Behavioral interviewing has drifted in a specific direction: away from rewarding well-told stories and toward demanding verifiable substance. Hiring teams got burned by polished candidates who underdelivered, and AI-assisted note-taking now makes every claim easy to revisit. So they grade for real proof, not pretty stories. The instruction interviewers receive is some version of: dig for specifics, distrust smoothness, reward numbers.
What "evidence" means in practice
An evidence-grade answer differs from a story-grade answer in checkable detail. Compare:
"I led a performance effort that made our API much faster and customers were really happy."
"Our p99 was 2.3 seconds and our biggest customer had it in their escalation. I profiled the hot path, found we were serializing the full object graph per request, and shipped a projection layer — p99 dropped to 400ms over three weeks, and the escalation closed."
The second answer isn't longer because it's padded. It's longer because every clause is a fact someone could verify: a baseline, a diagnosis, an action, a delta, an external confirmation. Baseline, action, delta, confirmation. That's the evidence pattern, and interviewers trained on 2026 rubrics are listening for exactly it.
Numbers you're allowed to estimate
Candidates freeze on quantification because they don't remember exact figures. Estimates are fine and expected. "Roughly," "about," and "on the order of" are all safe. What matters is that the kind of number is right: latency has a before and after, an incident has a duration and blast radius, a process change has a frequency ("deploys went from weekly to daily"). If you truly have no number, use an observable outcome instead. "The team stopped needing the workaround doc" is evidence too.
Never invent precision, though. "Exactly 34%" invites the follow-up "how was that measured?", and measured-claims-you-can't-back is the specific failure this whole trend exists to catch.
When the honest answer is small
Not every story has a big delta, and inflating a small one is the losing move. The evidence-era answer to a modest result is owning its size and locating the value elsewhere: "the optimization itself bought maybe 10% — the real outcome was the profiling harness, which the team still uses." Interviewers report this pattern, accurate small numbers plus clear-eyed assessment, as more credible than a parade of triumphs. In a round designed to catch embellishment, calibrated honesty stands out.
Prepare your stories as fact sheets
For each of your five or six core stories, write down before the loop: the baseline metric, what you specifically did, the delta, who confirmed it, and what you'd do differently. Five lines per story. This front-loads the recall you can't do live, and it forces you to notice which story has no substance before the interview. Better to discover that at your desk than in front of a bar raiser.
The polished-narrative era trained candidates to rehearse delivery. This era rewards rehearsing facts. The candidates who struggle are rarely the ones with weak experience. They're the ones who never itemized what their experience actually proves.