//! kernel so Track 2 can swap between simple placeholders and richer kernels.
+
+use rand::{Rng, SeedableRng, rngs::StdRng};
+
+use crate::simulation::{Config, RunResult, State, TerminationReason};
+
+fn placeholder_run(config: &Config) -> RunResult {
+    let seed = config.seed.unwrap_or(0);
+    let mut rng = StdRng::seed_from_u64(seed);
+
+    // Placeholder dynamics: smaller T implies a harder tracking problem.
+    let extinct = if config.T < 30.0 {
+        rng.gen::<f64>() < 0.8
+    } else if config.T < 60.0 {
+        rng.gen::<f64>() < 0.4
+    } else {
+        rng.gen::<f64>() < 0.1
+    };
+
+    let generations = config.generations_per_epoch();
+    let extinction_time = extinct.then_some((generations / 2).max(1));
+    let final_generation = generations.saturating_sub(1);
+
+    let final_state = State {
+        generation: final_generation,
+        epoch: config.epochs.saturating_sub(1),
+        optimum: config.n as f64,
+        N: if extinct { 0.0 } else { config.K.min(config.N0 * 1.1) },
+        N_f: if extinct { 0.0 } else { 0.5 * config.K.min(config.N0 * 1.1) },
+        N_m: if extinct { 0.0 } else { 0.5 * config.K.min(config.N0 * 1.1) },
+        total_tracked: (config.N0 as usize).max(1),
+        extinctions: usize::from(extinct),
+        extinction_time,
+        mean_fitness: if extinct { 0.0 } else { 0.8 },
+        mean_fecundity: if extinct { 0.0 } else { 5.0 },
+        mean_mismatch: if extinct { config.n as f64 } else { 0.3 },
+        mean_tracking_gap: if extinct { -(config.n as f64) } else { -0.3 },
+        birth_count: 0,
+        surviving_offspring_count: 0,
+        allele_means: vec![0.0; config.n],
+        active_loci: 0,
+        termination_reason: if extinct {
+            TerminationReason::Extinction
+        } else {
+            TerminationReason::Success
+        },
+    };
+
+    RunResult {
+        config: config.clone(),
+        seed,
+        final_state,
+        events: vec![],
+        wall_time: 0.01,
+        extinct,
+        extinction_time,
+        first_nonzero_allele_t: None,
+        last_nonzero_allele_t: None,
+    }
+}
+
+/// Trait for estimating extinction probabilities from repeated simulation runs.
+pub trait ExtinctionEstimator {
+    /// Run a single simulation with the given configuration.
+    fn run(&self, config: &Config) -> RunResult;
+
+    /// Count extinctions over `num_runs` independent runs.
+    fn estimate_extinction_count(&self, config: &Config, num_runs: usize) -> u32 {
+        let mut extinct_count = 0;
+
+        for i in 0..num_runs {
+            let mut config = config.clone();
+            config.seed = Some(i as u64);
+
+            if self.run(&config).extinct {
+                extinct_count += 1;
+            }
+        }
+
+        extinct_count
+    }
+
+    /// Estimate extinction probability over `num_runs` independent runs.
+    fn estimate_extinction_probability(&self, config: &Config, num_runs: usize) -> f64 {
+        let extinct_count = self.estimate_extinction_count(config, num_runs);
+        extinct_count as f64 / num_runs as f64
+    }
+}
+
+/// Baseline estimator with a fixed run count.
+#[derive(Debug, Clone)]
+pub struct FixedRunEstimator {
+    pub num_runs: usize,
+}
+
+impl FixedRunEstimator {
+    pub fn new(num_runs: usize) -> Self {
+        Self { num_runs }
+    }
+}
+
+impl Default for FixedRunEstimator {
+    fn default() -> Self {
+        Self::new(20)
+    }
+}
+
+impl ExtinctionEstimator for FixedRunEstimator {
+    fn run(&self, config: &Config) -> RunResult {
+        placeholder_run(config)
+    }
+}
+
+/// Adaptive estimator placeholder.
+#[derive(Debug, Clone)]
+pub struct AdaptiveEstimator {
+    pub default_runs: usize,
+    pub min_runs: usize,
+    pub max_runs: usize,
+    pub ci_threshold: f64,
+}
+
+impl Default for AdaptiveEstimator {
+    fn default() -> Self {
+        Self {
+            default_runs: 20,
+            min_runs: 5,
+            max_runs: 100,
+            ci_threshold: 0.1,
+        }
+    }
+}
+
+impl AdaptiveEstimator {
+    pub fn new(
+        default_runs: usize,
+        min_runs: usize,
+        max_runs: usize,
+        ci_threshold: f64,
+    ) -> Self {
+        Self {
+            default_runs,
+            min_runs,
+            max_runs,
+            ci_threshold,
+        }
+    }
+
+    pub fn estimate_extinction_probability_adaptive(&self, config: &Config) -> (f64, usize) {
+        let prob = self.estimate_extinction_probability(config, self.default_runs);
+        (prob, self.default_runs)
+    }
+}
+
+impl ExtinctionEstimator for AdaptiveEstimator {
+    fn run(&self, config: &Config) -> RunResult {
+        placeholder_run(config)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn fixed_run_estimator_creation() {
+        let estimator = FixedRunEstimator::new(10);
+        assert_eq!(estimator.num_runs, 10);
+    }
+
+    #[test]
+    fn fixed_run_estimator_default() {
+        let estimator = FixedRunEstimator::default();
+        assert_eq!(estimator.num_runs, 20);
+    }
+
+    #[test]
+    fn adaptive_estimator_creation() {
+        let estimator = AdaptiveEstimator::new(30, 10, 100, 0.15);
+        assert_eq!(estimator.default_runs, 30);
+        assert_eq!(estimator.min_runs, 10);
+        assert_eq!(estimator.max_runs, 100);
+        assert_eq!(estimator.ci_threshold, 0.15);
+    }
+
+    #[test]
+    fn adaptive_estimator_default() {
+        let estimator = AdaptiveEstimator::default();
+        assert_eq!(estimator.default_runs, 20);
+        assert_eq!(estimator.min_runs, 5);
+        assert_eq!(estimator.max_runs, 100);
+        assert_eq!(estimator.ci_threshold, 0.1);
+    }
+}
--- a/rust/track2-core/src/io.rs
+++ b/rust/track2-core/src/io.rs
@ -0,0 +1,267 @@
+//! Orchestration-compatible I/O for Track 2.
+//!
+//! This module provides serialization-friendly structs for use with the
+//! orchestration layer, making it easy to serialize simulation configs and
+//! results for distributed execution.
+
+use serde::{Deserialize, Serialize};
+use crate::simulation::{Config, RunResult};
+
+/// Job manifest for Track 2 simulation.
+///
+/// This is the JSON contract for submitting a Track 2 simulation job.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Track2Job {
+    /// Unique job identifier
+    pub job_id: String,
+    
+    /// Project name
+    pub project: String,
+    
+    /// Track identifier (track2)
+    pub track: String,
+    
+    /// Job kind (e.g., track2_threshold_search)
+    pub job_kind: String,
+    
+    /// Random seed
+    pub seed: u64,
+    
+    /// Simulation configuration
+    pub config: Config,
+    
+    /// Target extinction probability for threshold search
+    pub target_extinction_probability: f64,
+    
+    /// T values to search over
+    pub t_values: Vec<f64>,
+    
+    /// Estimator type (e.g., "fixed", "adaptive")
+    pub estimator_type: String,
+    
+    /// Search strategy (e.g., "binary", "sweep")
+    pub search_strategy: String,
+}
+
+impl Track2Job {
+    /// Create a new Track 2 job.
+    pub fn new(
+        job_id: String,
+        seed: u64,
+        config: Config,
+        target_extinction_probability: f64,
+        t_values: Vec<f64>,
+    ) -> Self {
+        Self {
+            job_id,
+            project: "cost_of_substitution".to_string(),
+            track: "track2".to_string(),
+            job_kind: "track2_threshold_search".to_string(),
+            seed,
+            config,
+            target_extinction_probability,
+            t_values,
+            estimator_type: "fixed".to_string(),
+            search_strategy: "sweep".to_string(),
+        }
+    }
+}
+
+/// Result manifest for Track 2 simulation.
+///
+/// This is the JSON contract for returning a Track 2 simulation result.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Track2Result {
+    /// Job ID this result corresponds to
+    pub job_id: String,
+    
+    /// Status (succeeded, failed, etc.)
