Hypothesis Generation

intelligence

Hypothesis Generation

Definition

Hypothesis Generation is the process of creating multiple competing interpretations, routes, or answers for a given decision problem. In the Prismatic Decision Core, this is deterministic -- the same input always produces the same set of hypotheses.

Technical Deep Dive

Standard Hypothesis Set

For every decision input, the generator produces at minimum:

  1. Positive ("Proceed") - The straightforward favorable interpretation
  2. Alternative ("Defer") - A plausible alternative suggesting more information is needed
  3. Contrary ("Reject") - The failure or negative scenario

Additional hypotheses are generated when:

  • Constraints are present - A conditional proceed hypothesis is added
  • Rich structured data (4+ fields) - A partial proceed (phased approach) hypothesis is added

Hypothesis Structure

Each hypothesis includes:

%{
  ordinal: 1,                          # Position in the set
  claim: "Proceed with...",            # The hypothesis statement
  reasoning_summary: "Because...",     # Brief justification
  assumptions: ["Market stable", ...], # What must be true
  supporting_signals: ["Revenue up"],  # Evidence in favor
  contradictory_signals: ["Risk X"]    # Evidence against
}

Signal Extraction

Signals are extracted from:

  • Structured data fields - Field existence and key names
  • Problem statement text - Keyword matching for opportunity/growth/benefit signals
  • Constraints - Constraint keys become conditional signals

Determinism Guarantee

The generator uses no randomness. Signal extraction, hypothesis ordering, and claim construction are all deterministic functions of the input data.

Connect & Contribute

Created by Tomas Korcak (korczis) | Open Source under GHL

Back to Glossary