# How the SCI score works

> SCI is a 0 to 100 score per skill, built from practice, proof, reliability and freshness, with an integrity check. A worked example, point by point.

Source: https://getlemma.io/blog/how-sci-works

SCI is a 0 to 100 score per skill, built from practice, proof, reliability and freshness, with an integrity check. A worked example, point by point.

Lemma team

21 September 2026 · 5 min read

![How the SCI score works](https://cdn.sanity.io/images/islu4peo/production/1afe0bfd59db0d9e1346c552ba81213407372799-2400x1350.jpg?w=2400&fm=webp&q=80&fit=max)

SCI is a score from 0 to 100 that Lemma computes for each skill on your [Skills Passport](https://getlemma.io/product/passport). It summarises how much evidence backs that skill, how hard that evidence was to produce, and how recent it is. You cannot type a number in, and every point traces back to the evidence that produced it.

This post explains each part of the score, then walks through one real calculation point by point.

## What are the parts of an SCI score?

Four components add points, each with its own cap. A fifth, integrity, can only subtract.

| Component | Cap | What feeds it |
| --- | --- | --- |
| Practice | 25 | Completed live sessions, weighted by rating; short AI tutor drills |
| Proof | 40 | Adaptive assessments, confirmations from distinct people, linked work, organisation endorsements |
| Reliability | 20 | Sessions you turned up to; recent no-shows and late cancellations subtract |
| Freshness | 15 | How recently you last produced evidence |
| Integrity | 0 to -30 | Patterns that look like gaming; zero for honest use |

The total is held between 0 and 100. The caps are the most important design choice in the score: they stop any one kind of evidence from carrying a skill on its own.

## How does practice work?

Practice comes from [live sessions](https://getlemma.io/product/sessions), each worth points according to the rating the other person gave it. A five-star session is worth 5 points, four stars 4, three stars 2.5, two stars 1 and one star nothing. An unrated session counts as 3.

Sessions where you teach count one and a half times, because explaining a skill well takes a deeper grasp of it. Sessions can add up to 20 points. Short drills with the [AI tutor](https://getlemma.io/product/ai-tutor) add 1 point each, up to 5, so practice with software can never stand in for practice with people.

## How does proof work?

Proof is the largest component, because it is the hardest evidence to produce. Four sources feed it, each capped:

-   **Adaptive assessments.** Your best skill assessment adds 8 points for a pass (6.5 out of 10 or better), 10 for 7.5 or better and 12 for 9 or better. Each failed attempt subtracts 2, up to 6. See [adaptive assessments](https://getlemma.io/product/assessments) for how the difficulty moves.
-   **Confirmations from people.** Each confirmation from a distinct person adds 4, up to 12. Three people can confirm a skill; a fourth adds nothing.
-   **Linked work.** A GitHub repository, Figma file or Vercel deployment, connected through your own account, adds 3 each, up to 6.
-   **Organisation endorsements.** An endorsement from an organisation adds 5, up to 10.

Proof is capped at 40 overall.

## How do reliability and freshness work?

Reliability measures whether you show up. Each completed session adds 1.5 points, up to 15. Ten or more sessions with no no-show or late cancellation ever adds a 5-point bonus. A no-show in the last 30 days subtracts 6, and a late cancellation subtracts 3. Penalties older than 30 days stop subtracting, but they still rule out the bonus.

Freshness measures how recent your evidence is:

| Latest evidence | Freshness |
| --- | --- |
| Within 7 days, with 5 or more sessions | 15 |
| Within 7 days | 10 |
| Within 30 days | 7 |
| Within 90 days | 3 |
| Older than 90 days | 0 |

If you stop practising a skill, freshness falls and the score drifts down. When new evidence arrives, it recovers.

## What does a worked example look like?

Here is a sample Python record, the one shown on this post's cover. It is illustrative, not a real person, but every number follows the current rules.

**The evidence.** Six live sessions in the last two months: four as the learner, each rated four stars, and two as the teacher, each rated five stars. One adaptive assessment passed with 8.1 out of 10. Two GitHub repositories linked through the member's own account. The latest session was three days ago. No no-shows or late cancellations.

