Limits#

class scikitplot.cleanprompt.Limits(max_input_chars=1000000, max_spans=100000, max_entries=50000, max_literal_terms=10000, max_literal_length=4096)[source]#

Hard bounds on one redaction pass.

Parameters:
max_input_charsint, default=1_000_000

Longest text accepted. Exceeding it raises LimitExceededError.

max_spansint, default=100_000

Most detections accepted before overlap resolution.

max_entriesint, default=50_000

Most distinct values newly detected in one text. Values carried in from earlier turns or pieces (a seed) do not count.

max_literal_termsint, default=10_000

Most user-supplied literal terms accepted.

max_literal_lengthint, default=4_096

Longest single literal term accepted.

Raises:
PolicyError

If any bound is not a positive integer.

Parameters:

  • max_input_chars (int)

  • max_spans (int)

  • max_entries (int)

  • max_literal_terms (int)

  • max_literal_length (int)

Notes

Developer notes. Bounds exist so that adversarial input fails loudly and early rather than consuming the process. They are policy, not constants, so a caller who genuinely needs a larger document can raise them explicitly and own that decision. Nothing in this submodule silently truncates: the truncated field on RedactionResult is always False and exists so a caller can assert that.

The default max_input_chars matches the default document ceiling of the optional spaCy tier, so the base tier and the NER tier fail at the same size rather than at two different, surprising ones.

as_dict()[source]#

Return a JSON-safe dictionary of these limits.

Returns:
dict

Field names to values.

Return type:

dict[str, Any]

max_entries: int = 50000#
max_input_chars: int = 1000000#
max_literal_length: int = 4096#
max_literal_terms: int = 10000#
max_spans: int = 100000#