Bias Evolution API Reference¶

MCP Tool Reference (25 tools)¶

All tools are exposed under the adapt_bias MCP skill package. Mutating tools require premium or trialing access.

Core Operations¶

bias_snapshot¶

Collect the current bias landscape from all registered components.

Returns: { snapshot: dict, component_count: int }

bias_observe¶

Classify the current system situation via context_engine and always-on memory.

Returns: { situation: str, is_novel: bool, confidence: float, memory_insight: str }

bias_understand¶

Structural analysis of a component's bias via self_model and deep_understanding.

Returns: { contracts: dict, verification: dict, domain_spec: dict, memory_patterns: str }

bias_evaluate¶

Multiobjective fitness scoring across speed, cost, and quality dimensions.

Returns: { speed: object, cost: object, quality: object, pareto: object }

bias_evolve ¶

Run the full evolution cycle: propose mutations via multi-agent debate, evaluate, Pareto-select.

Returns: { proposal: dict, frontier: list[dict], version: int, best_score: float }

bias_apply ¶

Apply a specific bias mutation with 4-layer safety verification.

Returns: { applied: bool, version: int, safety_layers_passed: int, healing_report: object }

bias_rollback ¶

Revert a component to a previous bias version.

Returns: { version: int, snapshot: dict }

bias_ground_check¶

Check a candidate bias against all 6 theoretical foundation principles.

Returns: { overall_passed: bool, results: list[PrincipleResult], warnings: list[str] }

Each PrincipleResult:

{
  "name": "compression_conjecture",
  "passed": true,
  "confidence": 0.8,
  "explanation": "Parameter count: 3 -> 3. Within compression bounds."
}

bias_self_heal¶

Run automated self-debugging and health checks after a bias mutation.

Returns: { audits: list, heals: list, health: object, auto_rollback: bool }

Evolution Management¶

bias_list_history¶

List the version history for a component's bias.

Returns: { history: list[{ version: int, timestamp: str, summary: str }] }

bias_get_frontier¶

Get the current Pareto frontier of top-performing bias candidates.

Returns: { frontier: list[{ bias: dict, score: float }] }

bias_compare¶

Compare two bias versions side-by-side.

Returns: { diff: dict, a: dict, b: dict }

Grounding & Safety¶

bias_list_principles¶

List all 6 theoretical grounding principles and their descriptions.

Returns: { principles: dict[str, str] }

bias_check_principle¶

Run a single grounding principle check against a candidate.

Returns: { name: str, passed: bool, confidence: float, explanation: str }

bias_safety_status¶

Report the status of all 4 safety layers.

Returns: { layers: list[{ name: str, available: bool, status: str }], required: int }

bias_validate_candidate¶

Dry-run validation of a candidate bias without applying it. Runs all safety layers and grounding checks.

Returns: { valid: bool, safety_layers_passed: int, grounding_report: dict, warnings: list[str] }

Memory & Observability¶

bias_memory_query¶

Query always-on memory for bias evolution history and patterns.

Returns: { answer: str, sources: list[str] }

bias_memory_ingest¶

Manually ingest a note into bias evolution always-on memory.

Returns: { ingested: bool }

bias_metrics¶

Get observability metrics for bias evolution operations.

Returns: { total_mutations: int, successful: int, rejected: int, rollbacks: int, avg_safety_score: float }

bias_audit_log¶

Retrieve the full audit log of bias mutations.

Returns: { entries: list[{ timestamp: str, component: str, op: str, version: int, outcome: str }] }

Batch & Configuration¶

bias_evolve_batch ¶

Run evolution across multiple components in a single call.

Returns: { results: dict[str, { proposal: dict, version: int }] }

bias_get_config¶

Get the current BiasEvolutionConfig.

Returns: BiasEvolutionConfig as JSON

bias_update_config ¶

Update bias evolution configuration at runtime. Changes are not persisted to disk.

Returns: { updated: bool, config: object }

bias_export¶

Export bias data for a component as a portable JSON bundle.

Returns: { component: str, current_bias: dict, history: list[dict] | null }

bias_import ¶

Import a previously exported bias bundle.

Returns: { imported: bool, version: int, warnings: list[str] }

REST Endpoint Reference (7 endpoints)¶

All endpoints are prefixed with /api/v1/bias. Mutating endpoints require premium or trialing authentication plus confirm_token.

