Ten Points, Five Parts: How a Points Rubric Scores a Screen

In July 2026 we changed how many HireQwik screens add up. Until then, almost every job used a weighted average: score each dimension from 0 to 10, multiply by its weight, and average. It worked, but there was a question the average answered badly: “Which part of the screen did this candidate lose marks on?” A 6.2 average could hide a strong motivation answer and a weak domain answer, or the reverse, and the two candidates looked identical.

Points-based interview scoring answers that question directly. Each part of the screen gets its own budget, the evaluator scores within that budget, and the total is a plain sum. This post explains where all 10 points go, how a written rubric level becomes a number of points, why communication has a floor, what the domain bonus does, and when a weighted average is still the better choice.

What points-based interview scoring means

Points-based interview scoring is a method where each part of a screening interview has a fixed point budget, the candidate is scored within each budget against written level descriptions, and the scores are added into one total out of 10. In HireQwik the total is then compared with the job’s bands to set the verdict tier.

The key difference from a weighted average is visibility. With points, you can read the breakdown and see that a candidate earned 2.5 points on question one but only 1 on question two. With an average, you get one blended number and have to dig for the parts.

The budget: where all 10 points sit

HireQwik’s points rubric follows a fixed layout from our screening playbook. The budgets do not change from job to job; what changes is the question text and the level descriptions written for each role.

The three core questions plus communication make up 10 points. The bonus can push the raw sum above 10, but the total is capped there, so the bonus works as a cushion: it can make up for a point lost elsewhere, but it can never lift anyone past a perfect score.

Why the three questions are not worth the same

People sometimes ask why the third question is worth only 2 points. The answer is about what each question can reliably show in a short voice screen.

The first two questions, motivation and ownership, are where an entry-level candidate has the most room to show real substance. A fresher with no job history can still explain why they want this role and describe a time they took responsibility for something in college, an internship or a part-time job. Those answers separate candidates well, so they carry the most weight.

The domain check is different. It asks for conceptual understanding of the field, such as how a sales pipeline works or why a support ticket gets escalated. That matters, but in a first screen it is more about whether the basics are in place than about ranking people finely. A smaller budget keeps one tough concept question from dominating the result, especially for freshers who may not have met the vocabulary yet. Tuning those expectations for early-career pools is covered in calibrating a rubric for freshers.

How a rubric level becomes points

For each core question, the job’s rubric contains written descriptions of four levels, numbered 0 to 3. Level 3 describes a strong answer, level 0 a failing one, and levels 1 and 2 sit in between. The evaluator reads the candidate’s answer, decides which level description it matches best, and scales that level to the question’s budget. Decimals are allowed, so a half point is a normal result.

On a 3-point question the scaling is direct: level 2 is worth about 2 points. On the 2-point third question, the same level is scaled proportionally, so level 2 is worth roughly two-thirds of the budget, a little over 1.3 points. That is why you will sometimes see a third-question score that is not a round number.

Here is an invented level set for the ownership question on a customer support job:

• Level 3: Describes a specific situation they owned, the actions they personally took, and a clear result, with some reflection on what they would do differently.
• Level 2: Describes a real situation and their actions, but the result or their personal role is vague.
• Level 1: Talks about ownership in general terms (“I always take responsibility”) with no concrete example.
• Level 0: Cannot give an answer, or the answer is unrelated to the question.

A candidate who says “In my final-year project our data collection was late, so I called the remaining respondents myself over two weekends and we submitted on time” is a clear level 3. One who says “I am a very responsible person, my friends always rely on me” is a level 1, however warmly it is said.

One safety detail is worth knowing. The rubric text and question text that HR writes are passed to the evaluator as data, clearly marked off, with an instruction never to treat them as commands. A rubric is a description of good answers, not a place where anyone can change how the evaluator behaves.

Communication: why it starts at half a point

Communication is scored once, holistically, across all of the candidate’s answers, on a band from 0.5 to 2 points. The playbook sets the 0.5 floor, and I read its reasoning as simple: if someone took part in the screen and answered questions, their communication was at least minimally functional, so it earns something. In practice the floor stops a candidate who is quiet or nervous from being double-penalised on top of thin content answers.

There is one exception. A score of zero still appears when there was nothing to assess at all, for example a call that ended almost immediately. In that case the zero is a marker for missing evidence, not a judgement.

By default communication is judged from the transcript. Where an account has speech assessment enabled, a delivery score measured from audio is first converted into the 2-point budget, and the two halves are then averaged. Because the whole communication budget is 2 points, the voice side can move a points total by about one point at most. Reading the two halves of a communication score shows where each number lives.

