I refused to do R&D claims
the hard way one more time.

For 7 years, I ran a technology company. Every quarter, the R&D claim process felt less like compliance and more like creative writing.

I would sit down with my engineers and ask, "What was hard about that feature you built three months ago?"

They didn't remember. They had moved on. So we would piece together half-remembered challenges, polish them up with technical jargon, and hope it sounded convincing enough for HMRC.

I knew that if an auditor actually dug into the details, our "stories" were flimsy. Not because we weren't doing R&D, but because our memory of it was flawed.

The frustration peaked during a mock audit. I was struggling to justify a claim about a complex data pipeline. The narrative felt weak.

I opened the git history for that period. It was all there. The 12 failed attempts. The revert commits. The PR comments debating race conditions. The 2 a.m. hotfix that finally solved it.

The code didn't lie. It didn't forget. It contained the exact timeline of technical uncertainty and advancement that the tax authorities required.

We were spending weeks writing subjective stories, when the objective evidence was sitting in our version control system the whole time.

For years, this problem was unsolvable. You couldn't just send a tax inspector a zip file of your git log. The gap between raw code and a human-readable narrative was too wide.

This is where my PhD comes in. With the recent breakthroughs in Large Language Models, we can finally bridge that gap.

Evidencestack uses the kind of semantic analysis I studied for years to read your commits, understand the engineering intent, and translate it into the format HMRC requires.

It's the tool I wished I had. It turns the most painful part of my quarter into a 10-minute automated task, backed by evidence that actually stands up.