Skip to main content
ZeroWidthdocs
Get started

Reviews

A review is what happens when you put a dataset together with a rubric and a group of raters. Each rater scores each item against each criterion; Caliper aggregates the result.

Reviews live inside the dataset they belong to — there's no top-level "reviews" list. Open a dataset's detail page and you'll see the review cards for any review running against that dataset, plus an Active reviews strip on the datasets index that surfaces work in progress across all datasets.

Creating a review

  1. Open the dataset

    From the Datasets tab in the caliper header, click into the dataset you want to review.

  2. Click New review

    The action lives in the dataset's header. The new-review dialog asks for a name, a description, a visibility setting, and the rubric to pair the dataset against.

  3. Pick a rubric

    Any rubric you can see in the workspace works. Caliper snapshots the rubric onto the review at this moment — future edits to the live rubric don't change this review's results.

The rating form

When a rater opens a review (via invite or share link), they land on a focused per-item form:

  • One item displayed at a time, in the dataset's native shape (raw text, Q&A, chat, file attachment).
  • The review's snapshot rubric below — every criterion, with its scale and description visible.
  • Next / Finish buttons that advance the rater forward; the review's progress bar updates as ratings land.

Raters can stop and resume — their progress sticks per item. They don't need to finish in one session.

Reading results

The review overview page is the home for results. It has four parts:

  1. Primary actions

    Review (or View ratings once the review is complete), Share, Invite. Same row on every review so the next action is always one click away.

  2. Highlights band

    Three callouts: lowest-scored item overall, items with highest rater disagreement, and worst-by-criterion. Tells you where to look first.

  3. Per-item breakdown

    Sortable table — every dataset item is a row, with mean / min / stddev / rating count / raters columns. Sort by mean to see the best and worst items; sort by stddev to find the items raters disagreed about.

  4. Raters strip

    Avatars + names of everyone who's contributed. Names resolve asynchronously as the page loads; you'll see them populate in.

Status

Reviews are either In progress or Complete. Complete just means the review owner closed it off — there's no automatic transition. A complete review still accepts ratings if you change your mind; the status is editorial.

The review status chip is the small color-keyed marker next to the review's name on its overview page and on the dataset detail page's review cards.

What persists, what doesn't

Where to go next

4 min read