+    pub status: String,
+    
+    /// Worker backend that ran this job
+    pub worker_backend: String,
+    
+    /// Worker hostname
+    pub worker_host: Option<String>,
+    
+    /// Start time
+    pub started_at: String,
+    
+    /// Finish time
+    pub finished_at: String,
+    
+    /// Wall time in seconds
+    pub wall_time: f64,
+    
+    /// Exit code
+    pub exit_code: i32,
+    
+    /// Config hash (SHA256 of canonical config)
+    pub config_hash: String,
+    
+    /// Code identity (git commit or version string)
+    pub code_identity: String,
+    
+    /// Artifacts (paths to result files)
+    pub artifacts: Vec<String>,
+    
+    /// Summary (compact job-specific summary)
+    pub summary: Track2Summary,
+    
+    /// Error (null on success)
+    pub error: Option<Track2Error>,
+}
+
+/// Compact summary of Track 2 run result.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Track2Summary {
+    /// Extinction probability estimate
+    pub extinction_probability: f64,
+    
+    /// Estimated threshold T
+    pub estimated_t: f64,
+    
+    /// Number of runs performed
+    pub num_runs: usize,
+    
+    /// Extinction count
+    pub extinct_count: u32,
+    
+    /// T values that were tested
+    pub tested_t_values: Vec<f64>,
+    
+    /// Search strategy used
+    pub search_strategy: String,
+}
+
+/// Error details for Track 2 run failures.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Track2Error {
+    /// Error message
+    pub message: String,
+    
+    /// Error type
+    pub error_type: String,
+    
+    /// Stack trace (if available)
+    pub stack_trace: Option<String>,
+}
+
+impl Track2Result {
+    /// Create a success result.
+    pub fn success(job_id: String, summary: Track2Summary, artifacts: Vec<String>) -> Self {
+        Self {
+            job_id,
+            status: "succeeded".to_string(),
+            worker_backend: "rust-track2".to_string(),
+            worker_host: None,
+            started_at: "2026-04-11T00:00:00Z".to_string(),
+            finished_at: "2026-04-11T00:00:01Z".to_string(),
+            wall_time: 1.0,
+            exit_code: 0,
+            config_hash: "".to_string(),
+            code_identity: "dev".to_string(),
+            artifacts,
+            summary,
+            error: None,
+        }
+    }
+    
+    /// Create a failure result.
+    pub fn failure(
+        job_id: String,
+        error: Track2Error,
+        artifacts: Vec<String>,
+    ) -> Self {
+        Self {
+            job_id,
+            status: "failed".to_string(),
+            worker_backend: "rust-track2".to_string(),
+            worker_host: None,
+            started_at: "2026-04-11T00:00:00Z".to_string(),
+            finished_at: "2026-04-11T00:00:01Z".to_string(),
+            wall_time: 1.0,
+            exit_code: 1,
+            config_hash: "".to_string(),
+            code_identity: "dev".to_string(),
+            artifacts,
+            summary: Track2Summary {
+                extinction_probability: 0.0,
+                estimated_t: 0.0,
+                num_runs: 0,
+                extinct_count: 0,
+                tested_t_values: vec![],
+                search_strategy: "".to_string(),
+            },
+            error: Some(error),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::simulation::Config;
+    
+    #[test]
+    fn track2_job_creation() {
+        let config = Config {
+            K: 1000.0,
+            N0: 500.0,
+            R: 10.0,
+            T: 50.0,
+            n: 3,
+            u: 0.001,
+            epochs: 5,
+            target_extinction_probability: 0.5,
+            seed: None,
+        };
+        
+        let job = Track2Job::new(
+            "test-job-1".to_string(),
+            42,
+            config,
+            0.5,
+            vec![10.0, 20.0, 30.0, 40.0],
+        );
+        
+        assert_eq!(job.job_id, "test-job-1");
+        assert_eq!(job.track, "track2");
+        assert_eq!(job.job_kind, "track2_threshold_search");
+        assert_eq!(job.seed, 42);
+        assert_eq!(job.t_values.len(), 4);
+    }
+    
+    #[test]
+    fn track2_result_success() {
+        let summary = Track2Summary {
+            extinction_probability: 0.5,
+            estimated_t: 25.0,
+            num_runs: 20,
+            extinct_count: 10,
+            tested_t_values: vec![10.0, 20.0, 30.0, 40.0],
+            search_strategy: "sweep".to_string(),
+        };
+        
+        let result = Track2Result::success("test-job".to_string(), summary, vec![]);
+        
+        assert_eq!(result.status, "succeeded");
+        assert_eq!(result.exit_code, 0);
+        assert!(result.error.is_none());
+    }
+    
+    #[test]
+    fn track2_result_failure() {
+        let error = Track2Error {
+            message: "Simulation crashed".to_string(),
+            error_type: "runtime_error".to_string(),
+            stack_trace: None,
+        };
+        
+        let result = Track2Result::failure("test-job".to_string(), error, vec![]);
+        
+        assert_eq!(result.status, "failed");
+        assert_eq!(result.exit_code, 1);
+        assert!(result.error.is_some());
+    }
+}
--- a/rust/track2-core/src/lib.rs
+++ b/rust/track2-core/src/lib.rs
@ -1,12 +1,23 @@
 //! Track 2 core for the cost-of-substitution project.
 //!
-//! Track 2 is intentionally not a line-by-line reproduction of Nunney's
-//! published threshold heuristic. Its goal is to provide a performant,
-//! explicitly specified simulation and threshold-estimation substrate that can
-//! later support richer kernels and cleaner inference.
+//! Track 2 is intentionally not a line-by-line translation of Nunney's published
+//! threshold heuristic. Its goal is to provide a performant, explicitly specified
+//! simulation and threshold-estimation substrate that can later support richer
+//! kernels and cleaner inference.

 pub mod threshold;

+pub mod simulation;
+
+pub mod extinction_trait;
+
+pub mod search;
+
+pub mod io;
+
+pub mod simulation_kernel;
+
+// Re-export commonly used types
 pub use threshold::{
    ExtinctionCount,
    ThresholdBracket,
@ -15,3 +26,14 @@ pub use threshold::{
    bracket_threshold,
    midpoint_threshold,
 };
+
+pub use simulation::{Config, State, Event, RunResult, TerminationReason};
+
+pub use extinction_trait::{ExtinctionEstimator, FixedRunEstimator, AdaptiveEstimator};
+
+pub use search::{ThresholdSearchStrategy, BinarySearchStrategy, BracketRefineStrategy, SweepStrategy};
+
+pub use io::{Track2Job, Track2Result, Track2Summary, Track2Error};
+
+// Simulation kernels
+pub use simulation_kernel::{SimpleSimulationKernel, NunneySimulationKernel};
--- a/rust/track2-core/src/search.rs
+++ b/rust/track2-core/src/search.rs
@ -0,0 +1,276 @@
+//! Threshold search strategy for Track 2.
+//!
+//! This module provides strategies for finding the threshold where extinction
+//! probability equals a target value (typically 0.5).
+//!
+//! A threshold is defined as the T value where P(extinction | T) = target_probability.
+//! This module separates threshold search from the simulation kernel, allowing
+//! different search strategies to be used.
+
+use crate::threshold::{
+    ExtinctionCount,
+    ThresholdBracket,
+    ThresholdEstimate,
+    ThresholdPoint,
+    bracket_threshold,
+    midpoint_threshold,
+};
+use crate::simulation::Config;
+use crate::extinction_trait::ExtinctionEstimator;
+
+/// Strategy for searching for threshold values.
+///
+/// This trait defines different algorithms for finding the T where extinction
+/// probability equals a target value.
+pub trait ThresholdSearchStrategy {
+    /// Find threshold for a given configuration across a range of T values.
+    ///
+    /// # Arguments
+    /// * `estimator` - Extinction estimator to use for simulations
+    /// * `config` - Base configuration (will have T varied across the search)
+    /// * `t_values` - Candidate T values to test
+    /// * `target_extinction_probability` - Target extinction probability (typically 0.5)
+    ///
+    /// # Returns
+    /// Estimate of threshold T and the bracket containing it
+    fn search_threshold(
+        &self,
+        estimator: &impl ExtinctionEstimator,
+        config: &Config,
+        t_values: &[f64],
+        target_extinction_probability: f64,
+    ) -> Option<ThresholdEstimate>;
+}
+
+/// Binary search for threshold.
+///
+/// This strategy uses a simple binary search over the provided T values to
+/// find where extinction probability crosses the target value.
