Skip to content

Draft: FAQ and support boundaries

Publication state: DRAFT. This page sets a technical and privacy boundary for a future public launch. It does not create a support commitment, service level, data-processing agreement, public upload channel, or publication authorization.

What is WolfXL?

WolfXL is a Python-facing, Rust-backed spreadsheet execution engine with an openpyxl-style API. Its documented scope is the supported workflow surface in this repository. The scope is not every workbook, every Excel feature, every future OOXML extension, or every spreadsheet-adjacent document format.

Does WolfXL replace LibreOffice?

No such claim is made here. LibreOffice is a capable full office suite with broader document and format workflows. The launch study concerns a bounded set of spreadsheet operations where an agent might otherwise orchestrate a LibreOffice process. It does not compare WolfXL with Codex’s private spreadsheet engine, and it does not make a document-suite replacement claim.

Why are the Codex macOS and stock-LibreOffice Linux results separate?

They are different runtime and platform groups. Hardware, operating system, office build, source revision, setup, sample count, and process behavior can all affect an observation. The study presents them separately and does not calculate a pooled result or a cross-platform headline.

Do the recorded timing rows apply to my workbook?

No. A timing row applies only to its recorded source, runtime, scenario, fixture, cache mode, and measurement protocol. The local and Linux observations in docs/trust/codex-libreoffice-evidence-2026-09-02.json are dated evidence. Before relying on a result, reproduce a narrowly defined task with your own runtime and retain the generated receipt.

Is a formula result Excel-correct?

A comparison with LibreOffice is not Excel adjudication. The Codex/LibreOffice receipt records formula matches and differences, including unresolved scope; it does not establish broad formula correctness. An Excel formula claim needs Excel-authored or Excel-calculated evidence for the stated family and tolerance.

Does an unchanged-package result prove that no workbook can be harmed?

No. A package diff is evidence about selected input/output files and the requested operation. It does not prove that every workbook, feature interaction, or application rendering outcome is safe. A changed package part is also not itself proof of corruption. Keep the requested changes, unrelated drift, reopenability, and known limitations visible together.

What about rendering and PDF export?

LibreOffice render comparisons are not Microsoft Excel ground truth. The recorded study treats PDF export as a declared LibreOffice-only route. Do not infer a general WolfXL PDF capability from a comparison table. Choose a supported engine, a declared fallback, or a visible refusal for the operation at hand.

Which platforms are supported?

Use the release artifacts and their source-attested receipts for the exact platform and Python combination you intend to use. A result from one platform does not certify another. Codex runtime availability is also environment specific; a missing runtime is a visible blocker, not permission to substitute a different one without recording it.

Can I ask for help with a workbook fidelity failure?

Start with a privacy-preserving summary. A future public issue form asks for a sanitized receipt, runtime identity, workbook feature class, expected outcome, observed outcome, and a synthetic or redacted reproduction. It does not ask for a workbook upload.

This is a public boundary:

  • Do not attach a confidential workbook, sample data, formulas, names, email addresses, tokens, connection strings, file paths, hostnames, or screenshots containing private information.
  • Do not paste raw diagnostic output if it contains cell values, sheet names, paths, or customer metadata.
  • Do not assume an issue tracker is private merely because an issue is new or unlisted.
  • A workbook upload is never automatic. If a confidential sample is necessary, stop before sharing it publicly and use an owner-approved private channel with a separately agreed data-handling process.

How do I reduce a failing workbook safely?

Use the conservative local minimizer described in docs/operations/failure-minimizer.md. It writes a separate no-clobber output, records structural hashes rather than workbook content or local paths, and fails closed on several high-risk package classes. The minimized output can still contain sensitive material; inspect it yourself before sharing anything.

A useful public report can often contain only:

  1. a sanitized receipt or structural hashes;
  2. the exact WolfXL version, Python version, operating system, and architecture;
  3. the declared operation and expected versus observed result;
  4. the feature class, such as external links, a pivot cache, macro preservation, or a formula family; and
  5. a synthetic workbook or redacted reproduction steps that contain no private data.

Where should a report go?

When a public issue tracker is explicitly approved and available, use the Workbook fidelity report (privacy-first) form. Until then, do not infer a public destination from this draft or send a workbook to an unverified address.

Is there a commercial support or licensing promise?

Not in this draft. It makes no warranty, procurement, support-response, data processing, or commercial licensing representation. Consult the actual selected release and repository materials before relying on any such term.

What makes a high-signal report?

A report is actionable when it binds a narrow operation to a reproducible, privacy-safe observation. Include the selected engine and fallback decision, source/output hashes where safe, requested mutation, unrelated-drift result, terminal status, and a minimal synthetic reproduction. Keep red, unsupported, and inconclusive outcomes intact.