Statistical Inference Engine

Architecture Overview

The inference engine is modular and deterministic.

Core layers

• Loader → dataset retrieval
• Preprocessing → NA handling
• Assumptions → statistical validation
• Dispatcher → pure routing logic
• Test modules → statistical computation
• Formatter → console output
• Exporter → Markdown / PDF generation

Execution Flow

flowchart TD
    A[Load Indexed Dataset] --> B{Multiple Files?}
    B -- Yes --> C[Merge on Key]
    B -- No --> D[Continue]
    C --> D
    D --> E[Apply NA Strategy]
    E --> F[Dispatch Test]
    F --> G[Check Assumptions]
    G -- Pass --> H[Run Parametric Test]
    G -- "Fail (AutoRoute)" --> I[Run Nonparametric Alternative]
    G -- Fail --> H
    H --> J[Compute Statistics]
    I --> J
    J --> K[Create InferenceResult]
    K --> L[Format Output]
    L --> M{Export?}
    M -- Yes --> N[Generate MD/PDF]
    M -- No --> O[CLI Display]

Example Usage

Run Pearson correlation:

indexly infer-csv dataset.csv --test correlation --x height --y weight --use-raw

Example output:

============================================================
TEST: pearson_correlation
------------------------------------------------------------
Statistic : 0.842193
P-value   : 0.000012
95% CI    : [0.712301, 0.913882]
------------------------------------------------------------
Interpretation:
  Strong positive linear association.
============================================================

Design Guarantees

• Always returns a unified InferenceResult
• Fisher Z-transform for Pearson CIs
• T-distribution for mean CIs
• Explicit alpha tracking
• No side effects in dispatcher
• Reproducible metadata included

Next

Continue to How It Works to understand:

• Which test to choose
• Required arguments
• Example CLI commands
• Advanced options