Graham McNicoll

We spent three years building our own experimentation platform at education.com. Then we found a bug in the stats package. The immediate question was not a small one: did we just invalidate every product decision we made over the last three years? Fortunately it was minor. The risk was real, and it forced an honest question. We were experts in building products for our edtech product. Statistical analysis was a different domain entirely. That gap is where errors live. This is what most engineering teams underestimate when they choose to build. You can estimate the engineering effort. But a bug in a homegrown stats package can invalidate every result you've made. You may not find out until years of decisions have already been made on bad data. GrowthBook's statistical library is open source, vetted by data science teams across the industry, and available for you to audit yourself. That is what battle-tested means. Your product decisions rest on a foundation you can actually verify. Take a look at the stats library yourself at growthbook.io

Only one snapshot so far — the engagement-over-time curve appears once the daily scrape has captured this post at least twice.