In plain terms: My portfolio used to showcase a beautifully presented project that, honestly, only existed as a design — it was never built. I removed it and replaced it with a list of smaller, less glamorous things that really happened, each traceable to actual saved work. Real and modest beats impressive and unverifiable.

My portfolio’s AI page used to feature a “Maritime API Contract Inspector” — a fully-designed five-stage pipeline with polished diagrams. It read great. It was never built.

I deleted it. Polished presentation of something unverifiable doesn’t add credibility, it spends it: anyone who asks one follow-up question finds the hole, and now everything else on the page is suspect too.

What replaced it is a data file of evidence entries sourced from actual diffs and commit messages — the payroll parity bug an agent audit caught (including the human revert of its one wrong fix), the HTTPS downgrade root-caused on this very site, the authorization hardening pass before a release. Each entry traces to a commit someone could go read.

Two things made this work:

  • Structure as a forcing function. The evidence lives in a typed data array, not free-form page copy. Adding an entry means filling in fields like what happened and where’s the proof — there’s no slot for vibes.
  • Failures count as evidence. The entry where the agent’s fix was wrong and a human reverted it is more convincing than the successes, because it shows the review loop actually runs.

What I learned

A shorter list of true things beats a longer list of plausible ones. Deleting my most impressive-looking page made the portfolio more convincing, not less — because every remaining claim survives a follow-up question. It’s the same principle this whole garden runs on: everything traces back to something I actually shipped or broke, and the failures earn more trust than the wins.

Related: Letting an AI hunt for the bugs humans skim past, Write the plan before the code