CSV Return Review — free command-line validation preview

A small offline Python preview for developer-coordinators who receive translated CSV files while the master text keeps changing. We are seeking feedback from up to five different game teams during a 14-day validation window. This is a limit on the feedback round, not a claim that only five download copies exist.

Free, with payments and tips disabled. Source code, not a graphical app or a translation service.

Is this your workflow?

You recently coordinated an external translator's return, keep all three files—the sent baseline, today's master, and the returned sheet—and use stable IDs with one source and one target column. You can run a Python command in a terminal. Your existing shared sheet, translation platform, or Git review does not already make this step straightforward.

If you have no saved baseline, no stable IDs, a multi-language spreadsheet you cannot intentionally reduce to three columns, or a workflow already covered by a shared source of truth/TMS, this preview is unlikely to help. A translator who does not control the master is not the primary intended user. If you need a GUI, a Windows/macOS installer, automatic merging, or translation quality checking, this preview does not provide them.

What it does

Read three UTF-8 CSV files with exactly id,source,target. Match by ID and produce JSON and escaped CSV reports: candidate target changes, current source drift, differing concurrent target edits, deleted/missing/unknown keys, and out-of-scope source edits. Invalid CSV/schema or duplicate/empty IDs block the report. Your input files are not changed. Every decision remains yours.

One deliberately constructed example

The included example has one ID and a multiline source with eight context lines. The baseline says Open the red door.; the current source says Open the blue door.; the returned target is 打开红门。 (“Open the red door”). In the recorded run, both raw Git and CSV-aware ID-sorted Git merged without a conflict. The preview reported HOLD / SOURCE_DRIFT.

The added warning comes from comparing source strings, not from understanding either language. The context lines deliberately separate the physical edits. This shows a warning can supplement clean text merging; it does not show superiority over a person's complete Git/spreadsheet review, detect every wrong translation, or establish measured time savings. The evidence image is a rendering of actual saved fields, not an application screenshot. Full inputs and raw outputs are included.

Download and try

The csv-return-review-0.1.0-validation-preview.zip file contains the unchanged Python CLI, synthetic examples, expected results, tests, reproduction commands, this example's actual outputs, English instructions, and a conditional preview license. Read README.md, then run the supplied synthetic example first. An exit code of 1 means the report asks for review; it is expected for the red/blue-door example.

Only Debian 13.6 / Linux x86_64 with Python 3.12.14 was actually tested. Python standard library only; Git 2.52.0 was used for the optional comparison. Other Linux configurations, Windows, macOS, GUI use, and desktop spreadsheet workflows are untested. Maximum 5 MiB and 20,000 records per input file. Exact three-column schema; no XLSX or additional language columns. Unknown IDs cannot reliably be mapped to renamed IDs; ID swaps, a wrong/forged baseline, and translation quality cannot be verified from these files alone.

Optional feedback—no private files

First try the synthetic example. If this matches a recent workflow, you may choose to compare an authorized local copy privately. Do not upload original CSVs, game text, reports, screenshots, names, email addresses, client details, or payment information.

An optional public comment may contain only these coarse answers: recently handled returns (yes/no); all three files available (yes/no); Python CLI feasible (yes/no); synthetic-only or own authorized local workflow; current method (Git / spreadsheet / TMS / other); report changed or clarified one decision (yes/no); existing method already sufficient (yes/no); reason category (source drift / target conflict / deleted key / none). Please respond once per team. Comments are public and itch displays account names; this is not an anonymous submission service. We retain only aggregate counts, not handles or customer files. No response is required to download.

This round tests workflow fit and usefulness, not willingness to pay. We do not promise future features, platform ports, individual debugging, response deadlines, or ongoing updates. Synthetic reproductions of defects may inform whether the preview should be withdrawn. No purchase or tip is accepted, so this round has no paid entitlement or purchase-refund flow.

Development used generative AI assistance for code, documentation, and the evidence illustration. The CLI itself makes no network or AI-service calls. This is an independent experimental tool, with no official Git, engine, or translation-platform endorsement.