Adapt Eurisko¶

Adapt Eurisko — mvp.adapt_eurisko

Cluster: ML & Optimisation | Type: component | MCP Tools: 35

Overview¶

EURISKO-inspired heuristic discovery engine that maintains a library of scored, categorised heuristics and applies them to problem contexts to generate ranked discoveries and recommendations. Heuristics track Bayesian confidence via a Beta-Binomial model — confidence updates after each application, rewarding successful use and down-weighting failures over time. Supports self-modification guardrails (frozen flag, generation depth cap) and three execution types: recommendation, registered transform, and LLM-guided.

When to use:

• Augmenting a solver or goal-engine step with heuristic guidance drawn from a curated knowledge base
• Discovering which problem-solving strategies are most relevant given a context dictionary
• Evaluating proposed solutions against problem objectives using registered evaluator transforms
• Building a self-improving heuristic library that adapts its confidence scores from live feedback

Example:

from mvp.adapt_eurisko import AdaptEuriskoBlock, EuriskoInput

block = AdaptEuriskoBlock(name="eurisko")
result = block.infer(EuriskoInput(
    operation="discover",
    problem={"domain": "optimisation", "constraint": "non-convex", "budget": "small"},
    top_k=3,
))
# result.value.discoveries → list of top-3 heuristic recommendations
# result.value.confidence_updates → updated confidence per applied heuristic

Works well with: goal_engine, adapt_experta, cegis

Public API¶

AdaptEuriskoBlock(AIBlock[EuriskoInput, EuriskoOutput, dict])¶

EURISKO-inspired heuristic discovery engine.

Methods:

infer(data: EuriskoInput) -> Result[EuriskoOutput]¶

Heuristic(BaseModel)¶

A single heuristic rule with confidence tracking.

Methods:

bayesian_confidence() -> float¶

Posterior mean of Beta(alpha, beta_param).

uncertainty() -> float¶

Approximate width of 95% credible interval.

EuriskoInput(BaseModel)¶

Input to AdaptEuriskoBlock.

EuriskoOutput(BaseModel)¶

Output from AdaptEuriskoBlock.

AdaptEuriskoMCPBlock(AIBlock[MCPEuriskoInput, MCPEuriskoOutput, dict])¶

29-op EURISKO heuristic engine with SQLite persistence.

Methods:

infer(data: MCPEuriskoInput) -> Result[MCPEuriskoOutput]¶

MCPEuriskoInput(BaseModel)¶

MCPEuriskoOutput(BaseModel)¶

EuriskoStore¶

Sync SQLite store with 6 tables.

Constructor:

Methods:

save_session(name: str, heuristics_json: str, units_json: str, notes: str) -> str¶

load_session(name: str) -> dict[str, Any] | None¶

list_sessions() -> list[dict[str, Any]]¶

delete_session(name: str) -> bool¶

create_unit(name: str, isa_json: str, worth: int, english_description: str, slots_json: str) -> str¶

get_unit(name: str) -> dict[str, Any] | None¶

update_unit(name: str, slots_json: str | None = None, worth: int | None = None) -> bool¶

list_units(isa_filter: str = '') -> list[dict[str, Any]]¶

add_task(op: str, params_json: str, priority: int, reasons_json: str) -> str¶

get_task(task_id: str) -> dict[str, Any] | None¶

update_task(task_id: str, status: str, result_json: str) -> None¶

list_tasks(status_filter: str = '') -> list[dict[str, Any]]¶

add_outcome(heuristic_name: str, success: bool, problem_json: str, notes: str, domain: str = '') -> str¶

query_outcomes(heuristic_name: str = '', limit: int = 50) -> list[dict[str, Any]]¶

summarize_outcomes(heuristic_name: str = '') -> dict[str, Any]¶

summarize_outcomes_weighted(heuristic_name: str = '', decay: float = 0.95) -> dict[str, Any]¶

Recency-weighted outcome summary. Recent outcomes weighted more.

summarize_outcomes_by_domain(heuristic_name: str = '') -> list[dict[str, Any]]¶

Per-domain success rate breakdown.

add_mutation(parent_name: str, child_name: str, mutation_type: str, description: str) -> str¶

list_mutations(parent_name: str = '') -> list[dict[str, Any]]¶

get_generation_depth(heuristic_name: str) -> int¶

Walk mutation lineage via recursive CTE, return max depth.

count_mutations_since(since_iso: str) -> int¶

Count mutations recorded after the given ISO timestamp.

record_verification(heuristic_name: str, transform_name: str, domain: str, retracted: bool, reason: str) -> str¶

list_verifications(heuristic_name: str = '') -> list[dict[str, Any]]¶

text_search(query: str, top_k: int = 5) -> list[dict[str, Any]]¶

search_units(query: str, top_k: int = 5) -> list[dict[str, Any]]¶

count_all() -> dict[str, int]¶

MCP Tools¶