Getting started

Your first data

The SDK is installed and signals are flowing. Here is what to do with them.

The SDK is installed and signals are flowing. Here's what to do with them.

Read your scores

Open the dashboard. Every tracked page has a confusion score from 0 to 100. Higher is worse. A page at 15 is fine. A page at 60 has real friction. A page at 80 is actively costing you conversions.

The score accounts for signal type, frequency, and recency. A single rage_click on your checkout button today matters more to the score than 10 passive_drift signals from last month.

New sites take 24-48 hours to produce meaningful scores. You need enough signal volume for the clustering to kick in.

Look at your first issues

Issues are what you act on. Each one has a title, an affected element, a signal count, and a severity score. Sort by severity first -- those are the ones where friction is concentrated on a specific element, not just diffuse across a page.

The hypothesis on each issue is the scoring engine's best guess at root cause. It's not always right. Use it as a starting point, not a verdict.

Issues work like tickets. Assign them, update the status as you investigate, mark them resolved after you deploy a fix. The scoring engine re-checks resolved issues after each deploy to confirm the fix held.

Wire conversion data

If you want revenue impact estimates, add track() calls around your key conversion events:

track('plan_intent', { plan_id: 'scale', amount_cents: 9900, billing: 'monthly' })
track('subscription_started', { plan_id: 'scale', amount_cents: 9900, billing: 'monthly' })
track('checkout_abandoned', { plan_id: 'grow', amount_cents: 3900 })

The scoring engine correlates friction signals with abandonment and estimates how much revenue each open issue is likely costing. It takes a few days of data to produce stable estimates.

Set up at least two alerts

Don't wait until users complain. Set up a spike alert for your most critical page, and a budget alert as a ceiling:

Spike alert -- fires when a page score increases sharply after a deploy. Catches regressions before they compound.

Budget alert -- fires when a page crosses a score threshold you set. Your checkout page probably shouldn't ever be above 50.

Both take 2 minutes to configure under Settings > Alert Rules. Route the spike alert to Slack or email. Add PagerDuty to the budget alert if the page is critical enough.

Tag your deploys

When you ship a fix, tag the deploy with a version string:

init({
  key: process.env.NEXT_PUBLIC_FLUSTERDUCK_KEY!,
  segment: { app_version: '2026.06.10' },
})

Flusterduck records the confusion score before and after each tagged deploy. This is how you confirm a fix worked -- the scoring engine shows you the before/after delta and verifies that the affected issues actually resolved.

Connect your AI assistant

Once your site has a few days of data, the MCP integration pays off fast. Instead of checking the dashboard, ask Claude:

"What's the worst-performing page right now, and what's causing it?"

See [MCP setup](./mcp) for a 5-minute configuration walkthrough.

Add your teammates under Settings > Members. Engineers can triage issues. PMs can read scores and run weekly summaries. Admins can manage alert rules and integrations.