Download websites that use Split

Feature flags begin as an engineering habit and only later become something a company buys. A team wraps a risky change in a conditional, ships it switched off, turns it on for a slice of traffic and keeps the switch within reach. Once that habit spreads across a whole organisation somebody has to own the switches, and at that point a platform gets purchased.

The product recorded here belongs to that market, filed under A/B Testing, dating to 2015, sold as Proprietary software by Harness. A vendor name that differs from the product name is usually the trace an acquisition leaves in an index like ours, and the entry is worth reading with that in mind. The product name itself is a common English word attached to more than one piece of software, so what follows counts pages matching a specific signature rather than everything wearing the label.

There is a deeper reason to treat the number as a lower bound. Flagging is mostly a server-side discipline, and the decisions that matter frequently never reach the browser at all, because the server picks a branch and sends finished markup. Only the client-side SDK is visible to a crawler, so a company can run the platform across its entire backend and show us nothing. As of October 2026 we see the browser SDK on 15 sites, the visible corner of a much larger deployment surface.

A match still says something concrete about a site. A browser SDK for flagging on a production page means releases have been decoupled from deployments, that experiments are running against real traffic and that there is enough engineering process in the building to justify paying for the machinery. Sites in that group lean toward software companies and large retailers with in-house teams, and away from agency-built brochure sites, where a controlled rollout has nowhere to live.

Where the data gets genuinely interesting is in what the SDK reveals about itself. We record 7 distinct versions across the matched sites, with 10.24.1 recurring through the group. Version spread is a maintenance signal, because a JavaScript SDK is a dependency somebody has to bump deliberately, and a site sitting several releases back is telling you how often its front-end dependencies get touched at all.

A per-domain lookup answers a narrow question, namely whether one company runs flags in the browser. The downloadable list answers the questions that need a whole population at once. Who runs experimentation client-side and who does not. Which versions are still deployed when a security note goes out. Which visible users of the platform a competing vendor could approach. A single lookup cannot build a population, and the population is where those answers live.