Evals
Task scoreboard
Last updated 10 November 2026
How often Senova finishes the tasks people give it, by kind of task, over the last 7 and 30 days. Failed tasks become test cases we use to make Senova better, and this page shows whether that is working.
There isn't enough data to publish a figure yet. We started counting on 10 November 2026, and we only show a number once a category has at least 20 tasks.
1. Last 7 days
| Category | Tasks counted | Finished |
|---|---|---|
| Bookings | Not enough yet | No figure |
| Research | Not enough yet | No figure |
| Reminders | Not enough yet | No figure |
| Email drafts | Not enough yet | No figure |
| Phone calls | Not enough yet | No figure |
| All categories | Not enough yet | No figure |
2. Last 30 days
| Category | Tasks counted | Finished |
|---|---|---|
| Bookings | Not enough yet | No figure |
| Research | Not enough yet | No figure |
| Reminders | Not enough yet | No figure |
| Email drafts | Not enough yet | No figure |
| Phone calls | Not enough yet | No figure |
| All categories | Not enough yet | No figure |
3. How we measure
Which tasks count. Only chats from people who chose to help improve Senova, started after they chose, in personal workspaces. We leave out the same chats our privacy notice says we never use: temporary and team chats, the developer API, agents people publish for their customers, and chats with sensitive information such as health, money or identity details.
Which kind of task it is. We sort each request by the tools Senova used (a phone call, a reminder, an email draft, a web search or the browser) and then by its words (“book”, “remind me”, “look up”). Requests that fit none of the five kinds aren't counted.
What “finished” means. Senova gave a real answer and no step failed. A task counts as not finished when a browser step or another tool failed, Senova ran out of steps, the reply came back empty or cut off, the person pressed Stop, or the person later gave the reply a thumbs down.
What we publish. Counts and percentages only, rounded: task counts down to the nearest ten, rates to the nearest whole percent. A kind of task with fewer than 20 tasks in the period shows no figure. Nothing on this page comes from what anyone wrote.
Limits. These are real tasks, not a fixed test set, so the mix changes from week to week. Finishing a task isn't the same as getting it right: a reply nobody marked down still counts as finished. A thumbs down is counted on the day it is given.
4. Comparisons with other assistants
None yet. A fair comparison needs the same tasks run on each assistant, and we haven't done that, so we don't show figures for anyone else.
5. What happens to a failed task
For people who chose to help, a failed task is kept as a test case: their request with their own name, contact details, card and ID numbers and secrets masked and links cut back to the website name, the names of the tools used, and how it failed. Staff review it and either add it to our test set or discard it. Deleting the chat deletes its case. See How Senova keeps you safe.
Questions? Email hello