Evolutionary Chain-of-Thought Methods

• Evolutionary Chain-of-Thought is a framework that treats multi-step reasoning as an evolvable process using chaining and contextual focus.
• It employs iterative variation, evaluation, and selection to adapt reasoning trajectories, demonstrating significant improvements in fitness and diversity in simulated models.
• Advanced implementations utilize genetic operators like crossover, mutation, and reflective recombination to optimize reasoning prompts and enable effective scientific distillation.

Searching arXiv for the cited papers to ground the article in the current record. Search query: (Gabora et al., 2013) evolutionary chain-of-thought EVOC contextual focus Evolutionary Chain-of-Thought denotes a family of computational approaches in which a thought sequence is treated as an evolvable object rather than a single fixed inference. In the earliest formulation considered here, EVOC models the emergence of a “stream of thought” through chaining, followed by contextual focus (CF), the capacity to shift between divergent and convergent modes of thought during cultural evolution (Gabora et al., 2013). In later large-language-model work, the same general motif appears as evolutionary search over CoT prompts, self-evolving populations of reasoning trajectories, and evolutionary distillation of CoTs from multiple thinkers for scientific reasoning (Jin et al., 2024, Wang et al., 16 Apr 2026, Feng et al., 15 Oct 2025). This suggests an umbrella category rather than a single standardized formalism: the common structure is iterative variation, evaluation, and selection over intermediate reasoning artifacts.

1. Research lineage and object of evolution

Across the literature, “evolutionary” CoT methods differ primarily in what is being evolved. EVOC evolves multi-step actions in an agent-based model of cultural evolution; EoT evolves prompt strings per instance; CoTEvol evolves full reasoning trajectories as individuals in a genetic algorithm; and CoT-Evo evolves candidate CoTs produced by multiple LLM thinkers under novelty-driven selection and reflective refinement (Gabora et al., 2013, Jin et al., 2024, Wang et al., 16 Apr 2026, Feng et al., 15 Oct 2025).

Work	Object evolved	Core evolutionary machinery
EVOC	Action sequences	chaining, CF
EoT	CoT prompts	crossover, mutation, rewriting
CoTEvol	Reasoning trajectories	reflective global crossover, uncertainty-guided local mutation
CoT-Evo	Distilled CoTs	novelty-driven selection, reflective recombination, mutation

The historical sequence is also conceptually significant. EVOC frames chaining as a cognitive transition enabling open-ended cultural evolution. The later LLM systems transpose that logic into explicit optimization over prompts or reasoning traces. A plausible implication is that the phrase “evolutionary chain-of-thought” spans both a cognitive model of recursive idea extension and a family of engineering methods that operationalize CoT search with genetic-algorithm primitives.

2. Chaining in EVOC: formal action space and recursive extension

In EVOC, an elementary idea is a sub-action represented by a six-dimensional discrete vector

$D = (d_1,d_2,d_3,d_4,d_5,d_6),$

with each component $d_j \in \{-1,0,+1\}$ encoding the posture of one of six body parts: HD, LA/RA, LL/RL, and HP. An action is a finite sequence of sub-actions,

$A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$

and the total action space is the Kleene-closure

$S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$

The model therefore treats behavior as recursively extensible rather than single-step (Gabora et al., 2013).

Fitness is defined on a multi-peaked “rugged” landscape using 45 Royal-Road templates $T^i=(t_1^i,\dots,t_6^i)$ with $t_j^i \in \{-1,0,+1,*\}$. For a sub-action $D$, the template match indicator is

$$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$

and template order is

$\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$

Sub-action fitness is then

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$

For a valid chained action $d_j \in \{-1,0,+1\}$0, the chained fitness is

$d_j \in \{-1,0,+1\}$1

that is, the fitness of the last sub-action plus the number of sub-actions (Gabora et al., 2013).

