The best-looking thing on my portfolio was a Maritime API Contract Inspector. A five-stage pipeline, fully designed, polished diagrams, the works. It read beautifully.
It had never been built. Not a prototype, not a branch. Zero lines shipped.
So I deleted it.
The reasoning took me longer than it should have, because deleting your most impressive page feels like removing your best argument. It is the opposite. A polished presentation of something unverifiable does not add credibility, it spends it. Anyone who asks a single follow-up question finds the hole, and the moment they do, everything else on the page becomes suspect too. One unbacked claim is not one bad item in a list. It is a discount applied to the whole list.
What replaced it
Two things made that work, and neither of them was discipline.
Structure as a forcing function. The evidence lives in a typed data array, not in free-form page copy. Adding an entry means filling in fields: what happened, where is the proof. There is no slot for vibes. When the shape of the data has nowhere to put a vague claim, you stop writing vague claims. That is a better mechanism than intending to be honest.
Failures count as evidence. The entry where an agent's fix was wrong and I reverted it is more convincing than any of the successes, because it is the one that proves the review loop actually runs. Anyone can list wins. A list with no losses in it is a list nobody audited.