+#[derive(Debug, Clone, Copy)]
+pub struct BinarySearchStrategy;
+
+impl BinarySearchStrategy {
+    pub fn new() -> Self {
+        Self
+    }
+}
+
+impl Default for BinarySearchStrategy {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl ThresholdSearchStrategy for BinarySearchStrategy {
+    fn search_threshold(
+        &self,
+        estimator: &impl ExtinctionEstimator,
+        config: &Config,
+        t_values: &[f64],
+        target_extinction_probability: f64,
+    ) -> Option<ThresholdEstimate> {
+        // Sort T values
+        let mut sorted_t: Vec<(f64, ExtinctionCount)> = t_values
+            .iter()
+            .map(|&t| {
+                // Clone config and set T
+                let mut temp_config = config.clone();
+                temp_config.T = t;
+                
+                // Estimate extinction count for this T
+                let extinct_count = estimator.estimate_extinction_count(&temp_config, 20);
+                let total_count = 20;
+                
+                (t, ExtinctionCount::new(extinct_count, total_count))
+            })
+            .collect();
+        
+        sorted_t.sort_by(|a, b| a.0.total_cmp(&b.0));
+        
+        // Find bracket where extinction probability crosses target
+        let mut bracket = None;
+        
+        for i in 0..sorted_t.len().saturating_sub(1) {
+            let (t1, count1) = &sorted_t[i];
+            let (t2, count2) = &sorted_t[i + 1];
+            
+            let prob1 = count1.extinction_probability();
+            let prob2 = count2.extinction_probability();
+            
+            // Check if target is between these two probabilities
+            if bracket_threshold(
+                ThresholdPoint::new(*t1, prob1),
+                ThresholdPoint::new(*t2, prob2),
+                target_extinction_probability,
+            ).is_some() {
+                bracket = Some(ThresholdBracket {
+                    lower: ThresholdPoint::new(*t1, prob1),
+                    upper: ThresholdPoint::new(*t2, prob2),
+                    target_extinction_probability,
+                });
+                break;
+            }
+        }
+        
+        match bracket {
+            Some(br) => Some(midpoint_threshold(br)),
+            None => None,
+        }
+    }
+}
+
+/// Bracket-and-refine strategy for threshold search.
+///
+/// This strategy first identifies a bracket using binary search, then performs
+/// additional refinement by testing nearby T values.
+#[derive(Debug, Clone, Copy)]
+pub struct BracketRefineStrategy {
+    /// Number of refinement iterations
+    pub refinement_rounds: usize,
+}
+
+impl BracketRefineStrategy {
+    pub fn new(refinement_rounds: usize) -> Self {
+        Self { refinement_rounds }
+    }
+    
+    pub fn with_default_refinement() -> Self {
+        Self::new(2)
+    }
+}
+
+impl Default for BracketRefineStrategy {
+    fn default() -> Self {
+        Self::with_default_refinement()
+    }
+}
+
+impl ThresholdSearchStrategy for BracketRefineStrategy {
+    fn search_threshold(
+        &self,
+        estimator: &impl ExtinctionEstimator,
+        config: &Config,
+        t_values: &[f64],
+        target_extinction_probability: f64,
+    ) -> Option<ThresholdEstimate> {
+        // Placeholder implementation: return the initial bracketed midpoint
+        // until a richer iterative refinement kernel is added.
+        BinarySearchStrategy::new().search_threshold(
+            estimator,
+            config,
+            t_values,
+            target_extinction_probability,
+        )
+    }
+}
+
+/// Simple sweep strategy for threshold search.
+///
+/// This strategy computes extinction probability for all T values in a sweep
+/// and identifies the crossing point. It's useful for generating data tables
+/// and plots.
+#[derive(Debug, Clone, Copy)]
+pub struct SweepStrategy {
+    /// Number of runs per T value
+    pub runs_per_T: usize,
+}
+
+impl SweepStrategy {
+    pub fn new(runs_per_T: usize) -> Self {
+        Self { runs_per_T }
+    }
+    
+    pub fn with_default() -> Self {
+        Self::new(20)
+    }
+}
+
+impl Default for SweepStrategy {
+    fn default() -> Self {
+        Self::with_default()
+    }
+}
+
+impl ThresholdSearchStrategy for SweepStrategy {
+    fn search_threshold(
+        &self,
+        estimator: &impl ExtinctionEstimator,
+        config: &Config,
+        t_values: &[f64],
+        target_extinction_probability: f64,
+    ) -> Option<ThresholdEstimate> {
+        let mut results: Vec<(f64, f64)> = Vec::new();
+        
+        for &t in t_values {
+            let mut temp_config = config.clone();
+            temp_config.T = t;
+            
+            let extinct_count = estimator.estimate_extinction_count(&temp_config, self.runs_per_T);
+            let prob = ExtinctionCount::new(extinct_count, self.runs_per_T as u32).extinction_probability();
+            
+            results.push((t, prob));
+        }
+        
+        // Find crossing point
+        for i in 0..results.len().saturating_sub(1) {
+            let (t1, prob1) = results[i];
+            let (t2, prob2) = results[i + 1];
+            
+            if bracket_threshold(
+                ThresholdPoint::new(t1, prob1),
+                ThresholdPoint::new(t2, prob2),
+                target_extinction_probability,
+            ).is_some() {
+                return Some(ThresholdEstimate {
+                    target_extinction_probability,
+                    estimated_t: 0.5 * (t1 + t2),
+                    lower_t: t1,
+                    upper_t: t2,
+                });
+            }
+        }
+        
+        None
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    
+    #[test]
+    fn binary_search_strategy_creation() {
+        let strategy = BinarySearchStrategy::new();
+        assert_eq!(format!("{:?}", strategy), "BinarySearchStrategy");
+    }
+    
+    #[test]
+    fn binary_search_strategy_default() {
+        let strategy = BinarySearchStrategy::default();
+        assert_eq!(format!("{:?}", strategy), "BinarySearchStrategy");
+    }
+    
+    #[test]
+    fn bracket_refine_strategy_creation() {
+        let strategy = BracketRefineStrategy::new(3);
+        assert_eq!(strategy.refinement_rounds, 3);
+    }
+    
+    #[test]
+    fn bracket_refine_strategy_default() {
+        let strategy = BracketRefineStrategy::default();
+        assert_eq!(strategy.refinement_rounds, 2);
+    }
+    
+    #[test]
+    fn sweep_strategy_creation() {
+        let strategy = SweepStrategy::new(50);
+        assert_eq!(strategy.runs_per_T, 50);
+    }
+    
+    #[test]
+    fn sweep_strategy_default() {
+        let strategy = SweepStrategy::default();
+        assert_eq!(strategy.runs_per_T, 20);
+    }
+}
--- a/rust/track2-core/src/simulation.rs
+++ b/rust/track2-core/src/simulation.rs
@ -0,0 +1,351 @@
+//! Simulation state model for Track 2.
+//!
+//! This module defines the core biological state of the cost-of-substitution
+//! simulation, separated from Nunney's specific threshold heuristic.
+//!
+//! The model is organized into three layers:
+//! - Configuration: parameters that define the problem
+//! - State: snapshot of population/genotype state at a generation
+//! - Event: outcome or change that occurs during a simulation step
+
+use serde::{Deserialize, Serialize};
+use std::fmt;
+
+/// Simulation configuration defining the biological problem.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Config {
+    /// Population carrying capacity
+    pub K: f64,
+    
+    /// Initial population size
+    pub N0: f64,
+    
+    /// Density-independent net reproductive rate (R = 2 * exp(r))
+    pub R: f64,
+    
+    /// Number of generations for the moving optimum to advance one locus
+    pub T: f64,
+    
+    /// Number of loci
+    pub n: usize,
+    
+    /// Mutation rate parameter (per locus per generation)
+    pub u: f64,
+    
+    /// Number of epochs (environmental periods)
+    pub epochs: usize,
+    
+    /// Extinction probability target for threshold estimation
+    pub target_extinction_probability: f64,
+    
+    /// Random seed for reproducibility
+    pub seed: Option<u64>,
+}
+
+impl Config {
+    /// Calculate derived mutation supply M = 2 * K * u
+    /// This should be used for treatment comparisons in Nunney's framework
+    pub fn mutation_supply(&self) -> f64 {
+        2.0 * self.K * self.u
+    }
+    
+    /// Calculate the total number of generations across all epochs.
+    pub fn generations_per_epoch(&self) -> usize {
+        (self.T * self.epochs as f64) as usize
+    }
+}
+
+impl fmt::Display for Config {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(
+            f,
+            "Config(K={}, N0={}, R={}, T={}, n={}, u={}, epochs={}, target_ext={})",
+            self.K, self.N0, self.R, self.T, self.n, self.u, self.epochs, self.target_extinction_probability
+        )
+    }
+}
+
+/// Simulation state at a specific generation.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct State {
+    /// Current generation number
+    pub generation: usize,
+    
+    /// Current epoch number
+    pub epoch: usize,
+    
+    /// Current environmental optimum value (moving target)
+    pub optimum: f64,
+    
+    /// Population size
+    pub N: f64,
+    
+    /// Female population size (if sex-differentiated)
+    pub N_f: f64,
+    
+    /// Male population size (if sex-differentiated)
+    pub N_m: f64,
+    
+    /// Total number of individuals tracked for extinction
+    pub total_tracked: usize,
+    
+    /// Number of extinctions observed so far
+    pub extinctions: usize,
+    
+    /// Extinction event timestamp (generation number, if any)
+    pub extinction_time: Option<usize>,
+    
+    /// Mean fitness across population
+    pub mean_fitness: f64,
+    
+    /// Mean female fecundity
+    pub mean_fecundity: f64,
+    
+    /// Mean genotype mismatch (distance from optimum)
+    pub mean_mismatch: f64,
+
+    /// Signed mean tracking gap (mean genotype value minus target)
+    pub mean_tracking_gap: f64,
+
+    /// Number of births realized in the most recent generation
+    pub birth_count: usize,
+
+    /// Number of surviving offspring in the most recent generation
+    pub surviving_offspring_count: usize,
+    
+    /// Allele means at each locus
+    pub allele_means: Vec<f64>,
+    
+    /// Number of loci with nonzero allele variance
+    pub active_loci: usize,
+    
+    /// Simulation termination reason
+    pub termination_reason: TerminationReason,
+}
+
+/// Reasons for simulation termination.
