Promises
Marking secrets
A field counts as secret when it is marked in any of these ways:
Secret[str]orAnnotated[str, Secret];- Pydantic
SecretStrandSecretBytes; - your own types, listed in
secret_types("module:Class"); - field names matching
secret_fieldspatterns (a heuristic, opt-in).
A parameter counts the same way when its own annotation is marked, for example
def entry(level: str, key: Secret[str] | None = None). The check then plants the canary in that parameter.
never_contains(Secret)
No secret planted in the input appears in the output. The check fills every secret field with a canary value and searches the output for it in plain text, JSON-escaped, URL-encoded and base64-encoded. It also searches what the function logs and what each fake dependency received during the call.
preserves("numbers", "dates", "quotes", "links", both_ways=True)
Every number, date, quote or link of the input is still in the output. With both_ways=True (the default) the
output may not add items either. Thousands separators and the word "percent" are normalised, so "15%" equals
"15 percent" and "1,000" equals "1000".
holds(predicate, name="")
A predicate over the input and the output is true, called as predicate(input, output). The lock entry hashes
the predicate's source, so an edit shows in review.
@invariant(holds(lambda inp, out: out.total == sum(inp.items), name="total adds up"), dry_run=True)
def summarize(order: Order) -> Summary: ...
Inputs
A check calls each dry_run=True function with saved regressions first, then values from your factories, then
values Hypothesis builds from the types. Random values rarely look like real data, so name a factory for the types
that matter (see Configuration).