Behavioural interviews · Module 12a · Level 3: Company, level and delivery

Evidence quality: structure vs proof

Most candidates know STAR, so structure alone rarely separates them. The follow-ups do: "What did you do?", "Why not the simpler option?", "How do you know it mattered?" This module shows the difference between an answer that's well structured and one that proves the skill, with the same story told both ways.

A 3-minute conversation: Sam asks, Priya explains

0:00 / 3:19

Follow along on the page Pause on “your turn” questions

Sam I know STAR. Why do my answers still fall apart under follow-ups?
Priya Because structure isn't proof. Two answers can both have a situation, a task, actions and a result, and only one of them proves the skill. The follow-ups find out which: what did you do, why not the simpler option, and how do you know it mattered?
Priya Think of a courtroom. A lawyer's closing speech can be perfectly organised and still lose, because the jury decides on the exhibits. STAR is the order of your speech. The exhibits are the decision you made, the option you turned down, and the number that moved.
Sam Do interviewers really dig for that?
Priya They're trained to. The US government's structured interview guide tells interviewers to narrow in on the candidate's own actions, and to ask again when they hear "we did". Google's guide plans follow-ups in advance to get a high level of detail. And Amazon's recruiters say specifics are key: they want to know the numbers.
Sam So what does a strong answer prove?
Priya Three things. Ownership: a decision or action that was yours, said with "I". Judgement: a real alternative, why you turned it down, and what your choice cost. And impact: a before and after, how it was measured, and who it mattered to.
Sam How do I check my own sentences?
Priya Use the evidence ladder. Rung zero is adjectives: "I'm very proactive". Rung one is habits: "I usually profile first". Rung two is a specific action in a specific situation. Rung three adds checkable detail: forty customers, fourteen minutes down to fifty seconds. Put every action and result on rung two or three.
Sam Show me the difference.
Priya The same invoice export story, told twice. Answer A: the export was slow, we optimised it, it got much faster. Answer B: it paged me three nights running, I profiled it, and it went from fourteen minutes to fifty seconds. Then three follow-ups. A says "it was a team effort", "this was the obvious fix", and "customers seemed happier". B says "I chose to fix the cause", "a cache was quicker but could show stale totals", and "the same forty customers, and twelve tickets closed". A ends with one line of notes, B with seven.
Sam Can I practise that?
Priya Sort nine sentences by what they prove, then paste in one of your own answers. The check flags more "we" than "I", no alternative, a number without a baseline, and vague words like "improved" and "helped".
Priya Your turn. "Latency dropped to fifty milliseconds." What's missing?
Priya The baseline, how it was measured, and who it mattered to.
Sam The memory hook?
Priya "I, instead, so." Every key sentence needs an I, what you did; an instead, what you turned down; and a so, what changed, measured.
Sam Common mistakes?
Priya Mistaking structure for proof. Hiding behind "we". Claiming you led the whole thing when you built one part. A trade-off with no real alternative. A number with no baseline. And inventing detail: if you don't remember a number, say roughly.
Sam And the practice?
Priya Take your strongest story and run the evidence check. Write three sentences: I, instead of, so. Then tell it out loud and ask yourself the three follow-ups. Each answer should add new detail.

1. The idea

In court, a lawyer's closing speech can be perfectly organised and still lose, because the jury decides on the exhibits, not the speech. STAR is the order of your speech. Evidence is the exhibits: the decision you made, the option you turned down, the number that moved. Follow-up questions are the jury asking to see them.

Most candidates now know STAR, so structure alone rarely separates them. Two answers can both have a situation, a task, actions and a result, and only one of them proves the skill. Interviewers are trained to dig for the difference:

The US Office of Personnel Management's structured interview guide tells interviewers to "narrow in on the candidate's specific roles and actions", and to "seek clarification when candidates say 'we did…'" or are vague.
Google's re:Work guide describes follow-up questions planned in advance to "elicit a high level of detail", detailed notes, and rubrics so every interviewer scores answers the same way.
Amazon's recruiters put it plainly: "specifics are key", and "we want to know the numbers".

The notes are what get scored (Module 01). A sentence that gives the interviewer nothing to write down doesn't count, however well it's placed in the structure.