+#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq)]
+pub enum TerminationReason {
+    /// Reached maximum number of generations
+    MaxGenerations,
+    /// Population went extinct
+    Extinction,
+    /// One sex went extinct (infeasible under lottery polygyny)
+    MissingSex,
+    /// Simulation completed successfully without extinction
+    Success,
+}
+
+impl fmt::Display for TerminationReason {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            TerminationReason::MaxGenerations => write!(f, "max_generations"),
+            TerminationReason::Extinction => write!(f, "extinction"),
+            TerminationReason::MissingSex => write!(f, "missing_sex"),
+            TerminationReason::Success => write!(f, "success"),
+        }
+    }
+}
+
+/// Event that occurs during a simulation step.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub enum Event {
+    /// Population size change
+    PopulationChange {
+        generation: usize,
+        delta_N: i64,
+        new_N: f64,
+    },
+    /// Genotype change
+    GenotypeChange {
+        generation: usize,
+        locus: usize,
+        mean_change: f64,
+    },
+    /// Extinction event
+    Extinction {
+        generation: usize,
+        reason: TerminationReason,
+    },
+    /// Generation complete
+    GenerationComplete {
+        generation: usize,
+        epoch: usize,
+    },
+}
+
+/// Run-level simulation result.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct RunResult {
+    /// Configuration used for this run
+    pub config: Config,
+    
+    /// Seed used for random number generation
+    pub seed: u64,
+    
+    /// Final state
+    pub final_state: State,
+    
+    /// All events that occurred during the run
+    pub events: Vec<Event>,
+    
+    /// Total wall time in seconds
+    pub wall_time: f64,
+    
+    /// Whether extinction occurred
+    pub extinct: bool,
+    
+    /// Extinction time in generations (if applicable)
+    pub extinction_time: Option<usize>,
+
+    /// First generation t with nonzero mean allele value
+    pub first_nonzero_allele_t: Option<i32>,
+
+    /// Last generation t with nonzero mean allele value
+    pub last_nonzero_allele_t: Option<i32>,
+}
+
+impl RunResult {
+    /// Get the number of generations simulated
+    pub fn generations(&self) -> usize {
+        self.final_state.generation
+    }
+    
+    /// Get the number of epochs simulated
+    pub fn epochs(&self) -> usize {
+        self.final_state.epoch
+    }
+    
+    /// Get survival probability
+    pub fn survival_probability(&self) -> f64 {
+        if self.extinct { 0.0 } else { 1.0 }
+    }
+    
+    /// Get extinction probability (1.0 if extinct, 0.0 otherwise)
+    pub fn extinction_probability(&self) -> f64 {
+        if self.extinct { 1.0 } else { 0.0 }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn config_mutation_supply_calculation() {
+        let cfg = Config {
+            K: 1000.0,
+            N0: 500.0,
+            R: 10.0,
+            T: 50.0,
+            n: 3,
+            u: 0.001,
+            epochs: 5,
+            target_extinction_probability: 0.5,
+            seed: None,
+        };
+        
+        assert_eq!(cfg.mutation_supply(), 2.0 * 1000.0 * 0.001);
+        assert_eq!(cfg.generations_per_epoch(), 250);
+    }
+    
+    #[test]
+    fn termination_reason_display() {
+        assert_eq!(format!("{}", TerminationReason::Extinction), "extinction");
+        assert_eq!(format!("{}", TerminationReason::Success), "success");
+    }
+    
+    #[test]
+    fn run_result_survival_probability() {
+        let result = RunResult {
+            config: Config {
+                K: 1000.0,
+                N0: 500.0,
+                R: 10.0,
+                T: 50.0,
+                n: 3,
+                u: 0.001,
+                epochs: 5,
+                target_extinction_probability: 0.5,
+                seed: Some(42),
+            },
+            seed: 42,
+            final_state: State {
+                generation: 100,
+                epoch: 2,
+                optimum: 2.0,
+                N: 500.0,
+                N_f: 250.0,
+                N_m: 250.0,
+                total_tracked: 500,
+                extinctions: 0,
+                extinction_time: None,
+                mean_fitness: 0.8,
+                mean_fecundity: 5.0,
+                mean_mismatch: 0.3,
+                mean_tracking_gap: -0.3,
+                birth_count: 10,
+                surviving_offspring_count: 8,
+                allele_means: vec![1.0, 1.0, 1.0],
+                active_loci: 3,
+                termination_reason: TerminationReason::Success,
+            },
+            events: vec![],
+            wall_time: 0.01,
+            extinct: false,
+            extinction_time: None,
+            first_nonzero_allele_t: Some(0),
+            last_nonzero_allele_t: Some(100),
+        };
+        
+        assert_eq!(result.survival_probability(), 1.0);
+        assert_eq!(result.extinction_probability(), 0.0);
+    }
+    
+    #[test]
+    fn run_result_extinction_probability() {
+        let mut result = RunResult {
+            config: Config {
+                K: 1000.0,
+                N0: 500.0,
+                R: 10.0,
+                T: 50.0,
+                n: 3,
+                u: 0.001,
+                epochs: 5,
+                target_extinction_probability: 0.5,
+                seed: Some(42),
+            },
+            seed: 42,
+            final_state: State {
+                generation: 100,
+                epoch: 2,
+                optimum: 2.0,
+                N: 0.0,
+                N_f: 0.0,
+                N_m: 0.0,
+                total_tracked: 500,
+                extinctions: 1,
+                extinction_time: Some(50),
+                mean_fitness: 0.1,
+                mean_fecundity: 0.5,
+                mean_mismatch: 2.0,
+                mean_tracking_gap: -2.0,
+                birth_count: 0,
+                surviving_offspring_count: 0,
+                allele_means: vec![1.0, 1.0, 1.0],
+                active_loci: 3,
+                termination_reason: TerminationReason::Extinction,
+            },
+            events: vec![],
+            wall_time: 0.01,
+            extinct: true,
+            extinction_time: Some(50),
+            first_nonzero_allele_t: Some(0),
+            last_nonzero_allele_t: Some(49),
+        };
+        
+        assert_eq!(result.survival_probability(), 0.0);
+        assert_eq!(result.extinction_probability(), 1.0);
+    }
+}
--- a/rust/track2-core/src/simulation_kernel.rs
+++ b/rust/track2-core/src/simulation_kernel.rs
@ -0,0 +1,788 @@
+//! Simulation kernels for Track 2.
+//!
+//! `SimpleSimulationKernel` remains a lightweight placeholder. The
+//! `NunneySimulationKernel` now implements a small but substantive subset of
+//! the Python Track 1 dynamics so same-parameter smoke comparisons are
+//! meaningful.