The domain bonus, and the cap at 10

The domain bonus is worth up to 1 point and rewards direct, role-specific signal, such as a candidate for a fintech sales role who has already sold a financial product. The playbook’s default guidance is that direct domain work earns the full bonus, adjacent experience earns half, and none earns zero. A job can write its own anchors for the bonus.

Because the total is capped at 10, the bonus never inflates a score past the maximum. Its real effect shows up near the bands. Take an invented candidate with 2.5 and 2.5 on the opening pair of questions, then 1.5 and 1.5 on the domain check and communication, a total of 8.0. That is Go on the default points bands, which start Go at 7.5 and Strong Go at 9.0. Add a full domain bonus and the total becomes 9.0, which is Strong Go. A candidate with genuine domain background is moved up a tier, which is exactly what most hiring managers would want.

How the points total sets the verdict

With every part scored, the sum is checked against the job’s bands. The defaults on points jobs are 9.0 and above for Strong Go, 7.5 for Go and 6.0 for On Hold, with anything below 6.0 a No Go. A Strong Go on this path is also flagged as priority, so recruiters see it first. Bands can be changed per job, and the full sequence of checks is in the tier rules.

One guard applies only to points jobs. If fewer than two core questions got scored, the outcome is Incomplete instead of a low total, because a partial call should never look like a weak candidate. A call with two of the three scored still gets a total, with the missing question adding nothing, an edge case set out in the incomplete interview explainer.

Writing rubric levels that score fairly

Level descriptions decide how well a points rubric works. Here are four habits I would suggest to anyone writing one:

1. Describe behaviour, not adjectives. “Gives a specific example with their own actions and a result” can be matched against a transcript. “Excellent answer” cannot.
2. Make each level clearly different from the next. If levels 2 and 3 differ only by the word “very”, the evaluator has nothing to separate them with, and neither would a human panel.
3. Write level 0 and level 1 as carefully as level 3. Most real disagreement happens at the bottom of the scale, where “weak but trying” has to be told apart from “no answer at all”.
4. Keep the role’s reality in mind. A level 3 for a campus hire should be reachable by a strong final-year student, not only by someone with five years of experience.

This approach has a long history in human assessment. Behaviourally anchored rating scales, described in this overview of BARS, use concrete example behaviours at each level for the same reason: it makes ratings more consistent between raters. An AI evaluator benefits from clear anchors in exactly the way a human interviewer does.

Points vs weighted average: which to choose

Neither method is better in general. They suit different jobs:

Whichever you choose, the score rests on evidence from the transcript, and the arithmetic stays plain. The full scoring walkthrough follows both methods from transcript to total.

A worked example: one candidate, part by part

Here is how a full points screen might read for an invented candidate on a customer support job.

On the motivation question she explains that she wants support work because she enjoyed fixing classmates’ laptop problems, and she names two things she knows about the company’s product. That matches level 2: a real reason and some homework, but no link to what the job involves day to day. About 2 of 3 points.

On the ownership question she describes running her hostel’s food committee when the caterer cancelled before a festival, calling three alternatives and settling on one within a day. Specific, personal and with a result: level 3, so 3 of 3.

On the domain check, asked when a support ticket should be escalated, she says “when the customer is angry.” That is partly right but misses the idea of issues beyond the agent’s access or authority. Level 1 on a 2-point question, roughly 0.7.

Her communication across the call is clear and organised, though some answers run long: 1.5 of 2. She has no direct support experience, so no domain bonus.

Her total is about 7.2. That is On Hold on default bands, just under the 7.5 line for Go. A recruiter reading the breakdown sees immediately why: two strong answers, one thin concept answer. That is a far more useful picture than a single blended number, and it tells the recruiter exactly what to probe in a second conversation.

What a points rubric will not fix

A points layout makes scores easier to read. It does not make a bad question good. If the domain question uses jargon your candidates have never met, every candidate will lose the same points on it, and the rubric will faithfully record a problem with the question rather than with the people. And a points total is still a summary of one 15-20 minute conversation, not a measure of potential.

Read one points breakdown today

Sign in to HireQwik, open a completed interview on a points-based job, and read its breakdown part by part. For each core question, find the answer in the transcript and decide which level you would have given. If your levels match the evaluator’s on most questions, your rubric is working. If they keep differing on one question, rewrite that question’s level descriptions first. If you would like a second opinion on a rubric before launch, talk to us.