The chaining criterion requires each new sub-action to be both novel and successful:

$d_j \in \{-1,0,+1\}$2

$d_j \in \{-1,0,+1\}$3

Operationally, an agent begins from its current idea, generates a candidate by modifying the last element of the current action, appends the candidate only if it is novel and successful, and otherwise terminates the chain. In practice, at each invention step the current action is fed back into the agent’s neural network, body-part postures are probabilistically flipped according to learned biases favoring symmetry or movement, and the loop is applied again (Gabora et al., 2013).

This formalization makes “stream of thought” computationally explicit. Rather than representing invention as a single local perturbation, EVOC represents it as recursive elaboration constrained by minimal success criteria. A plausible implication is that chaining increases the effective dimensionality of cultural variation because the search object is no longer a single vector but an unbounded sequence.

3. Contextual focus in EVOC and the emergence of open-ended innovation

EVOC supplements chaining with contextual focus (CF), defined as the capacity to shift between a convergent (analytic) mode involving small, local modifications and a divergent (associative) mode involving large, global leaps. The control variable is the Rate of Creative Change (RCC), which determines how many components of a sub-action are flipped in one invention attempt. RCC is adapted by

$d_j \in \{-1,0,+1\}$4

followed by

$d_j \in \{-1,0,+1\}$5

with $d_j \in \{-1,0,+1\}$6 and, in the reported runs, $d_j \in \{-1,0,+1\}$7. Initialization is

$d_j \in \{-1,0,+1\}$8

with $d_j \in \{-1,0,+1\}$9. When fitness declines, RCC increases; when fitness improves, RCC decreases. With CF off, RCC is held fixed at $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$0, so on average one body part is modified per invention (Gabora et al., 2013).

The reported simulations used a population of $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$1 agents on a $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$2 toroidal grid, each with 8 neighbors, over $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$3 generations. At each generation every agent attempts invention; if the invented action’s fitness exceeds its current fitness, it adopts it, otherwise it attempts to imitate one randomly scanned neighbor whose fitness exceeds its own. Neural-network biases learn trends in symmetry and movement over time. All curves are averages over 500 independent runs (Gabora et al., 2013).

Quantitatively, mean fitness at $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$4 was approximately

$A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$5

Action diversity $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$6 was reported as follows:

Condition	$A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$7	$A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$8
None	1.2	1.0
Chain	12.5	15.2
Chain + CF	14.0	15.3

The qualitative pattern is equally central. Without chaining, mean fitness quickly plateaus and diversity collapses to a single fixed action. With chaining alone, mean fitness grows roughly linearly and diversity remains high. Adding CF has little effect on final plateau fitness or diversity once the landscape is stable, but it speeds early exploration and re-adaptation sharply when the fitness function is switched at $A = [D^1,D^2,\dots,D^n], \quad n \ge 1,$9 (Gabora et al., 2013). This supports the paper’s interpretation of chaining as a mechanism for open-ended innovation and of CF as a flexible exploration–exploitation controller that becomes especially valuable under environmental or task change.

4. Evolutionary prompting for zero-shot reasoning

“Zero-Shot Chain-of-Thought Reasoning Guided by Evolutionary Algorithms” introduces EoT, a per-instance evolutionary method for prompt search in which a small population of candidate CoT prompts is evolved in one round (Jin et al., 2024). The initial population contains two seed prompts, for example “Let’s think step by step.” and a PS+ prompt. The method then applies LLM_Crossover to the two seeds, LLM_Mutation to the crossover output, evaluates all candidates on the specific question, selects the best prompt, rewrites the question in light of that prompt, and performs final reasoning and answer extraction.

Formally, for a candidate prompt $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$0 and question $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$1, fitness is defined by

$S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$2

The selected prompt is

$S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$3

with an optional stochastic selection rule

$S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$4

The rewriting operation is formalized by replacing $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$5 with a rewritten $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$6 conditioned on the selected prompt $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$7 before final reasoning:

$S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$8

The method therefore evolves guidance strings rather than full reasoning traces (Jin et al., 2024).

