I'm Pratyush Shivam, CTO of Konfidio in Berlin. If I had to put my approach to building trustworthy systems in five words, it would be the line on this photo: be exact about what you prove.
Where trust goes wrong
Most broken trust in software doesn't come from a system lying. It comes from a system implying more than it actually checked. A green tick that only means "the file uploaded". A "verified" badge that only means "someone clicked a button". A record "on the blockchain" that people read as "true".
Each of those is a small gap between what was proved and what was claimed. Users fill the gap with their own assumptions, and sooner or later one of those assumptions is wrong.
Say the narrow thing
The fix is unglamorous: say the narrow thing, precisely. Instead of "verified", say what was checked, by whom, and when. Instead of "secure", say which property holds and under what condition.
I wrote a whole paper on this for permissioned ledgers. A ledger can prove that a document hasn't changed since it was anchored, that it existed by a certain time, which account approved it, and in what order. It cannot prove that the content is true, who was really behind an account, or that the document is legally valid on its own.
Three habits that help
- Write the guarantee as a sentence. If you can't state what a feature proves in one plain sentence, users won't know either.
- Write the non-guarantee next to it. Listing what a check does not establish is as useful as listing what it does.
- Show it in the interface. The words on the screen should match the guarantee, not the marketing.
Why it matters
Precise claims are easier to defend, easier to audit, and easier to build on. A system that promises a little and keeps every promise earns more trust than one that promises everything.
The full paper, "What a Permissioned Ledger Can and Cannot Prove About Business Documents", is open access: doi.org/10.5281/zenodo.24438467.
More of my writing is at pratyushshivam.com.
Top comments (0)