Skip to content

Spec Inference

chimera.training.spec_inference infers specifications from existing source code via static analysis. SpecInferrer analyzes function signatures, return statements, and patterns to generate invariant-based regression tests. All analysis is AST-based — no code execution required.

Key Classes

InferredInvariant

Dataclass representing an invariant discovered from source code.

Field	Type	Description
`function`	`str`	Function name the invariant applies to
`file`	`str`	Source file path
`invariant`	`str`	Human-readable description (e.g. “returns int”)
`pattern`	`str`	Machine pattern: `return_type`, `non_null`, `has_docstring`, `argument_count`
`confidence`	`float`	0.0 to 1.0
`test_code`	`str`	Generated test function source

SpecInferrer

Analyzes Python source and infers invariants. Detected patterns:

return_type — function has type annotations
non_null — function never returns None
has_docstring — function has a docstring (regression guard)
argument_count — fixed number of parameters

Key methods:

analyze(source, file_path) -> list[InferredInvariant]
generate_test_file(invariants) -> str
write_test_file(output_path, invariants) -> str

Usage

from chimera.training.spec_inference import SpecInferrer

inferrer = SpecInferrer()

source = open("calculator.py").read()
invariants = inferrer.analyze(source, file_path="calculator.py")

for inv in invariants:
    print(f"{inv.function}: {inv.invariant} (confidence: {inv.confidence})")

# Generate a regression test file
test_content = inferrer.generate_test_file()
print(test_content)

# Or write directly to disk
inferrer.write_test_file("tests/test_invariants.py")

Oracle — grow the test suite during synthesis
Mutation Testing — evaluate test suite quality
Validation — detect overfitting with held-out tests