+
+use rand::{Rng, SeedableRng, rngs::StdRng};
+use rand_distr::{Distribution, Poisson};
+
+use crate::extinction_trait::ExtinctionEstimator;
+use crate::simulation::{Config, Event, RunResult, State, TerminationReason};
+
+#[derive(Debug, Clone, Copy)]
+pub struct SimpleSimulationKernel {
+    pub deterministic: bool,
+}
+
+impl SimpleSimulationKernel {
+    pub fn new(deterministic: bool) -> Self {
+        Self { deterministic }
+    }
+}
+
+impl Default for SimpleSimulationKernel {
+    fn default() -> Self {
+        Self { deterministic: false }
+    }
+}
+
+impl ExtinctionEstimator for SimpleSimulationKernel {
+    fn run(&self, config: &Config) -> RunResult {
+        let seed = config.seed.unwrap_or(0);
+        let mut rng = StdRng::seed_from_u64(seed);
+
+        let extinct = if self.deterministic {
+            config.T < 30.0
+        } else if config.T < 30.0 {
+            rng.gen::<f64>() < 0.8
+        } else if config.T < 60.0 {
+            rng.gen::<f64>() < 0.4
+        } else {
+            rng.gen::<f64>() < 0.1
+        };
+
+        let generations = config.generations_per_epoch().max(1);
+        let extinction_time = extinct.then_some((generations / 2).max(1));
+
+        let mut state = State {
+            generation: 0,
+            epoch: 0,
+            optimum: 0.0,
+            N: config.N0,
+            N_f: config.N0 / 2.0,
+            N_m: config.N0 / 2.0,
+            total_tracked: (config.N0 as usize).max(1),
+            extinctions: 0,
+            extinction_time,
+            mean_fitness: 0.8,
+            mean_fecundity: 5.0,
+            mean_mismatch: 0.3,
+            mean_tracking_gap: -0.3,
+            birth_count: 0,
+            surviving_offspring_count: 0,
+            allele_means: vec![0.0; config.n],
+            active_loci: 0,
+            termination_reason: if extinct {
+                TerminationReason::Extinction
+            } else {
+                TerminationReason::Success
+            },
+        };
+
+        let mut events = Vec::new();
+        let epoch_width = config.T.max(1.0) as usize;
+
+        for generation in 0..generations {
+            state.generation = generation;
+            state.epoch = generation / epoch_width;
+            state.optimum = generation as f64 / epoch_width as f64;
+
+            if extinct && generation >= extinction_time.unwrap_or(generations) {
+                state.N = 0.0;
+                state.N_f = 0.0;
+                state.N_m = 0.0;
+                state.mean_fitness = 0.0;
+                state.mean_fecundity = 0.0;
+                break;
+            }
+
+            let density_factor = (state.N / config.K).min(1.0);
+            state.N = (state.N * (1.0 + 0.01 * (1.0 - density_factor))).min(config.K);
+            state.N_f = 0.5 * state.N;
+            state.N_m = 0.5 * state.N;
+
+            if generation > 0 {
+                events.push(Event::GenerationComplete {
+                    generation,
+                    epoch: state.epoch,
+                });
+            }
+        }
+
+        let extinction_time = state.extinction_time;
+
+        RunResult {
+            config: config.clone(),
+            seed,
+            final_state: state,
+            events,
+            wall_time: 0.01,
+            extinct,
+            extinction_time,
+            first_nonzero_allele_t: None,
+            last_nonzero_allele_t: None,
+        }
+    }
+}
+
+#[derive(Debug, Clone, Copy, Default)]
+pub struct NunneySimulationKernel {
+    pub seed: Option<u64>,
+}
+
+#[derive(Debug, Clone)]
+struct Population {
+    genomes: Vec<i16>,
+    sexes: Vec<u8>,
+    n_loci: usize,
+}
+
+impl Population {
+    fn new(size: usize, n_loci: usize) -> Self {
+        Self {
+            genomes: vec![0; size * 2 * n_loci],
+            sexes: vec![0; size],
+            n_loci,
+        }
+    }
+
+    fn size(&self) -> usize {
+        self.sexes.len()
+    }
+
+    fn genome_offset(&self, individual: usize, chrom: usize, locus: usize) -> usize {
+        individual * 2 * self.n_loci + chrom * self.n_loci + locus
+    }
+
+    fn allele(&self, individual: usize, chrom: usize, locus: usize) -> i16 {
+        self.genomes[self.genome_offset(individual, chrom, locus)]
+    }
+
+    fn set_allele(&mut self, individual: usize, chrom: usize, locus: usize, value: i16) {
+        let idx = self.genome_offset(individual, chrom, locus);
+        self.genomes[idx] = value;
+    }
+
+    fn female_count(&self) -> usize {
+        self.sexes.iter().filter(|&&sex| sex == 0).count()
+    }
+
+    fn male_count(&self) -> usize {
+        self.size().saturating_sub(self.female_count())
+    }
+
+    fn extinct(&self) -> bool {
+        self.size() == 0 || self.female_count() == 0 || self.male_count() == 0
+    }
+}
+
+#[derive(Debug, Clone, Copy)]
+struct StepSummary {
+    t: i32,
+    n: usize,
+    female_count: usize,
+    male_count: usize,
+    fecundity: f64,
+    mean_fitness: f64,
+    target_value: f64,
+    mean_allele_value: f64,
+    mean_genotype_value: f64,
+    mean_tracking_gap: f64,
+    extinct: bool,
+}
+
+impl NunneySimulationKernel {
+    pub fn new(seed: Option<u64>) -> Self {
+        Self { seed }
+    }
+
+    fn intrinsic_growth_rate(config: &Config) -> f64 {
+        (config.R / 2.0).ln()
+    }
+
+    fn t_int(config: &Config) -> i32 {
+        config.T.round().max(1.0) as i32
+    }
+
+    fn total_generations(config: &Config) -> usize {
+        let t = Self::t_int(config);
+        (config.epochs as i32 * t + t / 2).max(1) as usize
+    }
+
+    fn resolved_a_max(config: &Config) -> i16 {
+        config.epochs.max(1) as i16
+    }
+
+    fn initialize_population(config: &Config, rng: &mut StdRng) -> Population {
+        let mut population = Population::new(config.N0.max(0.0) as usize, config.n);
+        for sex in &mut population.sexes {
+            *sex = if rng.gen::<f64>() < 0.5 { 0 } else { 1 };
+        }
+        population
+    }
+
+    fn female_fecundity(r: f64, n_population: usize, k: f64) -> f64 {
+        if n_population == 0 || k <= 0.0 {
+            return 0.0;
+        }
+        let ratio = n_population as f64 / k;
+        2.0 * (r * (1.0 - ratio.powf(1.0 / r))).exp()
+    }
+
+    fn genotype_fitness(population: &Population, config: &Config, t: i32) -> Vec<f64> {
+        let r = Self::intrinsic_growth_rate(config);
+        let target = t as f64 / config.T;
+        let n = config.n as f64;
+        let mut out = Vec::with_capacity(population.size());
+        for individual in 0..population.size() {
+            let mut squared_distance_sum = 0.0;
+            for locus in 0..config.n {
+                let locus_mean =
+                    0.5 * (population.allele(individual, 0, locus) as f64
+                        + population.allele(individual, 1, locus) as f64);
+                let delta = locus_mean - target;
+                squared_distance_sum += delta * delta;
+            }
+            out.push((-(r / n) * squared_distance_sum).exp());
+        }
+        out
+    }
+
+    fn tracking_metrics(population: &Population, config: &Config, t: i32) -> (f64, f64, f64, f64) {
+        let target_value = t as f64 / config.T;
+        if population.size() == 0 {
+            return (target_value, 0.0, 0.0, -target_value);
+        }
+
+        let allele_sum: f64 = population.genomes.iter().map(|&v| v as f64).sum();
+        let mean_allele_value = allele_sum / population.genomes.len() as f64;
+
+        let mut genotype_sum = 0.0;
+        for individual in 0..population.size() {
+            for locus in 0..config.n {
+                genotype_sum += 0.5
+                    * (population.allele(individual, 0, locus) as f64
+                        + population.allele(individual, 1, locus) as f64);
+            }
+        }
+        let mean_genotype_value = genotype_sum / (population.size() * config.n) as f64;
+        let mean_tracking_gap = mean_genotype_value - target_value;
+        (
+            target_value,
+            mean_allele_value,
+            mean_genotype_value,
+            mean_tracking_gap,
+        )
+    }
+
+    fn summarize(population: &Population, config: &Config, t: i32) -> StepSummary {
+        if population.size() == 0 {
+            return StepSummary {
+                t,
+                n: 0,
+                female_count: 0,
+                male_count: 0,
+                fecundity: 0.0,
+                mean_fitness: 0.0,
+                target_value: t as f64 / config.T,
+                mean_allele_value: 0.0,
+                mean_genotype_value: 0.0,
+                mean_tracking_gap: -(t as f64 / config.T),
+                extinct: true,
+            };
+        }
+
+        let fitness = Self::genotype_fitness(population, config, t);
+        let fecundity = Self::female_fecundity(Self::intrinsic_growth_rate(config), population.size(), config.K);
+        let female_count = population.female_count();
+        let male_count = population.male_count();
+        let (target_value, mean_allele_value, mean_genotype_value, mean_tracking_gap) =
+            Self::tracking_metrics(population, config, t);
+
+        StepSummary {
+            t,
+            n: population.size(),
+            female_count,
+            male_count,
+            fecundity,
