# Technical skills assessment template and rubric

> A copy-ready technical skills assessment template: a four-level scoring rubric with behavioural anchors, an evidence weighting table and a scorecard.

Source: https://getlemma.io/blog/technical-skills-assessment-template

A copy-ready technical skills assessment template: a four-level scoring rubric with behavioural anchors, an evidence weighting table and a scorecard.

Lemma team

30 September 2026 · 6 min read

![Technical skills assessment template and rubric](https://cdn.sanity.io/images/islu4peo/production/492b950e71ecfc2fb82ee623e09d28e3026bf967-2400x1350.jpg?w=2400&fm=webp&q=80&fit=max)

A technical skills assessment template has four parts: the tasks the role needs, a scoring rubric with a description of each level, a rule for weighting different kinds of evidence, and a scorecard each assessor fills in independently. This post gives you all four, ready to copy, built on the same principles Lemma uses to score skills.

Use it for a hiring screen, a promotion conversation or a team skills review. Adapt the criteria to the role; keep the structure.

## What makes a technical assessment fair and useful?

Four properties, in this order. If one is missing, the others cannot make up for it.

1.  **It measures the job.** Every criterion traces back to a task the role actually needs. "Knows React" is not a task. "Builds and ships a stateful interface with clear component boundaries" is.
2.  **Levels are described, not just numbered.** A score of 3 means something only if the rubric says what a 3 looks like. Behavioural anchors make two assessors agree.
3.  **It uses more than one source.** A single sitting is noisy. Combine a work sample with a structured conversation, or with evidence the candidate already has.
4.  **Scores are recorded before discussion.** Each assessor scores alone, with notes on what they saw. Discussion comes after, so the loudest voice does not set the result.

These are also the principles behind Lemma's scoring: evidence over claims, every score traceable to what produced it, and limits so that no single source can carry the result. The [methodology](https://getlemma.io/methodology#how-evidence-is-weighed) explains how Lemma applies them.

## Step 1: How do you define what to assess?

Write the role as three to five tasks. Then, for each task, pick the criteria from the rubric below that tell you whether someone can do it. Drop criteria that do not map to any task.

For example, a backend role might have: design and ship an API endpoint end to end; diagnose a production issue from logs; review a colleague's pull request. That maps to correctness, code quality, debugging, design and communication. It probably does not need a separate algorithms criterion.

## Step 2: What does the scoring rubric look like?

Here is a four-level rubric for software engineering work. Score each criterion from 1 to 4 based only on what you observed. If you saw no evidence for a criterion, write "not observed" rather than guessing.

| Criterion | 1\. Not yet | 2\. Working | 3\. Strong | 4\. Exceptional |
| --- | --- | --- | --- | --- |
| Problem understanding | Starts coding before understanding the task; misses stated requirements | Understands the main requirement; asks few questions | Clarifies requirements and edge cases before starting | Reframes the problem, surfaces hidden constraints and agrees scope |
| Correctness | Solution does not run or fails core cases | Core cases work; edge cases missed | Handles edge cases; verifies with tests or checks | Proves correctness systematically; anticipates failure modes |
| Code quality | Hard to read; unclear names; duplicated logic | Readable but uneven structure | Clear structure and names; easy to change | Code a teammate could extend without asking questions |
| Debugging | Changes things at random | Finds the bug with guidance | Forms and tests hypotheses; isolates the cause | Finds root cause quickly and explains how to prevent it recurring |
| Design and trade-offs | No visible reasoning about choices | Makes reasonable choices without explaining them | Explains trade-offs between at least two options | Weighs trade-offs against the real context: scale, team, cost |
| Communication | Hard to follow; silent when stuck | Explains when asked | Narrates reasoning; responds well to hints | Adjusts explanation to the listener; teaches as they go |
| Tool use, including AI | Accepts tool output without checking | Uses tools; checks output occasionally | Verifies and adapts tool output; knows its limits | Uses tools to go faster without losing understanding |

Copy the rows you need. If the role is not engineering, keep the four level names and rewrite the criteria; the anchors should always describe behaviour you can observe.

## Step 3: How should you weight different kinds of evidence?

Not all evidence is equally hard to produce, so not all of it should count equally. A hiring decision weighs evidence by two things: how close it is to the job, and how hard it is to fake or prepare for. The table applies that to the sources most hiring processes use.

| Evidence source | Example | Suggested weight | Why |
| --- | --- | --- | --- |
| Work sample or assessment | Take-home task, live exercise, adaptive assessment result | Highest (about 40%) | Hardest to fake; closest to the job |
| Structured technical interview | Same questions and rubric for every candidate | High (about 25%) | Shows reasoning and communication |
| Prior evidence of practice | Repositories, shipped projects, rated sessions | Medium (about 20%) | Shows sustained work, not one sitting |
| Recency | When the candidate last used the skill | Lower (about 15%) | A skill used last month says more than one used five years ago |

Why this order? The work sample or assessment sits highest because it is the stage you designed for this role and observed directly. The structured interview comes next because, run well, it is one of the strongest predictors of performance, but it captures less hands-on work. Prior practice matters because one sitting is noisy; it shows the candidate did the work more than once, in other settings. Recency is lowest because it adjusts the others rather than standing alone.