The empirical evaluation covers ten reasoning datasets: arithmetic, commonsense, and symbolic tasks. On GPT-3.5-turbo arithmetic reasoning, EoT attains an average accuracy of 83.5, exceeding Zero-shot CoT (80.7), Zero-shot PS+ (81.2), Zero-shot RE2 (82.3), Few-shot Manual-CoT (82.2), and narrowly exceeding Few-shot AuTo-CoT (83.4). On GPT-3.5-turbo commonsense and symbolic tasks, EoT reaches 72.7 on CommonsenseQA, 69.9 on StrategyQA, 76.8 on Last Letters, and 98.9 on Coin Flip. On GPT-4 arithmetic sets, EoT reports 75.9 on AQuA, 97.7 on AddSub, and 92.7 on SVAMP (Jin et al., 2024).

Ablation results attribute the highest accuracy to the full combination of rewrite, crossover, and mutation. Removing rewrite produces a small drop of approximately 0.5–1.1 points, while removing crossover or mutation yields larger drops of approximately 3–4 and 3–6 points, respectively. The paper also reports that larger population size improves accuracy at the price of additional inference calls, and that EoT with self-consistency outperforms CoT+SC, PS+SC, and RE2+SC by 2–3 points on AddSub, AQuA, SingleEq, and SVAMP (Jin et al., 2024). In this setting, evolutionary CoT is an adaptive prompting strategy rather than a training procedure.

5. Population-based search over reasoning trajectories

“CoTEvol: Self-Evolving Chain-of-Thoughts for Data Synthesis in Mathematical Reasoning” casts CoT generation as a population-based search over reasoning trajectories (Wang et al., 16 Apr 2026). For each math question $S = \bigcup_{n=1}^{\infty} \{-1,0,+1\}^6.$9, a population $T^i=(t_1^i,\dots,t_6^i)$0 of $T^i=(t_1^i,\dots,t_6^i)$1 candidate CoTs is maintained. Each individual

$T^i=(t_1^i,\dots,t_6^i)$2

is a sequence of reasoning steps, where each step $T^i=(t_1^i,\dots,t_6^i)$3 is a contiguous span of tokens indexed by $T^i=(t_1^i,\dots,t_6^i)$4. The paper explicitly identifies the genome as the entire token sequence of $T^i=(t_1^i,\dots,t_6^i)$5 and a gene as a single reasoning step $T^i=(t_1^i,\dots,t_6^i)$6.

Its reflective global crossover recombines two parent trajectories $T^i=(t_1^i,\dots,t_6^i)$7 and $T^i=(t_1^i,\dots,t_6^i)$8 into an offspring $T^i=(t_1^i,\dots,t_6^i)$9 using self-reflection feedback. The answer-correctness label is

$t_j^i \in \{-1,0,+1,*\}$0

Three feedback cases are used: Elite Merging when both parents are correct, Success–Error Fusion when one is correct and one is not, and Failure Pattern Summary when both are incorrect. If $t_j^i \in \{-1,0,+1,*\}$1 is the critic-generation operator and $t_j^i \in \{-1,0,+1,*\}$2 the LLM generation operator, then

$t_j^i \in \{-1,0,+1,*\}$3

The crossover therefore operates at the trajectory level rather than only at the token or prompt level (Wang et al., 16 Apr 2026).

The uncertainty-guided local mutation identifies the most uncertain reasoning step by token-level entropy

$t_j^i \in \{-1,0,+1,*\}$4

step-level aggregation

$t_j^i \in \{-1,0,+1,*\}$5

and selection of

$t_j^i \in \{-1,0,+1,*\}$6

Only the top-uncertainty step is mutated, with

$t_j^i \in \{-1,0,+1,*\}$7

The mutation temperature is

$t_j^i \in \{-1,0,+1,*\}$8

after which the prefix is frozen and the uncertain step, and optionally subsequent steps, are resampled (Wang et al., 16 Apr 2026).

Fitness is lightweight and task-aware:

$t_j^i \in \{-1,0,+1,*\}$9

Here $D$0 if the final answer is $D$1 and $D$2 otherwise; $D$3 depends on token length $D$4 relative to the maximum length $D$5 in the population, with separate cosine-shaped rewards for correct and incorrect trajectories. Typical hyperparameters are $D$6, $D$7, $D$8, and $D$9 (Wang et al., 16 Apr 2026).