+            mean_fitness: fitness.iter().sum::<f64>() / fitness.len() as f64,
+            target_value,
+            mean_allele_value,
+            mean_genotype_value,
+            mean_tracking_gap,
+            extinct: population.extinct(),
+        }
+    }
+
+    fn sample_poisson(lambda: f64, rng: &mut StdRng) -> usize {
+        if lambda <= 0.0 {
+            return 0;
+        }
+        let dist = Poisson::new(lambda).expect("lambda must be positive");
+        dist.sample(rng) as usize
+    }
+
+    fn choose_mutant_allele(current: i16, a_max: i16, rng: &mut StdRng) -> i16 {
+        if a_max <= 0 {
+            return current;
+        }
+        let draw = rng.gen_range(0..a_max);
+        if draw < current { draw } else { draw + 1 }
+    }
+
+    fn produce_offspring(
+        population: &Population,
+        mother_index: usize,
+        father_index: usize,
+        config: &Config,
+        a_max: i16,
+        rng: &mut StdRng,
+    ) -> (Vec<i16>, usize) {
+        let mut offspring = vec![0i16; 2 * config.n];
+        let mut mutation_count = 0usize;
+
+        for locus in 0..config.n {
+            let maternal_chrom = rng.gen_range(0..2);
+            let paternal_chrom = rng.gen_range(0..2);
+            offspring[locus] = population.allele(mother_index, maternal_chrom, locus);
+            offspring[config.n + locus] = population.allele(father_index, paternal_chrom, locus);
+        }
+
+        for allele in &mut offspring {
+            if rng.gen::<f64>() <= config.u {
+                *allele = Self::choose_mutant_allele(*allele, a_max, rng);
+                mutation_count += 1;
+            }
+        }
+
+        (offspring, mutation_count)
+    }
+
+    fn offspring_fitness(offspring: &[i16], config: &Config, t: i32) -> f64 {
+        let r = Self::intrinsic_growth_rate(config);
+        let target = t as f64 / config.T;
+        let n = config.n as f64;
+        let mut squared_distance_sum = 0.0;
+        for locus in 0..config.n {
+            let locus_mean = 0.5 * (offspring[locus] as f64 + offspring[config.n + locus] as f64);
+            let delta = locus_mean - target;
+            squared_distance_sum += delta * delta;
+        }
+        (-(r / n) * squared_distance_sum).exp()
+    }
+
+    fn simulate_one_generation(
+        population: &Population,
+        config: &Config,
+        t: i32,
+        rng: &mut StdRng,
+    ) -> (Population, StepSummary, usize, usize) {
+        let summary = Self::summarize(population, config, t);
+        if summary.extinct {
+            return (population.clone(), summary, 0, 0);
+        }
+
+        let female_indices: Vec<usize> = population
+            .sexes
+            .iter()
+            .enumerate()
+            .filter_map(|(idx, &sex)| (sex == 0).then_some(idx))
+            .collect();
+        let male_indices: Vec<usize> = population
+            .sexes
+            .iter()
+            .enumerate()
+            .filter_map(|(idx, &sex)| (sex == 1).then_some(idx))
+            .collect();
+
+        let a_max = Self::resolved_a_max(config);
+        let mut next_genomes = Vec::new();
+        let mut next_sexes = Vec::new();
+        let mut realized_mutation_count = 0usize;
+        let mut birth_count = 0usize;
+        let offspring_t = t + 1;
+
+        for &mother_index in &female_indices {
+            let births = Self::sample_poisson(summary.fecundity, rng);
+            birth_count += births;
+            for _ in 0..births {
+                let father_index = male_indices[rng.gen_range(0..male_indices.len())];
+                let (offspring, mutation_count) = Self::produce_offspring(
+                    population,
+                    mother_index,
+                    father_index,
+                    config,
+                    a_max,
+                    rng,
+                );
+                realized_mutation_count += mutation_count;
+                let survival = Self::offspring_fitness(&offspring, config, offspring_t);
+                if rng.gen::<f64>() <= survival {
+                    next_genomes.extend_from_slice(&offspring);
+                    next_sexes.push(if rng.gen::<f64>() < 0.5 { 0 } else { 1 });
+                }
+            }
+        }
+
+        (
+            Population {
+                genomes: next_genomes,
+                sexes: next_sexes,
+                n_loci: config.n,
+            },
+            summary,
+            birth_count,
+            realized_mutation_count,
+        )
+    }
+}
+
+impl ExtinctionEstimator for NunneySimulationKernel {
+    fn run(&self, config: &Config) -> RunResult {
+        let seed = self.seed.or(config.seed).unwrap_or(0);
+        let mut rng = StdRng::seed_from_u64(seed);
+        let mut population = Self::initialize_population(config, &mut rng);
+        let t_half = Self::t_int(config) / 2;
+        let mut t = -t_half;
+        let mut last_summary = Self::summarize(&population, config, t);
+        let mut last_birth_count = 0usize;
+        let mut last_surviving_offspring_count = population.size();
+        let mut extinction_time = None;
+        let mut events = Vec::new();
+        let mut first_nonzero_allele_t = (last_summary.mean_allele_value != 0.0).then_some(last_summary.t);
+        let mut last_nonzero_allele_t = first_nonzero_allele_t;
+
+        for _ in 0..Self::total_generations(config) {
+            let (next_population, summary, birth_count, _mutation_count) =
+                Self::simulate_one_generation(&population, config, t, &mut rng);
+            last_summary = summary;
+            if last_summary.mean_allele_value != 0.0 {
+                if first_nonzero_allele_t.is_none() {
+                    first_nonzero_allele_t = Some(last_summary.t);
+                }
+                last_nonzero_allele_t = Some(last_summary.t);
+            }
+            last_birth_count = birth_count;
+            last_surviving_offspring_count = next_population.size();
+            population = next_population;
+            events.push(Event::GenerationComplete {
+                generation: t.max(0) as usize,
+                epoch: ((t.max(0) as f64) / config.T.max(1.0)).floor() as usize,
+            });
+            if population.extinct() {
+                let terminal_t = if last_summary.extinct { t } else { t + 1 };
+                last_summary = Self::summarize(&population, config, terminal_t);
+                if last_summary.mean_allele_value != 0.0 {
+                    if first_nonzero_allele_t.is_none() {
+                        first_nonzero_allele_t = Some(last_summary.t);
+                    }
+                    last_nonzero_allele_t = Some(last_summary.t);
+                }
+                extinction_time = Some(terminal_t.max(0) as usize);
+                break;
+            }
+            t += 1;
+        }
+
+        let active_loci = (0..config.n)
+            .filter(|&locus| {
+                let first = population
+                    .genomes
+                    .get(locus)
+                    .copied()
+                    .unwrap_or(0);
+                population
+                    .genomes
+                    .chunks_exact(2 * config.n)
+                    .any(|genome| genome[locus] != first || genome[config.n + locus] != first)
+            })
+            .count();
+
+        let final_state = State {
+            generation: last_summary.t.max(0) as usize,
+            epoch: ((last_summary.t.max(0) as f64) / config.T.max(1.0)).floor() as usize,
+            optimum: last_summary.target_value,
+            N: last_summary.n as f64,
+            N_f: last_summary.female_count as f64,
+            N_m: last_summary.male_count as f64,
+            total_tracked: population.size(),
+            extinctions: usize::from(last_summary.extinct),
+            extinction_time,
+            mean_fitness: last_summary.mean_fitness,
+            mean_fecundity: last_summary.fecundity,
+            mean_mismatch: last_summary.mean_tracking_gap.abs(),
+            mean_tracking_gap: last_summary.mean_tracking_gap,
+            birth_count: last_birth_count,
+            surviving_offspring_count: last_surviving_offspring_count,
+            allele_means: vec![last_summary.mean_allele_value; config.n],
+            active_loci,
+            termination_reason: if last_summary.extinct {
+                TerminationReason::Extinction
+            } else {
+                TerminationReason::Success
+            },
+        };
+
+        RunResult {
+            config: config.clone(),
+            seed,
+            final_state,
+            events,
+            wall_time: 0.01,
+            extinct: last_summary.extinct,
+            extinction_time,
+            first_nonzero_allele_t,
+            last_nonzero_allele_t,
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use rand::SeedableRng;
+
+    fn smoke_config() -> Config {
+        Config {
+            K: 1000.0,
+            N0: 500.0,
+            R: 10.0,
+            T: 50.0,
+            n: 3,
+            u: 0.001,
+            epochs: 5,
+            target_extinction_probability: 0.5,
+            seed: Some(0),
+        }
+    }
+
+    fn zero_population(size: usize, n_loci: usize) -> Population {
+        let mut population = Population::new(size, n_loci);
+        for (idx, sex) in population.sexes.iter_mut().enumerate() {
