Evolution

65 public types in this namespace, organized by kind.

Models & Types (56)

Type Summary
DelegateProgramFitnessEvaluator Adapts a caller-supplied function into an IProgramFitnessEvaluator.
EmbeddingCosineGenomeDistance A cosine distance over program embeddings, served from a primed cache so no comparison blocks on I/O.
EvolveBlockExtractionResult The outcome of scanning a program source for evolve-block markers.
FencedCodeBlock One fenced code block found in a model response.
FencedCodeExtractionResult The code recovered from a model response, together with every fence that was considered.
FileSystemProgramArtifactStore An artifact store that keeps small content in a per-genome index and large content in its own file.
InMemoryProposalProvenanceSink A bounded in-memory IProposalProvenanceSink for tests and short-lived runs.
InputOutputProgramFitnessEvaluator Scores a candidate program by running it on input/output examples through a caller-supplied engine.
JsonLinesProposalProvenanceSink Writes provenance records as JSON Lines into a directory of crash-safe, immutable segment files.
LlmJudgeProgramFitnessEvaluator<T> Blends a language model’s judgement of a program into the score a measured evaluator produced.
LlmProgramNoveltyJudge<T> Asks a language model whether a candidate is a meaningful change from the incumbent it most resembles.
LlmProgramVariationOperator<T> Proposes the next candidate program by asking a chat model to edit or rewrite the parent.
NoveltyGatingProgramFitnessEvaluator Consults a novelty policy before an inner evaluator is allowed to spend anything on a candidate.
ProgramArtifact One named piece of evidence an evaluation produced, held as bytes so binary output survives.
ProgramArtifactDescriptor What a store knows about one artifact without reading its bytes.
ProgramArtifactStoreObserver Persists each evaluation’s artifacts so they outlive the run and stay retrievable by genome.
ProgramDescriptorSet A validated, ordered collection of program descriptors evaluated together.
ProgramDiffApplyResult The program produced by applying a set of edit blocks, plus every block that did not apply.
ProgramDiffBlock One parsed SEARCH/REPLACE edit: the text to find and the text that replaces it.
ProgramDiffFailure Names one SEARCH/REPLACE block that could not be parsed or applied, and why.
ProgramDiffParseResult The blocks recovered from one model response together with every block that was rejected.
ProgramDiffTargetSplit The result of routing a reply’s edit blocks to the program or to the changes description.
ProgramDiversityDescriptor Measures how far a candidate program sits from a fixed reference set of programs.
ProgramEvolutionElite One archived program from a finished run, with its score, its cell, and a bounded copy of its source.
ProgramEvolutionLlmUsage Immutable totals describing how much language-model work a program-evolution run consumed.
ProgramEvolutionResult The reportable summary of a finished program-evolution run.
ProgramEvolutionTask Binds program genomes, a fitness evaluator, and program descriptors into one evolution task.
ProgramGenome An immutable program source treated as one evolvable candidate.
ProgramGenomeCodec Serializes ProgramGenome instances for portable evolution checkpoints.
ProgramLengthDescriptor Measures a candidate program by the character length of its normalized source.
ProgramLineEditDistance A normalized Levenshtein distance over whole lines; order sensitive, and still entirely local.
ProgramMetricAggregationResult The scalar quality produced from a metrics dictionary, with the evidence behind it.
ProgramMetricAggregator Turns a dictionary of evaluator metrics into the single scalar quality an archive ranks on.
ProgramMetricIssue One metric the aggregation could not use, together with the reason it was set aside.
ProgramMetricValue One reported metric value, tagged as a number, a boolean flag, or free text.
ProgramNoveltyDecision One novelty decision, with the rung that produced it and what that rung cost.
ProgramNoveltyPolicy The three-rung novelty gate: a free structural check, then an optional embedding check, then a judge.
ProgramPromptArtifact One named piece of output an evaluation produced, offered back to the model as evidence.
ProgramPromptAttempt One earlier attempt summarized for the prompt: what it changed, what it measured, how it compared.
ProgramPromptContext Everything one prompt is built from: the parent program, its scores, its history, and its evidence.
ProgramPromptExample One other program shown in a prompt, together with why it was chosen and how it scored.
ProgramPromptResult One rendered prompt: the messages to send, plus the record of how they were assembled.
ProgramPromptTemplate One piece of prompt text with named {placeholder} slots, parsed once and rendered many times.
ProgramPromptTemplateSet The complete, validated collection of prompt texts and phrases a program-evolution run renders.
ProgramProposalAttempt One recorded request-and-answer round while a language model was asked to improve a program.
ProgramRunOutputObserver Writes the best program to disk at every checkpoint and once more when the run stops.
ProgramRunOutputRecord What one best-program write produced: where the files went and which candidate they describe.
ProgramRunOutputWriter Writes the best program of a run to disk, with an info document describing it.
ProgramTokenComplexityDescriptor Measures a candidate program by the number of lexical tokens in its normalized source.
ProgramTokenSetDistance A Jaccard distance over the lexical tokens of two programs; no provider, no model, no network.
ProposalProvenanceLineage The full ancestry of one program, rebuilt from a recorded provenance stream.
ProposalProvenanceLineageStep One accepted edit in a reconstructed lineage: the parent, the child, and the exchange that joined them.
ProposalProvenanceReadResult What a read of a provenance stream produced, including what it could not use.
ProposalProvenanceRecord Everything known about one request-and-answer round that a language model was asked to turn into a candidate program: who asked, what was asked, what came back, what it cost, and what the answer turned into.
SandboxedProgramFitnessEvaluator Scores a candidate program by running it on input/output examples through the asynchronous sandbox contract, and records why each failing case failed.
ScriptProgramFitnessEvaluator Scores a candidate by running a caller-supplied evaluator script inside the same sandbox, under the same limits, and reading the JSON metrics it prints.

Structs (2)

Type Summary
EvolveBlockMarkers The start and end comment markers that delimit the editable region of a program.
EvolveBlockRegion One matched evolve block, split into the text before it, the block body, and the text after it.

Helpers & Utilities (7)

Type Summary
EvolveBlock Finds, extracts, and rewrites the marked editable region of a program source.
FencedCodeExtractor Extracts fenced code blocks from a model response and picks the best full rewrite.
ProgramDiff Parses SEARCH/REPLACE edit blocks from a model response and applies them to a program.
ProgramLanguageDetector Infers a program’s language and maps languages to file extensions, comment prefixes, and fence labels.
ProgramPromptBuilder Turns a parent program and its surroundings into the bounded, redacted messages sent to a model.
PromptTextRedactor Strips terminal control sequences and credential-shaped text before it reaches a prompt or a log.
ProposalProvenanceReader Reads a recorded provenance stream back and rebuilds the ancestry it describes.