HypothesisGenerator

Category: Reasoning | Module: mycontext.templates.free.reasoning

Generates well-formed, testable hypotheses from observations. Implements the scientific method: primary hypothesis (H1), null hypothesis (H0), alternative hypotheses, testable predictions, experimental design, success criteria, and rival hypotheses. Every hypothesis produced is falsifiable by design.

When to Use

• Research and experimentation planning
• A/B test hypothesis formulation
• Product analytics investigations
• Scientific research methodology
• Business hypothesis-driven decision making
• Explaining unexpected observations

Quick Start

from mycontext.templates.free.reasoning import HypothesisGenerator

gen = HypothesisGenerator()

result = gen.execute(
    provider="openai",
    observation="Sales increase significantly when we send email reminders 24 hours before subscription renewal",
    domain="e-commerce",
)
print(result.response)

Methods

build_context(observation, domain="general", context=None)

ctx = gen.build_context(
    observation="Users who complete onboarding in under 10 minutes have 3x higher 90-day retention",
    domain="SaaS product",
    context="Sample of 10,000 users over 6 months",
)

Parameters:

Parameter	Type	Default	Description
observation	str	""	The observed phenomenon to explain
domain	str	"general"	Domain context — shapes variable selection and examples
context	str | None	None	Additional context about the observation

execute(provider, observation, domain="general", context=None, **kwargs)

result = gen.execute(
    provider="openai",
    observation="Mobile users have 2x higher bounce rate than desktop users",
    domain="web analytics",
    context="E-commerce site, Q4 data, all user segments",
)

Generated Output Structure

Each run produces a 10-section scientific document:

1. Observation Analysis — What exactly was observed, context, significance
2. Background Knowledge — Relevant theory, prior findings, possible mechanisms
3. Hypothesis Formulation
   • H1 (Primary): Cause-effect statement, variables, mechanism
   • H0 (Null): No-effect statement
   • H2, H3 (Alternatives): Other possible explanations
4. Testable Predictions — Specific outcomes if H1 is true vs. false
5. Variables & Controls — Independent, dependent, control, confounding variables
6. Experimental Design — Method, sample size, conditions, measurements
7. Success Criteria — Statistical thresholds and practical significance
8. Limitations & Assumptions — Boundary conditions
9. Rival Hypotheses — Competing explanations and how to distinguish them
10. Next Steps — Immediate, short-term, and long-term actions

Examples

Product A/B Test Hypothesis

result = gen.execute(
    provider="openai",
    observation="Users who see the pricing page before signing up convert at 18% vs 12% for direct sign-up",
    domain="SaaS conversion optimization",
)

Scientific Research

result = gen.execute(
    provider="anthropic",
    observation="Patients who exercise 3x/week report 40% reduction in chronic pain scores after 8 weeks",
    domain="clinical research",
    context="Retrospective study, N=200, ages 40-65",
)

Business Analytics

result = gen.execute(
    provider="gemini",
    observation="Customer support tickets spike every Tuesday afternoon by 300%",
    domain="customer operations",
    context="B2B SaaS product, global customer base",
)

The Falsifiability Principle

All hypotheses generated follow the Popperian principle of falsifiability — they must be testable and potentially refutable by evidence. The pattern enforces this through:

• Testable predictions: Specific outcomes that would confirm or deny H1
• Null hypothesis: Explicit no-effect baseline for statistical testing
• Success criteria: Quantitative thresholds for accepting or rejecting H0
• Rival hypotheses: Alternative explanations that must also be tested

API Reference

Method	Returns	Description
build_context(observation, domain, context)	Context	Assembled context
execute(provider, observation, domain, context, **kwargs)	ProviderResponse	Execute and return hypotheses
generic_prompt(observation, domain, context_section)	str	Zero-cost prompt string