These weights differ from a general skill score on purpose. A hiring decision is one point in time for one role, so the stages you run yourself deserve the most weight. The percentages are a starting point, not a standard. What matters is deciding them before you see any candidate, writing them down, and applying them the same way to everyone. Self-description and endorsements belong in context, not in the score.

## Step 4: What goes on the scorecard?

One scorecard per assessor, per candidate, filled in before any discussion. Copy this table into your document or applicant tracking system.

| Field | What to write |
| --- | --- |
| Candidate and role | Name, role, date |
| Assessor | Name, which part of the process they ran |
| Tasks covered | Which of the role's tasks this stage tested |
| Criterion scores | 1–4 for each criterion, or "not observed" |
| Evidence notes | One or two sentences per score: what you saw that justifies it |
| AI and tool rules | What was allowed, and whether the candidate followed it |
| Concerns | Anything to probe in a later stage |
| Recommendation | Proceed, do not proceed, or need more evidence, with the main reason |

The evidence notes are the important column. A score without a note cannot be checked, challenged or calibrated later.

## How do you turn scores into a decision?

Set the rule before you start. A simple, defensible version:

-   **Must-have criteria** need at least a 3. For a senior backend role, that might be correctness, debugging and design.
-   **Everything else** needs at least a 2, with no 1s.
-   **"Not observed"** on a must-have means more evidence, not a fail. Add a stage that tests it.

Adaptive assessments do something similar automatically. Lemma's assessments run on a 1 to 10 difficulty scale, ask between 15 and 25 questions, and weight harder questions more. A weighted score of 6.5 out of 10 or better passes, and the score sets a level from beginner to expert. See [adaptive assessments](https://getlemma.io/product/assessments) for how the difficulty moves.

## How do you keep assessors consistent?

Calibrate before and after.

1.  **Before.** Have every assessor score the same sample answer with the rubric. Discuss differences until the anchors mean the same thing to everyone.
2.  **During.** Score alone. Write evidence notes as you go.
3.  **After.** Review a few scorecards each month. If one assessor's scores are always higher, recalibrate rather than adjusting their results.

Integrity matters too. Decide in advance what candidates may use, including AI tools, and score how they used them. For online assessments, log signals such as tab switches and pastes and have a person review flags, rather than cutting scores automatically. That is how Lemma handles [assessment integrity](https://getlemma.io/methodology#trust-and-integrity).

## Where does existing evidence fit?

Before you schedule an assessment, look at what the candidate already has. A [Skills Passport](https://getlemma.io/product/passport) shows each skill's SCI with the sessions, assessments and linked work behind it, so you can skip what is already well evidenced and spend your interview time on the gaps.

If you are choosing a platform to run assessments, [HackerRank alternatives for evidence-based hiring](https://getlemma.io/blog/hackerrank-alternatives-evidence-based-hiring) compares the main options. For the interview stage itself, including task design and follow-up prompts, read [how to assess technical skills in an interview](https://getlemma.io/blog/assess-technical-skills-in-interview). For skill-specific criteria, the skill pages for [React](https://getlemma.io/skills/react), [Python](https://getlemma.io/skills/python) and [SQL](https://getlemma.io/skills/sql) describe what competence looks like.

## Frequently asked questions

What should a technical skills assessment include?

The tasks the role needs, written down before you assess anyone; a rubric that describes what each level looks like for each criterion; more than one source of evidence; and a rule for how the scores combine into a decision.

How many levels should a technical scoring rubric have?

Four works well. An even number stops assessors defaulting to the middle, and four levels are enough to separate not yet, working, strong and exceptional without arguing over fine distinctions.

Should candidates see the rubric before the assessment?

Sharing the criteria, not the exact anchors, is usually fair. Candidates do better work when they know what is valued, such as tests, readable code or explaining trade-offs, and you learn more about their real ability.

How do I stop a technical assessment rewarding AI-generated answers?

Decide the rules first: either allow AI tools and score how well the candidate checks and explains the output, or disallow them and use formats that make that visible, such as a live walkthrough of their own code.

![](https://cdn.sanity.io/images/islu4peo/production/2a68eb30a708fe7c4f76ebc6040dab700ab31948-1672x941.png?rect=0,53,1672,836&w=1600&h=800&fm=webp&q=70&fit=crop)

## Hire on evidence you can check.

Candidates on Lemma bring a Skills Passport: each skill scored from sessions, assessments and linked work, every item traceable to its source. Join the early-access list for hiring teams.

[See the verification API](https://getlemma.io/developers)

## Read more

-   [Hiring · 6 min read

    ### How to assess technical skills in an interview](https://getlemma.io/blog/assess-technical-skills-in-interview)
-   [Hiring · 8 min read

    ### Skills-based hiring: a practical guide](https://getlemma.io/blog/skills-based-hiring-guide)
-   [Hiring · 6 min read

    ### HackerRank alternatives for evidence-based hiring](https://getlemma.io/blog/hackerrank-alternatives-evidence-based-hiring)