Analytics filters: scope by model and category, plus Sanitize Execution
Two updates to the Analytics page that make the numbers you read more honest.
Filter by Models and Categories
The new filter row lets you scope every KPI, every table, and every chart on the page by:
Models — pick which LLMs to include (ChatGPT Fast, Claude, Perplexity, etc.). The filter applies workspace-wide on the page, so the Mean Sentiment card reflects only your selection.
Categories — pick which categories from your Experiments to include. Useful when you want to look at "Competitor Alternatives" categories in isolation, or zoom out to a single product line.
Both filters chip up so you can see your selection at a glance, and clear with one click.
Sanitize Execution
The bigger update is the Sanitize Execution toggle underneath the filter row.
When you run an Experiment across multiple LLMs, individual model calls sometimes fail — rate limits, timeouts, transient model errors. The default behavior is to include all results that came back successfully. But this can create an apples-to-oranges comparison: an execution where ChatGPT failed gets averaged in alongside one where every model returned cleanly.
Toggle Sanitize Execution on and Analytics only counts executions where every selected model returned at least one successful query result. The trendline gets a lot quieter, because the noisy executions are excluded — and the comparisons stop lying.
Why it matters
Trends are only useful if every point on them is comparable. Sanitize Execution exists because we kept finding teams reading drops or spikes that were really just one model failing to return that week. With the toggle on, what you see is what's actually moving.
Available today
Both controls are at the top of the Analytics page. The filter selections persist across tabs; Sanitize Execution defaults to off so you can compare with the unfiltered view at any time. The Insights Agent respects both settings when answering questions over the same data.