Three things a strong answer proves

Proof	The follow-up that tests it	Counts as evidence	Doesn't count
Ownership	"What did you do, as opposed to the team?"	A decision or action that was yours, said with "I" and a verb, and what would have happened without you	"We", "I was involved in", "I owned it" with nothing you did
Judgement	"What else did you consider? Why not that?"	A real alternative, why you turned it down, and what your choice cost	"It was the obvious choice", or an option nobody would pick
Impact	"How do you know it mattered?"	A before and after, how it was measured, and who it mattered to	"Much better", "everyone was happy", a number you can't explain

The evidence ladder

Every sentence in your answer sits on one of four rungs. A perfectly structured answer can sit entirely on the bottom two.

0Adjectives. "I'm very proactive." A claim about yourself; nothing to write down.
1Habits. "I usually profile before I optimise." Still a claim, not something that happened.
2A specific action in a specific situation. "I profiled the export and found one query running once per customer." Now there's a line for the notes.
3Checkable detail. "…on the same 40 customers, from 14 minutes to 50 seconds; billing reused the change in two other exports." Strong evidence.

Aim to put every action and every result on rung 2 or 3. Adjectives and habits are fine as a bridge, never as the proof.

Proof has to be trueNever invent a decision or a number to pass these follow-ups. Interviewers compare details across questions and rounds, and reference checks can ask about them. If you don't remember an exact figure, say it's approximate (Module 12). If a story can't prove a skill, use a different story (Module 03).

2. How it works

The same story, told twice. Both versions follow STAR. Then the interviewer asks the three follow-ups, and you can watch what reaches each set of notes.

Answer A · structured

SOur weekly invoice export was slow and customers complained.

TWe needed to make it faster.

AWe analysed the problem, optimised the queries and improved the pipeline.

RThe export became much faster and customers were happier.

Answer B · structured and proven

SThe export for our 40 largest customers was timing out; it paged me on call three nights running, though it wasn't my team's service.

TI decided to fix the cause instead of restarting the job again.

AI profiled it, found one query running once per customer, and batched it into one, after ruling out a cache.

R14 minutes down to 50 seconds on the same 40 customers; 12 support tickets closed.

The full version of Answer B, with every follow-up answered in advance, is the model answer in section 5.

3. Try it yourself

Evidence check: what does each sentence prove?

Each sentence comes from a behavioural answer. Pick what it proves: ownership, a trade-off, impact, or nothing an interviewer could write down.

Check your own answer

Paste or type one of your answers. The check looks for the words that usually carry evidence, and the ones that usually hide it. It's a rough word check, not a score: it can't tell whether a claim is true, or yours.

4. What interviewers look for

Evidence is scored sentence by sentence: each one either gives the interviewer something to write down or it doesn't.

Signal	Weak	Good	Strong
Ownership	"We" throughout; your part unclear	"I" for some actions	Your decisions named, and what would have happened without you
Judgement	No alternative mentioned	An alternative named	A real alternative, why it lost, and what your choice cost
Impact	"Much better", "people were happy"	A number, without a baseline	Before and after, how it was measured, and who it mattered to
Specificity	Adjectives and habits	Some specific actions	Specific actions with checkable detail: systems, numbers, time frames, roles
Consistency	Details shift under follow-ups	Details hold	Each follow-up adds detail that fits what was already said

5. A model answer

"Tell me about a time you fixed something nobody asked you to fix." An invented answer, shown part by part. Read it for the shape, then build your own from a true story.

SituationThe weekly invoice export for our 40 largest customers was timing out. It wasn't my team's service, but I was on call, and it paged me three nights in a row.

OwnershipI decided to fix the cause instead of restarting the job again. I profiled it and found that one query ran once per customer: about 90% of the run time.

JudgementI considered a cache first. It was a one-day change, but invoices could have shown stale totals, and finance couldn't accept that. So I chose to batch the 40 queries into one, which took four days.

ActionBilling owned the code, so I wrote the change, paired with their tech lead on the review, and we shipped it behind a flag for one weekly run before switching it on for good.

