Fluidit.ai

Background

A page experiment needs more than generated copy: it needs rules for which changes are eligible, a way to deliver them, and measurements with a denominator.

Concept

The implementation generates page variants with AI, then allocates among enabled, approved experiment versions using Thompson sampling. Required check results gate allocation. The browser snippet currently applies copy and palette changes; layout and image payloads are not applied by this SDK.

Build story

Server and snippet share a runtime-validated response contract. Required checks cover brand rules, contrast and static payload validity; a missing check is not a pass, and static validation is not a browser render test. The measurement helper calculates rates from success and exposure counts, with a default minimum of 30 observations. The Autonomy Dial controls permissions, but there is no unattended background loop. CI runs regression tests and a guard for runtime exports with no production references.

Outcomes

Pre-launch. The source contains the delivery, measurement and allocation paths, with regression coverage for the shared contract, check gates, rate calculation and Thompson sampling. Source-code writeback remains a dry run with no external network calls. This review does not establish installed customer snippets, hosted migration state, production traffic or conversion lift.