+            *sex = if idx % 2 == 0 { 0 } else { 1 };
+        }
+        population
+    }
+
+    #[test]
+    fn simple_simulation_kernel_creation() {
+        let kernel = SimpleSimulationKernel::new(false);
+        assert!(!kernel.deterministic);
+    }
+
+    #[test]
+    fn simple_simulation_kernel_deterministic() {
+        let kernel = SimpleSimulationKernel::new(true);
+        assert!(kernel.deterministic);
+    }
+
+    #[test]
+    fn simple_simulation_kernel_default() {
+        let kernel = SimpleSimulationKernel::default();
+        assert!(!kernel.deterministic);
+    }
+
+    #[test]
+    fn nunney_simulation_kernel_creation() {
+        let kernel = NunneySimulationKernel::new(Some(42));
+        assert_eq!(kernel.seed, Some(42));
+    }
+
+    #[test]
+    fn nunney_simulation_kernel_default() {
+        let kernel = NunneySimulationKernel::default();
+        assert_eq!(kernel.seed, None);
+    }
+
+    #[test]
+    fn female_fecundity_matches_closed_form() {
+        let config = smoke_config();
+        let r = NunneySimulationKernel::intrinsic_growth_rate(&config);
+        let observed = NunneySimulationKernel::female_fecundity(r, 500, 1000.0);
+        let expected = 2.0 * (r * (1.0 - (0.5_f64).powf(1.0 / r))).exp();
+        assert!((observed - expected).abs() < 1e-12);
+    }
+
+    #[test]
+    fn genotype_fitness_is_one_at_target_and_decreases_with_distance() {
+        let config = Config { n: 1, ..smoke_config() };
+        let mut population = Population::new(2, config.n);
+        population.sexes = vec![0, 1];
+        population.genomes = vec![0, 0, 10, 10];
+        let near = NunneySimulationKernel::genotype_fitness(&population, &config, 0);
+        assert!((near[0] - 1.0).abs() < 1e-12);
+        assert!(near[1] < near[0]);
+    }
+
+    #[test]
+    fn choose_mutant_allele_never_returns_current_and_respects_bound() {
+        let mut rng = StdRng::seed_from_u64(7);
+        for _ in 0..200 {
+            let allele = NunneySimulationKernel::choose_mutant_allele(2, 5, &mut rng);
+            assert_ne!(allele, 2);
+            assert!((0..=5).contains(&allele));
+        }
+    }
+
+    #[test]
+    fn population_extinction_requires_zero_population_or_missing_sex() {
+        let empty = Population::new(0, 1);
+        assert!(empty.extinct());
+
+        let mut female_only = Population::new(3, 1);
+        female_only.sexes = vec![0, 0, 0];
+        assert!(female_only.extinct());
+
+        let mixed = zero_population(4, 1);
+        assert!(!mixed.extinct());
+    }
+
+    #[test]
+    fn total_generations_and_initial_t_match_track1_convention() {
+        let config = smoke_config();
+        assert_eq!(NunneySimulationKernel::total_generations(&config), 275);
+        assert_eq!(NunneySimulationKernel::t_int(&config), 50);
+    }
+
+    #[test]
+    fn simulate_one_generation_bookkeeping_is_internally_consistent() {
+        let config = Config { n: 1, ..smoke_config() };
+        let mut rng = StdRng::seed_from_u64(0);
+        let population = NunneySimulationKernel::initialize_population(&config, &mut rng);
+        let (next_population, summary, birth_count, mutation_count) =
+            NunneySimulationKernel::simulate_one_generation(&population, &config, 0, &mut rng);
+        assert_eq!(summary.n, population.size());
+        assert_eq!(summary.female_count + summary.male_count, population.size());
+        assert!(birth_count >= next_population.size());
+        assert!(mutation_count <= 2 * birth_count * config.n);
+    }
+
+    #[test]
+    fn offspring_inherits_only_parental_alleles_when_mutation_is_zero() {
+        let config = Config {
+            n: 1,
+            u: 0.0,
+            ..smoke_config()
+        };
+        let mut population = Population::new(2, config.n);
+        population.sexes = vec![0, 1];
+        population.genomes = vec![1, 2, 7, 8];
+        let mut rng = StdRng::seed_from_u64(3);
+        let (offspring, mutation_count) =
+            NunneySimulationKernel::produce_offspring(&population, 0, 1, &config, 5, &mut rng);
+        assert_eq!(mutation_count, 0);
+        assert!(matches!(offspring[0], 1 | 2));
+        assert!(matches!(offspring[1], 7 | 8));
+    }
+
+    #[test]
+    fn repeated_offspring_generation_has_zero_mutations_when_u_is_zero() {
+        let config = Config {
+            n: 2,
+            u: 0.0,
+            ..smoke_config()
+        };
+        let mut population = Population::new(2, config.n);
+        population.sexes = vec![0, 1];
+        population.genomes = vec![1, 2, 3, 4, 7, 8, 9, 10];
+        let mut rng = StdRng::seed_from_u64(11);
+        for _ in 0..100 {
+            let (offspring, mutation_count) =
+                NunneySimulationKernel::produce_offspring(&population, 0, 1, &config, 5, &mut rng);
+            assert_eq!(mutation_count, 0);
+            for locus in 0..config.n {
+                assert!(matches!(offspring[locus], 1 | 2 | 3 | 4));
+                assert!(matches!(offspring[config.n + locus], 7 | 8 | 9 | 10));
+            }
+        }
+    }
+
+    #[test]
+    fn summarize_empty_population_reports_extinction_and_negative_gap() {
+        let config = Config { n: 1, ..smoke_config() };
+        let population = Population::new(0, config.n);
+        let summary = NunneySimulationKernel::summarize(&population, &config, 10);
+        assert!(summary.extinct);
+        assert_eq!(summary.n, 0);
+        assert_eq!(summary.mean_allele_value, 0.0);
+        assert_eq!(summary.mean_tracking_gap, -(10.0 / config.T));
+    }
+
+    #[test]
+    fn immediate_missing_sex_extinction_produces_terminal_result() {
+        let config = Config {
+            N0: 1.0,
+            n: 1,
+            u: 0.0,
+            epochs: 1,
+            T: 10.0,
+            seed: Some(0),
+            ..smoke_config()
+        };
+        let kernel = NunneySimulationKernel::new(Some(0));
+        let result = kernel.run(&config);
+        assert!(result.extinct);
+        assert_eq!(result.final_state.termination_reason, TerminationReason::Extinction);
+        assert_eq!(result.final_state.N, 1.0);
+        assert!(result.final_state.N_f == 0.0 || result.final_state.N_m == 0.0);
+        assert!(result.extinction_time.is_some());
+    }
+
+    #[test]
+    fn offspring_survival_is_monotone_with_distance_from_target() {
+        let config = Config {
+            n: 1,
+            T: 10.0,
+            ..smoke_config()
+        };
+        let exact = vec![5, 5];
+        let near = vec![4, 4];
+        let far = vec![0, 0];
+        let exact_fit = NunneySimulationKernel::offspring_fitness(&exact, &config, 50);
+        let near_fit = NunneySimulationKernel::offspring_fitness(&near, &config, 50);
+        let far_fit = NunneySimulationKernel::offspring_fitness(&far, &config, 50);
+        assert!((exact_fit - 1.0).abs() < 1e-12);
+        assert!(exact_fit > near_fit);
+        assert!(near_fit > far_fit);
+    }
+
+    #[test]
+    fn run_result_generation_and_epoch_are_terminal_summary_values() {
+        let kernel = NunneySimulationKernel::new(Some(0));
+        let config = smoke_config();
+        let result = kernel.run(&config);
+        assert_eq!(result.generations(), result.final_state.generation);
+        assert_eq!(result.epochs(), result.final_state.epoch);
+        assert!(result.final_state.generation <= NunneySimulationKernel::total_generations(&config));
+    }
+
+    #[test]
+    fn non_extinction_run_finishes_at_expected_terminal_generation() {
+        let kernel = NunneySimulationKernel::new(Some(0));
+        let config = smoke_config();
+        let result = kernel.run(&config);
+        if !result.extinct {
+            let expected_terminal_t = (NunneySimulationKernel::total_generations(&config) as i32 - 1)
+                - (NunneySimulationKernel::t_int(&config) / 2);
+            assert_eq!(
+                result.final_state.generation as i32,
+                expected_terminal_t
+            );
+        }
+    }
+
+    #[test]
+    fn run_result_bookkeeping_fields_are_consistent() {
+        let kernel = NunneySimulationKernel::new(Some(0));
+        let config = smoke_config();
+        let result = kernel.run(&config);
+        assert_eq!(result.final_state.birth_count >= result.final_state.surviving_offspring_count, true);
+        assert_eq!(result.final_state.mean_mismatch, result.final_state.mean_tracking_gap.abs());
+        if let (Some(first), Some(last)) = (result.first_nonzero_allele_t, result.last_nonzero_allele_t) {
+            assert!(first <= last);
+        }
+    }
+
+    #[test]
+    fn nunney_simulation_kernel_matches_smoke_shape() {
+        let kernel = NunneySimulationKernel::new(Some(0));
+        let config = smoke_config();
+        let result = kernel.run(&config);
+        assert!(!result.extinct);
+        assert!(result.final_state.N > 0.0);
+        assert!(result.final_state.mean_mismatch < 1.0);
+        assert!(result.last_nonzero_allele_t.is_some());
+    }
+}
--- a/scripts/compare_track1_rust_smoke.py
+++ b/scripts/compare_track1_rust_smoke.py
@ -0,0 +1,242 @@
+#!/usr/bin/env python3
+"""
+Compare the Python Track 1 run summary against the Rust smoke binary over a
+small seed range.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import subprocess
+from pathlib import Path
