01 / Research practice
Keep source evidence, translation coverage and page geometry as three review questions
A clean geometry report cannot verify a quotation; an intact chapter map cannot certify a translation. Design the handoff around those distinctions.
Suppose a colleague hands you an excerpt, an English Markdown document, its Traditional Chinese version and the source PDF. “Looks good” is not a useful review result. It hides which object was checked and which question remains open.
A small review matrix
| Question | Record to inspect | What remains human |
|---|---|---|
| Can I find the cited claim? | Evidence ID, quote and source location | Open the actual filing and verify context |
| Are the intended chapters paired? | Section indices and object differences | Read meaning and confirm correspondence |
| What dimensions will this page expose? | Page boxes, scale and rotation | Decide whether the layout is intentional |
Use a synthetic case first: E02 lacks a period, the translated Methods chapter lacks an image target, and page 2 is rotated. These are three independent findings. Completing the period does not fix the missing image, and an ordinary rotation is not evidence that the underlying financial claim is correct.
Write three separate decisions
E02 / evidence: period still needs original-source review
section 1 ↔ 1 / structure: image omission needs a decision
page 2 / geometry: 90-degree rotation appears intentionalKeep the labels when you hand off the folder. A later reviewer can then close one item without accidentally approving the others. Do not collapse “no field checks” and “no structure differences” into a single green quality score.
Use the appropriate upstream workflow
Eqvero is the related public-disclosure research product. KolmoPDF can supply Markdown from a PDF through its separate cloud conversion service. KomoPDF is the related editing product when a page change is genuinely needed. None of those relationships makes the three local tools a single automatic pipeline.
The evidence ledger never reads the PDF. The chapter mapper never evaluates translation meaning. The geometry inspector never validates a quotation. The review matrix is useful precisely because it keeps those boundaries visible.