Extract formal math specifications from documents

DOC2MATH converts a narrative technical document into a grounded Mathematical Problem Specification (MPS): variables, operators, constraints, objectives, and uncertainty, each one classified and cited back to the exact source phrase that supports it. It runs a Zero-Inference Protocol: nothing that is not stated in the document exists in the output (closed world), every element must carry an evidence field quoting the source, unknown values become null and ambiguous types become "ambiguous" rather than being guessed, structural inferences are explicitly tagged inferred: true with an inference_basis, and elements that are mentioned but under-defined get a status: "MISSING" with a missing_reason. It never introduces an equation or value that is not in the source text.

Use it for requests like "formalize this problem statement into math," "extract the mathematical structure from this research paper section," "what variables, constraints, and objectives are in this spec?," or "convert this word problem to a structured MPS." It does not invent missing equations, domains, values, or assumptions absent from the source, and a sparse prompt is returned with explicit missing-information markers rather than a filled-in guess. It produces a formal specification only - not a solved optimization model and not a proof.

Input is document text, a research excerpt, a problem description, or a specification. The skill first classifies the problem_class (optimization, classification, simulation, proof, estimation, or other), then extracts variables, operators, constraints, objectives, and uncertainty as structured elements, each with an evidence field and, where relevant, an inferred flag and status. It also surfaces missing_information - what the document implies but does not state - and validation flags covering whether objectives and variable bounds are complete, whether every element has evidence, an inference count, a missing count, and an overall formalizability rating of HIGH, MEDIUM, or LOW. The final output is a single JSON object:

{
  "mps_version": "1.0",
  "source_title": "...",
  "problem_class": "optimization",
  "variables": [...],
  "operators": [...],
  "constraints": [...],
  "objectives": [...],
  "uncertainty": [...],
  "missing_information": [...],
  "validation_flags": {
    "overall_formalizability": "HIGH"
  }
}

No external service integrations - it works directly on the document text supplied to it. The skill's repository is thebrierfox/doc2math-skill (MIT license, built by IntuiTek1), which also hosts a bring-your-own-key hosted tool version at the ACE license server.

Researchers, analysts, and engineers who need a rigorous, auditable bridge from a written problem statement to a formal mathematical specification - one where every extracted element can be traced back to the sentence that justified it, and where missing structure is flagged rather than silently filled in.