+from statistics import mean, pstdev
+
+from renunney.track1_analysis import summarize_tracking
+from renunney.track1_reference import Track1Parameters, simulate_run
+
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+RUST_CRATE = REPO_ROOT / "rust" / "track2-core"
+RUST_TARGET = Path("/tmp/renunney-track2-compare")
+RUST_BIN = RUST_TARGET / "release" / "examples" / "smoke_compare"
+
+
+def build_rust_smoke_binary() -> None:
+    subprocess.run(
+        [
+            "cargo",
+            "build",
+            "--release",
+            "--example",
+            "smoke_compare",
+            "--target-dir",
+            str(RUST_TARGET),
+        ],
+        cwd=RUST_CRATE,
+        check=True,
+    )
+
+
+def parse_rust_output(text: str) -> dict[str, object]:
+    parsed: dict[str, object] = {}
+    for line in text.splitlines():
+        if "=" not in line:
+            continue
+        key, raw = line.split("=", 1)
+        raw = raw.strip()
+        if raw in {"true", "false"}:
+            parsed[key] = raw == "true"
+        elif raw.startswith("Some(") and raw.endswith(")"):
+            parsed[key] = int(raw[5:-1])
+        elif raw == "None":
+            parsed[key] = None
+        else:
+            try:
+                parsed[key] = int(raw)
+            except ValueError:
+                parsed[key] = float(raw)
+    return parsed
+
+
+def run_rust(seed: int, params: Track1Parameters) -> dict[str, object]:
+    proc = subprocess.run(
+        [
+            str(RUST_BIN),
+            "--K",
+            str(params.K),
+            "--N0",
+            str(params.N0),
+            "--R",
+            str(params.R),
+            "--T",
+            str(params.T),
+            "--n",
+            str(params.n),
+            "--u",
+            str(params.u),
+            "--epochs",
+            str(params.epochs),
+            "--seed",
+            str(seed),
+        ],
+        cwd=RUST_CRATE,
+        check=True,
+        capture_output=True,
+        text=True,
+    )
+    return parse_rust_output(proc.stdout)
+
+
+def run_python(seed: int, params: Track1Parameters) -> dict[str, object]:
+    summaries = simulate_run(params, seed=seed)
+    final = summaries[-1]
+    tracking = summarize_tracking(summaries)
+    return {
+        "extinct": bool(final.extinct),
+        "generation": int(final.t),
+        "N": int(final.N),
+        "female_count": int(final.female_count),
+        "male_count": int(final.male_count),
+        "target_value": float(final.target_value),
+        "mean_allele_value": float(final.mean_allele_value),
+        "mean_fitness": float(final.mean_fitness),
+        "fecundity": float(final.fecundity),
+        "mean_tracking_gap": float(final.mean_tracking_gap),
+        "birth_count": int(final.birth_count),
+        "surviving_offspring_count": int(final.surviving_offspring_count),
+        "first_nonzero_allele_t": tracking.first_nonzero_allele_t,
+        "last_nonzero_allele_t": tracking.last_nonzero_allele_t,
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description="Compare Python Track 1 and Rust smoke outputs over multiple seeds.")
+    parser.add_argument("--seed-start", type=int, default=0)
+    parser.add_argument("--seed-count", type=int, default=10)
+    parser.add_argument("--K", type=int, default=1000)
+    parser.add_argument("--N0", type=int, default=500)
+    parser.add_argument("--n", type=int, default=3)
+    parser.add_argument("--u", type=float, default=0.001)
+    parser.add_argument("--R", type=float, default=10.0)
+    parser.add_argument("--T", type=int, default=50)
+    parser.add_argument("--epochs", type=int, default=5)
+    args = parser.parse_args()
+
+    build_rust_smoke_binary()
+    params = Track1Parameters(
+        K=args.K,
+        N0=args.N0,
+        n=args.n,
+        u=args.u,
+        R=args.R,
+        T=args.T,
+        epochs=args.epochs,
+    )
+
+    rows: list[dict[str, object]] = []
+    for seed in range(args.seed_start, args.seed_start + args.seed_count):
+        py = run_python(seed, params)
+        rs = run_rust(seed, params)
+        rows.append(
+            {
+                "seed": seed,
+                "python": py,
+                "rust": rs,
+                "delta": {
+                    "N": float(rs["N"]) - float(py["N"]),
+                    "mean_allele_value": float(rs["mean_allele_value"]) - float(py["mean_allele_value"]),
+                    "mean_tracking_gap": float(rs["mean_tracking_gap"]) - float(py["mean_tracking_gap"]),
+                    "birth_count": float(rs["birth_count"]) - float(py["birth_count"]),
+                    "surviving_offspring_count": float(rs["surviving_offspring_count"]) - float(py["surviving_offspring_count"]),
+                },
+                "extinct_match": bool(rs["extinct"]) == bool(py["extinct"]),
+            }
+        )
+
+    summary = {
+        "parameters": {
+            "K": params.K,
+            "N0": params.N0,
+            "n": params.n,
+            "u": params.u,
+            "R": params.R,
+            "T": params.T,
+            "epochs": params.epochs,
+        },
+        "seed_start": args.seed_start,
+        "seed_count": args.seed_count,
+        "extinct_match_rate": mean(1.0 if row["extinct_match"] else 0.0 for row in rows),
+        "mean_abs_delta": {
+            "N": mean(abs(row["delta"]["N"]) for row in rows),
+            "mean_allele_value": mean(abs(row["delta"]["mean_allele_value"]) for row in rows),
+            "mean_tracking_gap": mean(abs(row["delta"]["mean_tracking_gap"]) for row in rows),
+            "birth_count": mean(abs(row["delta"]["birth_count"]) for row in rows),
+            "surviving_offspring_count": mean(abs(row["delta"]["surviving_offspring_count"]) for row in rows),
+        },
+        "aggregate": {
+            "python": {
+                "N": {
+                    "mean": mean(float(row["python"]["N"]) for row in rows),
+                    "sd": pstdev(float(row["python"]["N"]) for row in rows),
+                },
+                "birth_count": {
+                    "mean": mean(float(row["python"]["birth_count"]) for row in rows),
+                    "sd": pstdev(float(row["python"]["birth_count"]) for row in rows),
+                },
+                "surviving_offspring_count": {
+                    "mean": mean(float(row["python"]["surviving_offspring_count"]) for row in rows),
+                    "sd": pstdev(float(row["python"]["surviving_offspring_count"]) for row in rows),
+                },
+                "mean_allele_value": {
+                    "mean": mean(float(row["python"]["mean_allele_value"]) for row in rows),
+                    "sd": pstdev(float(row["python"]["mean_allele_value"]) for row in rows),
+                },
+                "mean_tracking_gap": {
+                    "mean": mean(float(row["python"]["mean_tracking_gap"]) for row in rows),
+                    "sd": pstdev(float(row["python"]["mean_tracking_gap"]) for row in rows),
+                },
+            },
+            "rust": {
+                "N": {
+                    "mean": mean(float(row["rust"]["N"]) for row in rows),
+                    "sd": pstdev(float(row["rust"]["N"]) for row in rows),
+                },
+                "birth_count": {
+                    "mean": mean(float(row["rust"]["birth_count"]) for row in rows),
+                    "sd": pstdev(float(row["rust"]["birth_count"]) for row in rows),
+                },
+                "surviving_offspring_count": {
+                    "mean": mean(float(row["rust"]["surviving_offspring_count"]) for row in rows),
+                    "sd": pstdev(float(row["rust"]["surviving_offspring_count"]) for row in rows),
+                },
+                "mean_allele_value": {
+                    "mean": mean(float(row["rust"]["mean_allele_value"]) for row in rows),
+                    "sd": pstdev(float(row["rust"]["mean_allele_value"]) for row in rows),
+                },
+                "mean_tracking_gap": {
+                    "mean": mean(float(row["rust"]["mean_tracking_gap"]) for row in rows),
+                    "sd": pstdev(float(row["rust"]["mean_tracking_gap"]) for row in rows),
+                },
+            },
+            "delta_of_means": {
+                "N": mean(float(row["rust"]["N"]) for row in rows)
+                - mean(float(row["python"]["N"]) for row in rows),
+                "birth_count": mean(float(row["rust"]["birth_count"]) for row in rows)
+                - mean(float(row["python"]["birth_count"]) for row in rows),
+                "surviving_offspring_count": mean(float(row["rust"]["surviving_offspring_count"]) for row in rows)
+                - mean(float(row["python"]["surviving_offspring_count"]) for row in rows),
+                "mean_allele_value": mean(float(row["rust"]["mean_allele_value"]) for row in rows)
+                - mean(float(row["python"]["mean_allele_value"]) for row in rows),
+                "mean_tracking_gap": mean(float(row["rust"]["mean_tracking_gap"]) for row in rows)
+                - mean(float(row["python"]["mean_tracking_gap"]) for row in rows),
+            },
+        },
+        "rows": rows,
+    }
+    print(json.dumps(summary, indent=2, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())