SCI: a skill score built from evidence.
SCI is Lemma's skill score: one number from 0 to 100 for each skill. Four capped components add points, integrity checks can take them away, and every point is stored with the evidence that produced it.
TypeScript
Skill record
- Practice22 / 25
10 sessions (avg 4.5) · 3 taught (1.5x) · 2 AI drills
- Proof26 / 40
2 peer confirms (+8) · skill assessment (+12) · 2 external proofs (+6)
- Reliability20 / 20
10 clean sessions (+15.0) · streak bonus (+5)
- Freshness10 / 15
active this week
- Integrity0
No deductions
Five components, each with a cap.
Practice, proof, reliability and freshness can add up to 100 points between them. Integrity can only subtract. The total is held between 0 and 100.
Practice
25 points max
How it's earned
Completed sessions on the skill, weighted by the rating your partner gives. Sessions where you teach count one and a half times. Passed AI practice drills add up to 5 points.
What lowers it
Low-rated sessions add little. Sessions alone stop adding points at 20, so volume cannot buy the whole score.
Proof
40 points max
How it's earned
Results confirmed outside your own account: peer confirmations from different people, your best skill or certification assessment and connected proofs from tools such as GitHub, Figma or Vercel.
What lowers it
Each failed assessment subtracts 2 points, up to 6. Only your best assessment counts, so retaking an easy test does not stack.
Reliability
20 points max
How it's earned
Completed sessions add points up to a cap, and a clean record of ten or more sessions with no no-shows or late cancellations adds a bonus. Your own evidence shows the exact points each session added.
What lowers it
In the last 30 days, each no-show subtracts 6 points and each late cancellation 3. Reliability never drops below zero.
Freshness
15 points max
How it's earned
How recently new evidence arrived for this skill. Full points need evidence this week and at least five completed sessions.
What lowers it
Points step down as evidence ages and reach zero after 90 days without anything new.
Integrity
−30 points max deduction
How it's earned
Starts at zero and stays there for ordinary activity. It never adds points.
What lowers it
Pattern checks deduct up to 30 points when a skill's evidence looks farmed, such as most sessions with one partner or confirmations traded in a closed loop.
Rebuilt from the record, every time.
Nobody edits a score by hand. It is recalculated from the skill's evidence log whenever something new arrives.
- New
Peer confirmation · Ingrid N.
#1042 · 24 Sep 14:02
Skill assessment · level 9.1
#1038 · 21 Sep 09:15
Append-only · 17 events
SCI
- practice22/2510 sessions (avg 4.5) · 3 taught (1.5x) · 2 AI drills
- proof26/402 peer confirms (+8) · skill assessment (+12) · 2 external proofs (+6)
- reliability20/2010 clean sessions (+15.0) · streak bonus (+5)
- freshness10/15active this week
- integrity0no deductions
Saved 14:02:10 · nightly replay matches
- 01
Evidence is appended
A completed session, an assessment result, a peer confirmation or a connected proof is added to the skill's evidence log. Earlier entries stay as they were.
- 02
A recompute is queued
The new event queues a recompute job for that skill. A background worker picks it up and retries if a step fails.
- 03
Components are rebuilt
All five components are calculated from the full log under a versioned ruleset. The ruleset version is saved with the result.
- 04
Drivers are stored
The score is saved with each component's value and the drivers behind it, such as "2 peer confirms (+8)". A scheduled check replays scores to catch drift.
Freshness keeps a score current.
A skill you used last week says more than one you last used two years ago. Freshness rewards recent evidence and lets old scores fade, so the number reflects what you can do now.
After 90 days without new evidence, the skill's status also changes to stale. One new session or assessment brings it back.
Freshness points by age of the newest evidence
- This week, with 5+ sessions15
- This week10
- Within 30 days7
- Within 90 days3
- Older than 90 days0
Integrity checks look for farmed evidence.
A score is only useful if it is hard to inflate. The checks look at patterns across a skill's evidence, not at single events, and they only ever subtract.
- Dmitri R.9 · 64%
- Petra M.2 · 14%
- Selin O.2 · 14%
- Emil E.1 · 7%
Repeated pair: 9 of 14 sessions with Dmitri R. · integrity deduction applied
Recomputed
Repeated pairs
If more than half of your completed sessions on a skill are with the same partner, the skill takes a deduction of up to 15 points.
Closed loops
If most of the people who confirmed your skill are people you confirmed back, the skill takes a deduction of up to 15 points.
One person, one confirmation
Peer confirmations count once per person. Ten confirmations from the same friend are worth the same as one.
Every point leads back to a record.
- +4.0
Ingrid N. confirmed TypeScript
60 min · discriminated unions · 24 Sep
- +4.0
Yusuf A. confirmed TypeScript
60 min · generics · 18 Sep
- skill assessment (+12)
- 2 external proofs (+6)
Correction: one session voided after review
Appended 26 Sep · earlier entries unchanged
Skills Passport document
TypeScript · SCI 78 · issued 2 Oct
Each driver on a score refers to entries in the skill's evidence log: the sessions, assessments, confirmations and connected proofs that produced it. The log is append-only. Corrections, revocations and disputes are added as new events instead of erasing old ones.
When a dispute voids a session, the correction is recorded first, then the score is recalculated without that session. Your Skills Passport shows the result, and the Passport document is signed so a viewer can check it came from Lemma.
Questions about SCI.
Can my SCI go down?
Why is proof worth the most?
Do more assessments mean more points?
Is SCI one score for all my skills?
What does a score not tell you?
Give your skills somewhere to live.
Bring your evidence together and start building a Skills Passport that can grow with you.