Cognotik | ProbabilisticReasoningTask

ExecutionConfig.json JSON

{
  "decision_context": "Cloud Migration Risk",
  "hypotheses": {
    "Success: Lower TCO": 0.4,
    "Failure: Egress Costs": 0.6
  },
  "evidence": [
    "Current egress is 50TB/mo",
    "Reserved instances available"
  ],
  "risk_tolerance": "medium",
  "calculate_expected_value": true,
  "identify_key_uncertainties": true,
  "suggest_experiments": true
}

→

Session UI (TabbedDisplay) Rendered

Overview Bayesian Update

✔ Posterior Probabilities Calculated

Hypothesis	Posterior
Success: Lower TCO	62.4%
Failure: Egress Costs	37.6%

"Evidence of reserved instances significantly increased the likelihood of TCO success..."

Live Results Showcase

Explore actual Bayesian analysis artifacts and logs generated by this task in the test workspace.

Configuration Parameters

Field	Type	Description
`hypotheses`*	`Map<String, Double>`	Map of hypotheses to prior probabilities. Must sum to 1.0.
`evidence`	`List<String>`	Observed data points used to update the prior beliefs.
`risk_tolerance`	`String`	"low", "medium", or "high". Influences decision recommendations (Default: "medium").
`calculate_expected_value`	`Boolean`	If true, quantifies outcomes and risks (Default: true).
`identify_key_uncertainties`	`Boolean`	If true, identifies critical unknowns (Default: true).
`suggest_experiments`	`Boolean`	If true, recommends tests to reduce uncertainty (Default: true).
`input_files`	`List<String>`	File patterns (e.g. `*/.kt`) to use as context.
`decision_context`	`String`	The problem statement or background for the analysis.

Task Lifecycle

Validation: Ensures probabilities are between 0 and 1 and sum to 1.0 (±0.01). Validates risk tolerance levels.

Context Loading: Reads specified input_files and inherits context from previous tasks in the orchestration.

Bayesian Update: The agent evaluates the diagnostic value of each piece of evidence and calculates posterior probabilities using Bayes' Theorem.

Multi-Stage Analysis: Sequentially performs Expected Value analysis, identifies Key Uncertainties, and designs Experiments based on the updated belief distribution.

Embedding Cognotik

Integrate ProbabilisticReasoningTask directly into your Kotlin applications or CI/CD pipelines using the UnifiedHarness.

Scenario A: Full Agent Planning

runPlan Example Kotlin

harness.runPlan(
    prompt = "Analyze the risks of migrating our legacy auth service to OAuth2.",
    cognitiveSettings = CognitiveModeConfig(type = CognitiveModeType.AdaptivePlanning),
    workspace = File("./project-analysis"),
    autoFix = true
)

Scenario B: Single Task Execution

runTask Example Kotlin

import com.simiacryptus.cognotik.plan.tools.reasoning.ProbabilisticReasoningTask.Companion.ProbabilisticReasoning

val config = ProbabilisticReasoningTaskExecutionConfigData(
    decision_context = "Market Entry Strategy",
    hypotheses = mapOf(
        "High Demand" to 0.3,
        "Moderate Demand" to 0.5,
        "Low Demand" to 0.2
    ),
    evidence = listOf("Competitor A just exited the market"),
    risk_tolerance = "low",
    suggest_experiments = true
)

harness.runTask(
    taskType = ProbabilisticReasoning,
    typeConfig = TaskTypeConfig(),
    executionConfig = config,
    workspace = File("./workspace")
)

Prompt Segment

The following logic is injected into the LLM context:

ProbabilisticReasoning - Reason under uncertainty using Bayesian analysis
  ** Specify hypotheses with prior probabilities (must sum to 1.0)
  ** Provide observed evidence to update beliefs
  ** Calculate expected values and quantify risks
  ** Identify key uncertainties that need resolution
  ** Suggest experiments to reduce uncertainty
  ** Useful for:
     - Risk assessment and management
     - Diagnostic reasoning (bug hunting)
     - A/B test analysis and decision making
     - Resource allocation under uncertainty
     - Technology adoption decisions