**Practice: 20 of 25.** The four learner sessions are worth 4 each, so 16. The two teaching sessions are worth 5 × 1.5 = 7.5 each, so 15. That makes 31, but sessions are capped at 20. There are no AI drills, so practice is 20.

**Proof: 16 of 40.** The assessment scored 8.1, which clears 7.5, so it adds 10. The two linked repositories add 3 each, so 6. There are no confirmations or endorsements yet. Proof is 16.

**Reliability: 9 of 20.** Six completed sessions at 1.5 each make 9. There are fewer than ten sessions, so no bonus yet.

**Freshness: 15 of 15.** The latest evidence is three days old and there are at least five sessions.

**Integrity: 0.** Nothing unusual. The sessions were with several different people.

**Total: 20 + 16 + 9 + 15 = 60.**

The breakdown also shows where the next points would come from. Practice is full, so more sessions alone add little there. Three confirmations from people who have worked with this person would add 12 to proof. Four more clean sessions would lift reliability to 15 and unlock the 5-point bonus. A stronger assessment, 9 or better, would add 2 more.

## What does the integrity check look for?

It looks for two patterns built to inflate a score rather than to learn. One is pair repetition: more than half of a skill's sessions with the same partner. The other is a confirmation ring: most of your confirmations coming from people you confirmed back. Each pattern can subtract up to 15 points, 30 in total, and the reason appears in the breakdown. For honest use the check is zero and has no effect.

## Why are scores versioned and traceable?

Every point comes from a specific item: a session on a date, an assessment result, a linked repository. Evidence is append-only. When something is corrected or disputed, the change is added to the record rather than replacing what was there, and the score is recomputed.

The rules that turn evidence into a score are versioned. The example above uses the current ruleset, version 3. Each Passport records which version produced its scores, so a score can always be read against the rules in force when it was computed. The [methodology](https://getlemma.io/methodology) documents them, and the [SCI product page](https://getlemma.io/product/sci) shows how the breakdown appears to you.

## What is SCI not?

SCI is not a ranking against other people, and it is not a pass or fail. It is a summary of the evidence for one skill, with that evidence visible underneath. It is also not a measure of talent. A new skill with little evidence has a low score because there is little to show yet, not because the person is weak at it. For the terms used here, see the [glossary](https://getlemma.io/glossary).

## Frequently asked questions

Can I raise my SCI score by doing one thing many times?

Only up to a point. Each component has its own cap, so repeating one kind of evidence stops adding to the score once that component is full. In the worked example, practice is already at its session cap of 20.

Does my subscription affect my SCI score?

No. The score is computed from evidence only. What you pay for does not change how evidence is counted, and every plan can reach 100.

Why did my score go down when I had not done anything wrong?

Most likely freshness. If a skill has no recent evidence, the freshness component falls in steps at 7, 30 and 90 days, so the score reflects current activity rather than past activity.

How much can AI tutor practice add to SCI?

At most 5 points out of 100. Each passed drill adds 1 point to practice, up to 5, so practice with the AI tutor supports a score but cannot carry it.

![](https://cdn.sanity.io/images/islu4peo/production/2b7af0b20be2229e3514f378699c1509c6891e0c-1536x1024.png?rect=0,128,1536,768&w=1600&h=800&fm=webp&q=70&fit=crop)

## See the rules applied to your own skills.

Every SCI point on a Skills Passport links back to the evidence that produced it, under a versioned ruleset anyone can read. Join the early-access list to build yours.

[See how SCI is broken down](https://getlemma.io/product/sci)

## Read more

-   [Methodology · 5 min read

    ### What is skill verification?](https://getlemma.io/blog/what-is-skill-verification)
-   [Verification · 6 min read

    ### LinkedIn Skill Assessments: Still Available in 2026?](https://getlemma.io/blog/linkedin-skill-assessments-no-longer-available)
-   [Learning · 7 min read

    ### Skill Swap: How It Works and Where to Do It](https://getlemma.io/blog/skill-swap-app)