All responses include the header:

X-G6-Warning: experimental-self-modification

GET /api/v1/bias/snapshot¶

Collect the current bias landscape.

Response:

{
  "snapshot": {
    "llm_router": {
      "name": "llm_router",
      "type": "LLMBlock",
      "strategy": "cascading_fallback",
      "evolvable_params": { ... },
      "version": 3
    }
  },
  "component_count": 84,
  "warnings": ["EXPERIMENTAL: ..."]
}

POST /api/v1/bias/observe¶

Classify the current system situation.

Request:

{
  "workflow_trace": {
    "quality_score": 0.72,
    "latency_ms": 1450,
    "tokens_used": 3200
  }
}

Response:

{
  "situation": "stable",
  "is_novel": false,
  "confidence": 0.85,
  "memory_insight": "Similar pattern seen 3 cycles ago with positive outcome."
}

POST /api/v1/bias/evaluate¶

Run multiobjective fitness evaluation.

Request:

{
  "workflow_trace": { "quality_score": 0.72, "latency_ms": 1450 },
  "weights": { "speed": 0.4, "cost": 0.3, "quality": 0.3 }
}

Response:

{
  "evaluation": {
    "speed": { "score": 0.8, "details": "..." },
    "cost": { "score": 0.65, "details": "..." },
    "quality": { "score": 0.72, "details": "..." },
    "pareto": { "rank": 1, "dominated_by": 0 }
  }
}

POST /api/v1/bias/evolve¶

Run the full evolution cycle for a component. Premium or trialing access required.

Request:

{
  "target_component": "llm_router",
  "confirm_token": "human-reviewed-secret",
  "evolution_config": { "max_iterations": 3 },
  "weights": { "speed": 0.5, "cost": 0.3, "quality": 0.2 }
}

Response:

{
  "proposal": { "name": "llm_router", "strategy": "cascading_fallback", "evolvable_params": { ... } },
  "frontier": [ { "bias": { ... }, "score": 0.87 } ],
  "version": 4,
  "best_score": 0.87,
  "warnings": ["EXPERIMENTAL: ..."]
}

POST /api/v1/bias/apply¶

Apply a specific bias mutation. Premium or trialing access required.

Request:

{
  "target_component": "llm_router",
  "confirm_token": "human-reviewed-secret",
  "bias_candidate": {
    "name": "llm_router",
    "type": "LLMBlock",
    "strategy": "cascading_fallback",
    "assumptions": ["Model availability varies", "Cost correlates with quality"],
    "evolvable_params": {
      "temperature": { "type": "float", "value": 0.7, "min": 0.0, "max": 2.0 }
    },
    "version": 4
  }
}

Response:

{
  "applied": true,
  "version": 4,
  "safety_layers_passed": 4,
  "healing_report": { "audits": [], "heals": [], "health": null },
  "warnings": ["EXPERIMENTAL: ..."]
}

POST /api/v1/bias/rollback¶

Revert a component to a previous bias version. Premium or trialing access required.

Request:

{
  "target_component": "llm_router",
  "confirm_token": "human-reviewed-secret",
  "rollback_version": 3
}

Response:

{
  "version": 3,
  "snapshot": { "llm_router": { ... } },
  "warnings": ["EXPERIMENTAL: ..."]
}

POST /api/v1/bias/ground-check¶

Check a candidate bias against theoretical foundations.

Request:

{
  "target_component": "llm_router",
  "bias_candidate": { ... }
}

Response:

{
  "overall_passed": true,
  "results": [
    { "name": "compression_conjecture", "passed": true, "confidence": 0.8, "explanation": "..." },
    { "name": "physical_symbol_system", "passed": true, "confidence": 0.9, "explanation": "..." },
    { "name": "category_theory", "passed": true, "confidence": 1.0, "explanation": "..." },
    { "name": "autopoiesis", "passed": true, "confidence": 0.85, "explanation": "..." },
    { "name": "human_controllability", "passed": true, "confidence": 0.9, "explanation": "..." },
    { "name": "antifragility", "passed": true, "confidence": 0.8, "explanation": "..." }
  ],
  "warnings": []
}