How do I tell if a candidate can actually think, not just prompt?
Short answer
Ask for the decisions, not the output. Someone who only prompted can show you working code but cannot tell you what they rejected or why. Someone who thought can name the option they dropped and the cost of dropping it. Reading a few weeks of their dated entries makes the difference obvious quickly.
How it works
Weak tell
It works, and no account of why it is built this way.
the output is fine; nothing sits behind it
↓
Strong tell
A named tradeoff, and what it cost.
you cannot borrow this from a model
↓
Strongest tell
A recorded change of mind, with the reason.
rare, and almost never faked
↓
Where to look
Weeks of dated entries, not one polished artefact.
consistency over time is the hard part to manufacture
On Proof today
4,100+
logs written
900+
engineers writing them
Counted live, and excluding one account that flooded the platform with automated rows in July–August 2026. Every public number on Proof excludes it.
What this does not do
Using AI well is a real skill and this is not a test for whether someone used it. It is a test for whether they can account for their own choices.
Has this actually worked for anyone? One hire, every step, with the denominator — including what it does not prove.
Related questions
Published
September 2026
Made by
Radhakrishnan Selvaraj
Candor and Innovation
PROOF.ZEROMAINTENANCEENGINEER.IN