Sitearc
Measure and learn

Every shipped change is measured against pages it didn't touch.

When a change ships, Sitearc saves a baseline and holdout pages. After the window, the server decides: Won, Backfired or Inconclusive.

Demo data. Harbor & Pine is a fictional boatyard.

What it shows

  • 01

    Changed and holdout pages

    The metric per page per day, for the pages that changed and the ones that didn't.

  • 02

    A baseline

    Search Console and GA4 readings from before the change, saved when it ships.

  • 03

    Three results

    Won, Backfired or Inconclusive. Inconclusive is a real answer, not a failure.

  • 04

    Predictions

    Each change predicts a direction and a window, never a percentage.

  • 05

    Confounders

    Algorithm updates, campaigns or incidents that could skew a window, logged with their dates.

  • 06

    What it taught

    Each result can become a learning that the next proposal sees.

What your agent can do

Your agent works with Experiments through tools like these. Some only read. Others may change something or spend credits.

app.sitearc.ai/mcp7 of 130 tools
  • ship_experimentMay change or spend

    Records an approved, merged change as an experiment and saves its baseline.

  • list_experimentsReads

    Lists running and concluded experiments, with their windows.

  • get_experimentReads

    Reads one experiment: its pull request, changed pages, holdout, baseline and result.

  • get_measurementsReads

    Reads the baseline and the measurements taken since the change shipped.

  • conclude_experimentMay change or spend

    Asks the server for the result. The agent cannot choose the outcome.

  • record_confounderMay change or spend

    Logs an event that could skew a window, like an algorithm update.

  • list_confoundersReads

    Lists confounders and the dates each one covers.

How it works

  1. 1

    A person approves the preview, and the pull request merges in GitHub.

  2. 2

    Your agent records it as shipped. Sitearc saves the baseline, the changed pages and the holdout.

  3. 3

    Measurements build up over the window, beside the pages that didn't change.

  4. 4

    The server concludes the result. Your agent can ask for it, but can't pick it.

Limits and credits

Know the edges before you rely on it. Credits pay for outside data and heavy work, and saved data never costs any.

Limits

  • Only changes shipped through a GitHub pull request are measured.
  • An unresolved confounder in the window forces Inconclusive. Only a person can resolve one.
  • Predictions name a direction and a window. They never promise a size.
  • Sitearc refuses to record a change as shipped if your agent's preview check failed.

Credits

0This feature spends no credits. Reading saved data is free on every plan.

One credit is $0.001. Pro includes 100,000 a month, and Free includes none. See pricing

Connect your first site free.

Add every site you run, at no cost. Your agent can start the same day.