Experiments and flags

1. The question this answers

Your experiment tool tells you that variant B converts 12 % worse. It does not tell you why, and it cannot show you a single person struggling with it.

This bridge closes that gap in the cheapest possible way: it reads the variant assignment out of the tool you already run and attaches it to the MousePlayer session. From then on, "show me sessions that saw variant B and rage-clicked" is a filter, not a research project.

2. The tools, and the one rule they share

Six A/B tools and five flag providers, all read in the browser, all through the tool's own public API.

Kind	Tools
A/B and experiments	Optimizely, VWO, AB Tasty, Kameleoon, Google Optimize successors, Convert
Feature flags	LaunchDarkly, ConfigCat, Flagsmith, Split, Unleash

Only assignments that really happened

VWO is the clearest example of the rule: we take a campaign only when its status is RUNNING and a combination has actually been assigned to this visitor. A campaign that is merely defined tells you nothing about the person watching, and writing it onto the session would make every later filter lie.

Some answers arrive late

ConfigCat resolves asynchronously; Snowplow answers by callback. Their values land in a buffer and are picked up on the next pass rather than being waited for - a bridge that blocks the page to learn a variant name has already cost more than it is worth.

3. What it costs your page

Nothing measurable. There is no network request of ours, no extra script, and no work at all when the bridge is switched off.

The tool you already embed does its own work whether we read from it or not. What we add is one read, a handful of times, until the tool has an answer or twenty attempts have passed.

Why twenty and not forever

Experiment tools often load after us. A single read would miss almost every assignment; an endless poll would keep a timer alive on every page for the sake of a value that is not coming. Twenty attempts at half-second intervals is the compromise, and it is the same one the consent adapters make for the same reason.

4. Where this earns its keep

• A variant that tests worse for a reason nobody can name. Filter to that variant, sort by rage clicks, watch three sessions. The reason is usually visible in the first one.
• A flag rolled out to 5 % of users. Those five percent are exactly the sessions you want to watch, and without this bridge there is no way to find them.
• A test that looks flat. Sometimes both variants are equally bad in the same place, and the experiment tool cannot tell you that - it only compares totals.