Arispec

Arispec extracts structured data from the fixed-width reports that banks, insurers, telecoms, and public agencies still produce — including the ones whose headings and values have drifted apart over decades of separate edits.

• 0runtime dependencies
• 121tests across report fixtures and a 24-source discovery corpus
• Dataspecifications can be stored, diffed, generated, and reviewed

Ask the way a person reads the report

A money value may end in column 78 on one row and 79 on the next because a trailing minus sign was appended after right-justification. Pinning either column can silently turn a negative into a positive. Arispec instead lets a specification ask for the last money-shaped token on the row.

section detail starts(/^GL\s+Tran #/):
    rows:
        field gl:     columns 0-24
        field amount: last money

Type declarations do more than convert a slice after extraction: they help delimit the value itself. Labels, types, sections, repeated regions, and across-or-down relationships describe the document without embedding a parser in application code.

Refusal is part of the result

A malformed amount becomes UNKNOWN for that cell with a diagnostic; it never becomes a silent zero and never sinks a hundred-thousand-line import. An ambiguous date stays ambiguous until the spec declares its convention. The parser reports what it could not honestly know instead of manufacturing a plausible value.

Install

pip install arispec

Every Python example in the README and documentation is executed by the test suite and compared with the text printed beneath it.