Skip to content

Series · 7 parts Published July 29, 2026

Page speed watcher

A small system that measures a short list of important pages on a fixed device and network profile, several times a run, and reports a regression only when it is larger than the noise. It names the resource that changed, and it never blocks a deploy. Seven posts on the same system — one diagram at a time — with a cost breakdown and an engineering reference at the end.

  1. 01

    A page speed watcher on AWS for a few dollars a month

    The whole system on one page — a measurer, a comparer and a reporter — and the reason it takes several samples: one measurement of a web page tells you almost nothing.

  2. 02

    How a page gets measured

    The device profile that must never change, how many samples is enough, and the four things that quietly make two runs incomparable.

  3. 03

    How a speed regression is detected

    Comparing against a page’s own rolling spread rather than a fixed threshold, why the baseline rolls, and the slow drift no single comparison will ever catch.

  4. 04

    How a performance budget is set

    Where a fixed threshold genuinely belongs, why budgeting bytes beats budgeting seconds, and why a budget is a prompt for a conversation rather than a gate.

  5. 05

    How the speed report reads

    Naming the resource rather than the metric, correlating a regression to a deploy, and the two numbers worth putting in front of anybody who is not an engineer.

  6. 06

    What the page speed watcher costs

    About $12 a month. One Bedrock read per page load is the only line that grows; the queue, the table, the mail and the storage are rounding errors. Plus the three ways the bill could surprise you.

  7. 07

    Engineering reference: the page speed watcher architecture

    Same system, drawn purely for engineers. Service names, region, Lambda inventory, IAM scopes, the schemas and the exact model id.

What is a page speed watcher?
A small serverless system that measures a fixed list of pages on a fixed device profile several times a run, compares against a rolling baseline, and reports a regression when the change is bigger than the measurement noise. It names what got heavier and does not block anything.
Why not just use a hosted speed tool?
Plenty are good and this is a build-or-buy decision. What a small custom one gives you is your own page list, a baseline drawn from your own site rather than a global percentile, and enough runs per measurement to distinguish a regression from noise — which is the thing most tools get wrong.
Does it fail a build?
No. A page speed number is noisy enough that gating deploys on it produces false failures, and a check that fails wrongly gets bypassed within a month. It reports after the fact, with the deploy it correlates to.
What device profile does it use?
One you choose and never change: a mid-range mobile on a throttled connection is the usual answer. The absolute number matters far less than measuring the same thing every time.
What does it cost to run?
A few dollars a month for a handful of pages measured daily. See part six.
All posts