The overall workflow initializes $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$0 solutions at temperature $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$1, filters duplicates with ROUGE-L $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$2, evaluates fitness, selects $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$3 parents via Boltzmann-tournament, creates one crossover child and one mutation child, retains the top-$$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$4 individuals, and terminates after $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$5 iterations or saturation. Reported implementation details include Qwen2.5-7B-Instruct as base model, $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$6, $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$7, $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$8, $$\Phi(T^i,D)= \begin{cases} 1 & \text{if for all } j,\ (t_j^i=d_j)\ \text{or}\ (t_j^i=*)\ 0 & \text{otherwise,} \end{cases}$$9, $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$0, max output length 2048 tokens, fallback to a distilled D-CoT trace in fewer than 7\% of failed evolutions, and SFT via OpenRLHF on 4 $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$1 NVIDIA H100 (80 GB) with CUDA 12.4 (Wang et al., 16 Apr 2026).

Empirically, on S1K the correct-CoT synthesis success rate improves from 0.359 to 0.825 for CoTEvol w/GT, described as +130\% relative; +46 pp. Offspring exhibit >20\% higher inter-trajectory edit distance than Best-of-N. A fine-tuned model based on Qwen2.5-7B improves from 46.4\% baseline accuracy by +6.8 pp on S1K and +6.3 pp on LIMO, for an overall average gain of 6.6\% across eight math benchmarks. The reported efficiency in FLOPs $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$2 is 1689.5 for Best-of-N, 733.9 for Self-Refine, and 453.8 for CoTEvol (Wang et al., 16 Apr 2026). In this formulation, evolutionary CoT is a data-synthesis and trajectory-optimization procedure designed to produce high-quality training traces at lower compute cost than brute-force sampling or iterative self-refinement.

6. Evolutionary distillation for scientific reasoning

“CoT-Evo: Evolutionary Distillation of Chain-of-Thought for Scientific Reasoning” targets a different regime: scientific reasoning, where direct CoT distillation from advanced models is described as unreliable because even strong teachers can generate incorrect or superficial explanations (Feng et al., 15 Oct 2025). CoT-Evo therefore initializes a diverse pool of candidate CoTs by querying multiple LLM thinkers $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$3:

$\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$4

and then augments some thinkers with automatically retrieved domain knowledge $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$5:

$\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$6

The resulting pool $\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$7 contains both pure CoTs and knowledge-augmented CoTs.

The composite fitness function is

$\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$8

Here

$\Omega(T^i)=\sum_{j=1}^{6} I[t_j^i \ne *].$9

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$0

with $F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$1 and $F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$2 set to the 15% and 85% token-length percentiles, and

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$3

This combines exact-match correctness, reasoning-length appropriateness, and knowledge-usage correctness (Feng et al., 15 Oct 2025).

Selection is novelty-driven. Each CoT is embedded as

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$4

its novelty score is the mean distance to its $F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$5 nearest neighbors,

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$6

and its local-competition score is

$F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$7

The Pareto front $F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$8 is formed over the bi-objective vector $F(D)=\sum_{i=1}^{45}\Phi(T^i,D)\cdot \Omega(T^i).$9, and parents are sampled with probability

$d_j \in \{-1,0,+1\}$00

This prevents selection from collapsing onto only the highest-fitness but behaviorally similar traces (Feng et al., 15 Oct 2025).

Offspring are produced by either reflective recombination or reflective mutation. Recombination is triggered only if the parent is incorrect, $d_j \in \{-1,0,+1\}$01; it identifies a binding point $d_j \in \{-1,0,+1\}$02, extracts informative snippets $d_j \in \{-1,0,+1\}$03 from a strategy provider, and forms

$d_j \in \{-1,0,+1\}$04

Mutation uses three modes:

$d_j \in \{-1,0,+1\}$05

$d_j \in \{-1,0,+1\}$06

$d_j \in \{-1,0,+1\}$07

The full algorithm runs for up to $d_j \in \{-1,0,+1\}$08 generations with population size $d_j \in \{-1,0,+1\}$09; in practice, $d_j \in \{-1,0,+1\}$10 and $d_j \in \{-1,0,+1\}$11 (Feng et al., 15 Oct 2025).

