StepByStepReasoner

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

Guides systematic problem solving through clear, logical steps. Implements Chain-of-Thought methodology with transparent reasoning — every step shows what, why, how, and the intermediate result. Ideal for math, logic, analysis, and any problem where showing your work matters.

Based on:

• "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" (Wei et al., 2023)
• IBM Zurich cognitive tools research (2025)

When to Use

• Mathematical and quantitative problems
• Multi-step logical reasoning
• Scientific analysis with intermediate calculations
• Problems where you need to verify the solution
• Educational contexts where transparent reasoning matters
• Debugging complex processes step-by-step

Quick Start

from mycontext.templates.free.reasoning import StepByStepReasoner

reasoner = StepByStepReasoner()

result = reasoner.execute(
    provider="gemini",
    problem="If a train travels 120km in 2 hours, then speeds up to travel 200km in the next 2.5 hours, what is its average speed for the whole journey?",
    domain="mathematical",
    temperature=0.3,
)
print(result.response)

Constructor

reasoner = StepByStepReasoner()

Methods

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

ctx = reasoner.build_context(
    problem="A company's revenue grew 15% in Y1 and 22% in Y2 from a base of $1M. What's the compound growth?",
    domain="mathematical",
)

# Export and use with any LLM or framework
print(ctx.to_openai())

Parameters:

Parameter	Type	Default	Description
problem	str	""	The problem to solve
context	str | None	None	Additional context or constraints
domain	str	"general"	Problem domain — affects framing

execute(provider, problem, context=None, domain="general", temperature=0.3, **kwargs)

result = reasoner.execute(
    provider="gemini",
    problem="Prove that the sum of any two consecutive odd numbers is divisible by 4",
    domain="mathematical",
    temperature=0.3,
)

Parameters:

Parameter	Type	Default	Description
provider	str	"gemini"	LLM provider (gemini recommended for reasoning)
problem	str	""	The problem to solve
context	str | None	None	Additional context
domain	str	"general"	Problem domain
temperature	float	0.3	Lower = more consistent reasoning. Keep between 0.2–0.4
**kwargs			Additional provider parameters

Domain Options

The domain parameter adjusts framing for the problem type:

Domain	Best for
"mathematical"	Calculations, proofs, quantitative problems
"logical"	Deductive/inductive reasoning, logic puzzles
"scientific"	Hypothesis testing, experimental analysis
"analytical"	Business analysis, data interpretation
"general"	Any problem not in above categories

Reasoning Framework

The pattern enforces a 5-phase problem-solving approach:

Phase 1: UNDERSTAND

Restate the problem, identify problem type, list all known information and constraints. Forces the model to engage with the problem before solving it.

Phase 2: PLAN

Define solution strategy and methods. Sketch the solution path visually:

[Input] → [Step A] → [Step B] → [Step C] → [Solution]

Phase 3: EXECUTE

Work through each step with four elements per step:

• What you're doing
• Why you're doing it
• How you do it (calculations, reasoning)
• The result of this step

Phase 4: VERIFY

Check against original problem. Apply alternative verification method if possible. Sanity-check scale and units.

Phase 5: CONCLUDE

State the final answer explicitly. Assess confidence level and explain it.

Examples

Mathematical Problem

result = reasoner.execute(
    provider="gemini",
    problem="""
    A store sells shirts at 40% markup over cost.
    During a sale, they discount 25%.
    What's the actual profit margin as a percentage of cost?
    """,
    domain="mathematical",
    temperature=0.3,
)

Logical Reasoning

result = reasoner.execute(
    provider="openai",
    problem="""
    All managers attend the Monday meeting.
    Sarah doesn't attend the Monday meeting.
    Sam attends the Monday meeting.
    Can we conclude that Sam is a manager?
    """,
    domain="logical",
)

Scientific Analysis

result = reasoner.execute(
    provider="anthropic",
    problem="""
    An experiment shows Drug A reduces blood pressure by 12mmHg in a sample of 150 patients.
    The standard deviation is 8mmHg. Calculate the 95% confidence interval
    and determine if results are statistically significant (p < 0.05).
    """,
    domain="scientific",
    temperature=0.2,
)

Temperature Guidance

Temperature matters for reasoning

Lower temperatures produce more deterministic, reproducible step-by-step solutions. For mathematical/logical problems:

• 0.2 — Most consistent, minimal creativity
• 0.3 — Recommended default (balance)
• 0.5+ — More varied but less reliable for calculations

Integration with Quality Metrics

Pair with QualityMetrics to measure solution thoroughness:

from mycontext.templates.free.reasoning import StepByStepReasoner
from mycontext.intelligence import QualityMetrics

ctx = StepByStepReasoner().build_context(
    problem="Calculate the ROI of a $500k investment with 12% annual return over 5 years",
    domain="mathematical",
)
result = ctx.execute(provider="openai")
metrics = QualityMetrics.evaluate(ctx, result.response)
print(f"Completeness: {metrics.completeness}")

Output Format

## Step 1: UNDERSTAND
**Restate the Problem**: ...
- What we're looking for: ...
- Key information given: ...

## Step 2: PLAN
**Strategy**: ...
**Sketch the Solution Path**:
[Input] → [Step A] → [Step B] → [Solution]

## Step 3: EXECUTE
### Step 3.1: [Description]
- **Action**: ...
- **Calculation**: ...
  = [intermediate result]
- **Result**: ...

## Step 4: VERIFY
...

## Step 5: CONCLUDE
🎯 **[Final Answer]**
Confidence: High

API Reference

Method	Returns	Description
build_context(problem, context, domain)	Context	Assembled context
execute(provider, problem, context, domain, temperature, **kwargs)	ProviderResponse	Execute reasoning
generic_prompt(problem, context_section, domain)	str	Zero-cost prompt string