sixma 0.3.0

Add your description here

Sixma

Probabilistic Correctness & Logical Falsification for Python.

"Stop writing unit tests. Start certifying reliability."

Sixma is a testing framework that replaces manual test cases with Generative Spaces and Statistical Certification. Instead of checking if f(2) == 4, you define the invariant f(x) == x^2 and Sixma proves it holds true with a specific Reliability and Confidence Level.

It is built on the Zero-Failure Reliability model: calculating exactly how many random trials are required to certify that a system is bug-free up to a certain probability threshold.

📦 Installation

uv add sixma
# or
pip install sixma

🚀 Quick Start

Write your tests as standard Python functions using the sixma.generators as type hints.

from sixma import certify, require, generators as g

# 1. Certify logic
@certify(reliability=0.999, confidence=0.95)
def test_drinking_age_logic(age: g.Integer(0, 120)):
    """
    Verifies legal drinking age logic.
    Target: 0.1% failure rate with 95% confidence (~2993 trials).
    """
    # Preconditions (Rejection Sampling)
    # If this is False, the input is discarded and regenerated.
    require(age >= 18)

    # Execution
    is_allowed = check_id(age)

    # Postconditions (Falsification)
    # If this fails ONCE, the certification fails immediately.
    assert is_allowed is True

Run it with standard pytest:

$ pytest -s

[Sixma] Target: 2993 successes (R=0.999, C=0.95)
[Sixma] Certified ✔️  (2993 passed, 41 discarded)
PASSED

🧠 The Philosophy

Standard property-based testing runs an arbitrary number of tests (e.g., 100). Sixma inverts this: You tell the framework how confident you want to be.

The number of trials is calculated dynamically using the Zero-Failure Testing formula:

$$ N = \left\lceil \frac{\ln(1 - C)}{\ln(R)} \right\rceil $$

Reliability	Confidence	Trials Required	Use Case
0.90	0.95	29	MVP / Quick Smoke Tests
0.99	0.99	459	Standard Business Logic
0.999	0.99	4,603	Core Algorithms
0.9999	0.999	69,075	Critical Infrastructure

• Reliability (): The probability that the code will NOT fail on a random input.
• Confidence (): The probability that our estimation of is correct.

🛠 Features

1. Smart Generators

Sixma generators are finite iterators first, and infinite streams second. They always yield edge cases (0, -1, empty strings, boundaries) before switching to random sampling.

# Will yield: 0, 10, 1, -1, 5, 8, ...
def test_integers(x: g.Integer(0, 10)): ...

# Will yield: "", "a", "   ", "xyz", ...
def test_strings(s: g.String(max_len=3)): ...

2. Clean Syntax

You don't need typing.Annotated. Just use the generator instance directly as the type hint.

@certify
def test_math(
    x: g.Integer(0, 100),
    y: g.Float(0.0, 1.0)
):
    assert x * y <= 100.0

3. Dependent Variables (g.Case)

Often, test inputs depend on each other (e.g., start < end or deadline > created_at). Instead of using require()—which wastes CPU cycles discarding invalid random samples—use g.Case to define a Generative Causal Chain.

@certify
def test_slicing(
    # 'case' generates a namespace where fields depend on previous ones
    case: g.Case(
        # 1. Independent Variable
        size = g.Integer(1, 100),

        # 2. Dependent: start must be within size
        start = lambda size: g.Integer(0, size - 1),

        # 3. Dependent: end must be > start
        end = lambda start, size: g.Integer(start + 1, size)
    )
):
    # Inputs are guaranteed to be valid! No require() needed.
    data = list(range(case.size))
    chunk = data[case.start : case.end]

    assert len(chunk) == case.end - case.start

4. Complex Data Structures

Model complex domains easily with List, Dict, and Object.

@dataclass
class Packet:
    id: int
    payload: str

# Generates fully populated class instances
def test_network(
    pkts: g.List(
        g.Object(
            Packet,
            id=g.Integer(1000, 9999),
            payload=g.String(max_len=255)
        ),
        min_len=1, max_len=5
    )
):
    assert len(pkts) >= 1

5. Time Travel (Temporal Testing)

Generate valid dates, times, and windows easily.

from datetime import date, timedelta

# Define a Business Quarter
Q1_2024 = g.Date(date(2024, 1, 1), date(2024, 3, 31))

@certify
def test_quarterly_report(day: Q1_2024):
    assert day.year == 2024
    assert 1 <= day.month <= 3

6. Reproducibility (Seeding)

Statistical tests must be reproducible. If a test fails, Sixma prints the Random Seed used.

Output on Failure:

❌ Falsified at trial 412!
   Seed: 84920174 (Set SIXMA_SEED=84920174 to reproduce)
   Inputs: {'x': -5}
   Error: assert -5 > 0

Reproduce it locally:

SIXMA_SEED=84920174 pytest tests/test_my_logic.py

📚 API Reference

@certify(reliability, confidence, max_discards)

The main decorator.

• reliability: Target probability of success (0.0 - 1.0).
• confidence: Statistical significance level (0.0 - 1.0).
• max_discards: Safety valve for infinite loops in require().

require(condition)

Used for Preconditions.

• If True: Continues execution.
• If False: Aborts the current trial, discards inputs, and samples again.

Generators (sixma.generators)

• Primitives: Integer, Float, Bool, String

• Combinators:

• List(gen, min_len, max_len)

• Dict(key=gen, ...)

• Object(Cls, field=gen, ...)

• Logic:

• Case(field=gen, dependent_field=lambda prev: gen)

• Temporal:

• Date(start, end)

• DateTime(start, end)

📄 License

MIT License.