Ordo

Trust · Accuracy

A missed item is recoverable. A false task is not.

Ordo is built for precision first: it would rather miss an item you can paste in than invent work you have to clean up. This page explains how we measure that, so any number we publish can be checked against a method.

What we measure.

01

Proposal approval rate

Proposals a person approved, with or without edits, divided by all proposals Ordo showed.

02

False-task rate

Proposals dismissed as “not a commitment”, divided by all proposals shown. This is the number we work hardest to keep low.

03

Recall

Real commitments in a labelled conversation that Ordo proposed, divided by all the commitments the labellers found.

Method

How the evaluation works.

  • We measure each connector separately (Gmail, Slack, pasted notes and so on), because a promise reads differently in an email than in a chat.
  • The evaluation set is made of real conversations used only with consent, with personal details removed before labelling.
  • Two people label each conversation independently. A third resolves any disagreement. A commitment is someone promising or being asked to deliver something specific.
  • A false task is a proposal for something nobody committed to: an FYI, a newsletter, a question with no ask, or work that was already done.
  • We run the set before every release that changes how Ordo reads messages, and compare it with how people actually approved and dismissed proposals in the product.
  • Customer content is never used to train models.

Results

The first evaluation is still running. Until it is published here, with its sample size and date, the Ordo site cites no accuracy figures. Questions about the method? Ask us.

Start free

Try commitment infrastructure on one conversation.

Paste one email thread. Ordo proposes the next steps, and nothing happens until you confirm.