Experimentation, without the hand-waving
Sample size math shown in full, architecture tradeoffs stated plainly, and honest vendor comparisons, including where we lose. Ten pieces in progress; outlines below.
A/B vs. multivariate testing, and when the extra complexity pays off
A/B tests are simpler, multivariate tests see interactions A/B can't. A traffic-based decision framework for choosing between them.
Read the outline
Why anonymous-to-identified linking silently corrupts your experiment results
The moment a visitor logs in, most experimentation stacks quietly double-count them. How identity resolution breaks experiments, and how to link without re-keying.
Read the outline
Client-side, server-side, or API-first, where your experiments should run
Snippet flicker, ad blockers, and ITP are eroding client-side testing. What each architecture can and can't test, and what it costs you.
Read the outline
Why you can never edit a running experiment
Changing weights, adding variants, or tweaking copy mid-flight doesn't refine an experiment, it invalidates it. The statistics behind config immutability.
Read the outline
Sample size math for multivariate tests, honestly
Most MVT guides hand-wave the sample size problem. Worked power calculations for factorial designs, and how fractional factorials cut the bill.
Read the outline
Control cohorts, why every campaign needs a true holdout
Without a holdout you're measuring attribution, not lift. Global vs per-campaign controls, and why TraqLyte makes the control a first-class cohort.
Read the outline
One user, one variant, every channel
Your email tool and your website are running different experiments on the same person. How to keep variant assignments consistent across email, web, and ads.
Read the outline
At-least-once delivery, or why your conversion counts are wrong
Retries, offline queues, and double-fired pixels inflate conversion counts unless you dedupe deliberately. Idempotency design for outcome events.
Read the outline
Securing a public tracker snippet, site IDs, origins, and signatures
A tracker's site ID ships in your page source, so it can't be a secret. The layered model that keeps a public identifier from becoming an open write endpoint.
Read the outline
The 2026 experimentation landscape for teams without a data science team
An honest map of the A/B testing and experimentation market by team size and technical capability, including where TraqLyte does and doesn't fit.
Read the outline