Comparing runs

How a comparison works#

Drop a before and an after capture. Both go through the exact same deterministic engine (see why that matters), then the two results are diffed: an overall verdict, the key changes ranked by impact, deltas for every headline metric (score, transfer bytes, request count, load time), and the request-level changes — what's new, what's gone, what got slower or heavier.

Because the engine is deterministic, any difference in the comparison is a real difference between the captures — capture both the same way (same browser, cold load, page finished) so the diff reflects your change, not your method.

Compare against a live capture#

Signed in on a Patrol or Fleet plan, the compare page also offers HAR vs live URL: pick a baseline — upload a HAR, or choose one of your existing analyses — give it a page URL, and Harpoon captures that URL live before diffing it against the baseline. It's the same headless-browser engine and the same usage limits as the Capture page, so a live comparison counts toward your plan's capture usage the same way a standalone capture does.

If the live capture fails — a timeout, a blocked navigation, anything the browser hits — you get the same failed-step detail as Capture, with a retry that re-runs just the capture. The baseline (your upload, or the resolved analysis) is never re-uploaded on retry.

The AI comparison report#

The deterministic diff renders instantly. Signed in, the AI's read on the comparison loads alongside it asynchronously — same trust rules as the analysis AI report: the model sees only the diff the engine produced, never the raw captures. Anonymous comparisons get the deterministic report inline, with no model call.

Permalinks, export & sharing#

• Signed-in comparisons are saved and get a permalink you can revisit and send to teammates.
• Export as PDF, HTML, Markdown, or CSV — see Sharing & export.
• Create an opt-in public link for people outside your org — see public share links.