After evolution, the distilled dataset

$d_j \in \{-1,0,+1\}$12

is used to fine-tune compact LLMs including Qwen3-8B, Qwen2.5-7B-Instruct, and Llama3.1-8B-Instruct. The reported recipe uses LLaMA-Factory, DeepSpeed ZeRO-2, FlashAttention2, AdamW with $d_j \in \{-1,0,+1\}$13, $d_j \in \{-1,0,+1\}$14, weight decay 0.1, peak learning rate $d_j \in \{-1,0,+1\}$15 with 10% warm-up and cosine decay, batch size 32, max sequence length 16,384, and 5 epochs on 4 $d_j \in \{-1,0,+1\}$16 A100 GPUs (Feng et al., 15 Oct 2025).

Evaluation is conducted on BioProBench and ChemCoTBench. For a Qwen3-8B student, the reported distilled performance is 0.649 on BioProBench PQA accuracy, 0.351 on ChemCoTBench understanding MAE, and 0.625 on Edit accuracy, compared with ST values of 0.508, 0.461, and 0.600, MT values of 0.603, 0.395, and 0.623, and BoK values of 0.603, 0.395, and 0.600. Relative to ST and MT, the paper reports up to 27\% error reduction on ChemCoTBench subtasks and 12.6\% PQA accuracy improvement on BioProBench. Ablations indicate that both recombination and mutation are essential, and that novelty-driven selection prevents premature convergence (Feng et al., 15 Oct 2025).

7. Conceptual issues, misconceptions, and open directions

A recurring misconception is that evolutionary CoT refers only to prompt search. The record surveyed here is broader: EVOC evolves multi-step actions through chaining and CF; EoT evolves prompts; CoTEvol evolves complete reasoning trajectories; and CoT-Evo evolves candidate CoTs for distillation (Gabora et al., 2013, Jin et al., 2024, Wang et al., 16 Apr 2026, Feng et al., 15 Oct 2025). The shared pattern is not a single representation but a common search logic over intermediate cognitive or linguistic structures.

A second misconception is that diversity alone explains performance. The methods consistently combine diversity with explicit control signals. In EVOC, CF is only effective when the fitness function changed; in CoTEvol, fitness combines answer correctness, format matching, and length-based reward; in CoT-Evo, novelty is coupled to local competition rather than used in isolation (Gabora et al., 2013, Wang et al., 16 Apr 2026, Feng et al., 15 Oct 2025). This suggests that evolutionary CoT systems are best understood as structured exploration mechanisms, not as unconstrained diversification.

Limitations are likewise method-specific. EoT states that only two classical EA operators, crossover and mutation, were explored; fitness relies on ground-truth answers; experiments are limited to GPT-3.5-turbo and GPT-4; and few-shot demonstrations were not combined with EoT because of cost constraints. Proposed future directions include richer operators such as differential evolution and multi-point crossover, unsupervised or self-evaluation fitness functions, multi-objective EoT, adaptive mutation rates $d_j \in \{-1,0,+1\}$17, tighter integration with retrieval or external tools, and population-level self-consistency (Jin et al., 2024). CoTEvol reports that current SFT uses only the single best-fitness CoT per problem and that naïve multi-trajectory training hurts performance due to logical noise; it proposes aggregation methods that leverage the full solution diversity and hybrid RL/evolutionary schemes (Wang et al., 16 Apr 2026).

Taken together, the literature characterizes evolutionary Chain-of-Thought as a convergence between evolutionary search and structured reasoning. In EVOC, the central claim is about a cognitive transition enabling open-ended cultural evolution. In LLM systems, the central claim is algorithmic: CoT prompts or trajectories can be iteratively varied and selected to improve instance-level reasoning, data synthesis, or distillation. A plausible implication is that the strongest unifying principle is not “chain-of-thought” as a fixed textual artifact, but the treatment of intermediate reasoning as a manipulable population whose variation and selection can be engineered.