Use this article to find, filter, and tag your Evals, create a new Eval, organize your Evals, import Simulations from a CSV (comma-separated values) file, add a Scorecard, and see how Fin is performing across your Evals.

The Evals page lists every Eval in your workspace, so you can see how each group of tests performed and run them again. An Eval is a themed group of tests, and each test inside it is called a Simulation. To open the page, go to Test > Evals.

The Evals table has these columns:

To organize your Evals, group Simulations into Evals by whatever theme makes sense for your business. Many teams start with what isn't working yet, then build out from there. Common starting points include:

Over time, build one or a few Evals as a "golden set" that covers the conversations that matter most to you. Re-run the golden set regularly, for example weekly or whenever you change Fin's configuration, to make sure your most important scenarios still work.

To find an Eval by name, type in the Search evals box at the top of the Evals page. To narrow the list, click Add Filter and choose Result or Tag.

To create an Eval, choose whether to build it by hand or to import a list of customer scenarios from a CSV file.

Each Eval can hold up to 50 Simulations. To import from a CSV file, see "How to import Simulations from a CSV file" below. To set up the Simulations inside an Eval, see the Set up Simulations for Evals article.

To import Simulations from a CSV file, upload it from the Evals page or from inside an Eval. Each row becomes one Simulation, and you review every column match before anything is saved. To import, you need Can view configuration and propose changes in a draft release or Can view configuration and publish changes directly to main (individually or in a release) under Fin AI Agent & Automation.

To import into a new Eval, go to Test > Evals, click New eval, and select Upload a CSV. To add Simulations to an Eval you already have, open it, select Import CSV from the options menu (...), and then choose that Eval under Import to.

A CSV import can fill 5 Simulation fields. Only the opening message is required.

To start from a ready-made file, click Download template in the Upload a CSV file dialog. The template has one column for each of the 5 fields. Column headers don't need to match the template exactly. Common names like question, prompt, persona, outcome, or ground truth are matched automatically. Comma-, semicolon-, and tab-separated files all work.

Each Simulation field is matched to one column, and columns you don't use are listed under Not used and ignored. If an expected answer is over 1,000 characters, you're asked what to do with it: Leave them out (the default) imports the row without that answer, Cut to 1,000 characters keeps the start of the answer, and Skip these rows leaves the whole row out.

If a Channel or Handoff value isn't one Fin recognizes, such as WhatsApp or Yes, a warning appears on that field.

You can also choose Skip rows to leave those rows out of the import.

After the import, a confirmation shows how many Simulations were created. Each row appears in the Eval as a Simulation named after its opening message, ready for you to click Run all. Imported Simulations use the Test customer and all brands, so open a Simulation to choose a specific customer. Tags are added to the Eval, not in the CSV file. When you import inside a Release, the Simulations are added to that Release.

If your rows would take an existing Eval over 50 Simulations, you see how many still fit and can select Import first [number] to import just those. If any rows are rejected when you import, nothing is saved. Each problem row is listed with the reason, and you can select Import without these rows to import the rest.

Tags group related Evals so you can find them quickly as your suite grows. For example, tag by topic (Refunds), team or owner (Billing team), type of test (Regression, Escalation), or language (French). An Eval can have as many tags as you need.

To see and create tags, click Tags at the top right of the Evals page. The Tags panel lists each tag and how many Evals use it, and you can find a tag with the Search tags box.

To rename or delete a tag, click Tags at the top of the Evals page, or select Manage tags in the tag picker. The panel shows how many Evals use each tag. Deleting a tag removes it from every Eval, including Evals in Releases.

A Scorecard adds quality scoring to an Eval, such as Clarification or Efficiency, on top of its pass/fail checks. To add one, open the Eval and click Add scorecard at the top right, next to New simulation. After you add it, a Quality evaluation section appears next to Simulation results. To create a Scorecard, add criteria, and set each criterion's score weighting and Critical setting, see the How to add a Scorecard to an Eval article.

To see how Fin performed across all of your Evals, look at the summary at the top of Test > Evals. It reflects the last time each Simulation was run, and it has two parts:

The summary updates with your search and filters, so filtering by a tag shows results for just those Evals. To compare earlier runs, open a Simulation and click the run history dropdown next to its result. It lists each run by date with its result.