ImpactThe export went from 14 minutes to 50 seconds on the same 40 customers. The timeouts and the pages stopped, support closed 12 open tickets, and billing reused the batching in two other exports.

Why it scoresIt's the same STAR shape as Answer A, but every key sentence answers a follow-up before it's asked. "What did you do?": I decided, I profiled, I wrote. "Why not the simpler option?": the cache, and why stale totals ruled it out. "How do you know it mattered?": a before and after on the same data, plus who noticed and what others reused.

6. Remember it

Memory hook: "I, instead, so"Every key sentence needs an I (what you did), an instead (what you turned down) and a so (what changed, measured).

How this hook works: what “I, instead, so” means

What the slogan means, phrase by phrase

“I”: Your action or decision, with a verb: decided, profiled, wrote, proposed. Credit the team, but say which part was yours.
“instead”: The alternative you turned down, and what your choice cost or saved.
“so”: The measured change: before, after, and who it mattered to.

Why this picture fits: The jury decides on the exhibits, not the speech. "I", "instead" and "so" are the three exhibits the follow-ups ask to see.

The story → the concept

The courtroom	Evidence quality
The closing speech	Your STAR structure
The exhibits	Evidence of ownership, judgement and impact
The jury asking to see an exhibit	A follow-up question
An exhibit that matches the testimony	Detail that fits what you already said
The verdict	The score, decided on evidence, not delivery

Watch it on an example

"I decided to fix the cause instead of restarting the job." → I
"I considered a cache; stale invoice totals ruled it out." → instead
"From 14 minutes to 50 seconds on the same 40 customers." → so

The trap it saves you from

A beautifully structured answer that sits on adjectives and habits: STAR with nothing to write down.

It still works when…

The result belonged to the team, or can't be measured. Say plainly which part was yours, and describe what you could observe: what stopped breaking, who stopped complaining, what others reused.

7. Common mistakes

Mistaking structure for proofSTAR tells the interviewer where to look. If there's nothing there, a tidy shape doesn't help.

Hiding behind "we""We" is right for the team's result. Your own actions need "I", or the interviewer can't credit you with them.

Over-claiming"I led the migration" when you built one part falls apart at "What did the others do?" Claim exactly your part; it's enough.

A trade-off with no real alternative"We could have done nothing" isn't a trade-off. Name an option a sensible engineer might have picked.

A number with no baseline"Latency was 50 ms" proves nothing on its own. From what, measured how, and did it matter to anyone?

Inventing detailMade-up numbers drift between follow-ups and rounds. Say "roughly", or pick a story you know well.

8. Practice exercise (25 minutes)

Pick your strongest story (2 min). The one you'd tell first in any loop.
Run the evidence check (8 min). Paste it into the tool above. Fix each flag, or note why it doesn't apply.
Write the three proofs (10 min). One sentence each: "I…" (your decision), "instead of…" (the alternative and its cost), "so…" (before, after, who it mattered to).
Say it out loud (5 min). Tell the story with the three sentences in it, then ask yourself the three follow-ups. Did each answer add new detail?

9. Self-check quiz

Two answers both follow STAR. What usually makes one score higher?

Structure shows where to look; the evidence is what gets scored.

Which sentence is the strongest evidence of ownership?

Your action, with a verb, and a sign it wouldn't have happened without you.

Which one is a real trade-off?

A sensible alternative, why it lost, and what the choice cost.

"Latency dropped to 50 ms." What's missing?

A number without a before and a reason it mattered is hard to score.

You can't remember the exact number. What's best?

Honest and approximate beats precise and invented.

10. Go deeper

Resource	Time	Why
A guide to structured interviewing (Google re:Work)	15 min	How interviewers plan follow-ups, take detailed notes and score against shared rubrics.
Structured interviews (US Office of Personnel Management)	20 min	The probes interviewers are trained to use, including clarifying "we did…".
11 Amazon interview tips from recruiters and hiring managers	10 min	"Specifics are key" and "we want to know the numbers", from the people who run the loops.
Module 04: Ownership and results	Revisit	The result sharpener turns a vague result into a before and after.
The question bank	Practice	Run the evidence check on one answer for each question type.