Biotraxion
Sign in Request a pilot

Guide

Importing a protocol: what to check before you approve

Published November 11, 2026

Otto reads a document (Word, PDF, scan) and proposes a structured protocol. It does not sign it and does not approve it: acceptance and saving remain human. Between the document and the approved protocol there is therefore a review, and that review is what this page describes.

It is written for QA and lab managers. It does not replace your review procedure or your validation, and the framework that applies depends on your standard and your sponsor.

A proposal is not a protocol

Before the import, you choose the language and give your instructions. File names stay visible, and the launch is confirmed after you accept. These steps frame the request, they do not guarantee the result. Extraction is done by a model, and our release notes publish this: two imports of the same document can differ. The number of steps, how the document is split and the tables can change from one run to the next.

One practical consequence: an earlier import is not a reference, even if it was reviewed. The reference is the source document, and the review is done against it, line by line. Until that has happened, the proposal is worth a draft.

What gets distorted in reading

Start by counting. Does each step in the document have its counterpart in the proposed protocol, and does none appear twice? When several complementary documents are imported together, read with the most care here: a step can go missing or show up twice. Step days are put in ascending order, but ascending is not necessarily right. Check the order and the day of each step against the document.

Then the tables. They are taken one by one, and the thermocycler program and the reaction mix give two separate steps. The total volume of a mix is calculated from the number of reactions, then recalculated at launch. Variables are normalized and references follow renaming. That calculation starts from what was extracted: a wrong volume per reaction gives a wrong total. So read the volumes, units, equipment and incubation times value by value.

For a long document, text import is limited to 20,000 characters, with a specific reason given when it is refused. A long protocol is processed in batches, with up to three retries, after which a message asks you to split the document. If you split it, you get several imports: check that no step is missing at the join between two parts.

What remains is what is missing. Ambiguous matches ask for your choice, and missing information is left for you to fill in. An empty field is an honest signal. The risk is the plausible value that slips through. When in doubt, go back to the document, not to your memory.

Who approves, and what to do with a mismatch

Good practice separates the author of the version from the approver: two people, and an approver who signs in their own name and can say why they approve. There is one exception. An administrator can approve their own version, after a warning, with an explicit reason and a personal signature, and the exception is identified in the history. A traced exception is still an exception: if it becomes routine, the question is for your organization, not for the tool.

When the review finds a mismatch, you correct it before approving, then reread the corrected part against the document. You never approve "to see". An approval says that someone has read it, and a wrong protocol should not enter the record while you wait for something better.

What we do not say

We do not say an import is accurate, or that two imports of the same document will give the same protocol. We do not put a number on the reliability of an extraction. Nor do we say Otto replaces the review: it prepares, you check and you approve.

Five or six questions for any import tool

  1. Who approves, and can the same account prepare and approve? If so, where is the exception traced?
  2. Do two imports of the same document give the same result, and is that limit published?
  3. Are tables taken one by one, or merged?
  4. Are totals and references recalculated when a value or a name changes?
  5. What happens with a document that is too long?
  6. What does the tool show when information is missing or a match is ambiguous?

Get in touch

Write to us: contact@biotraxion.com, or request a pilot from the home page.

Keep reading

  • Modular registry or one tool per function: what an audit gains
  • Signing every step: what an inspector can check
  • Bench execution record vs ELN: what changes at inspection
  • The Audit Pack: what a sponsor can verify on their own
  • Several sponsors, one laboratory
  • Declare a deviation at the moment it happens
  • ALCOA+ from the bench: what each letter asks of your hands

← Back to home

© 2026 Biotraxion, activité portée par KIMA HOLDING, SARL au capital de 3 300 €, RCS Toulouse 999 711 682, 5 rue Jules Claretie, 31400 Toulouse.