Skip to content

How the Sistum score works

Method · version 9Updated 17 November 2026

One number for the health of a design system. The method is public and versioned, so a score means the same thing wherever you see it.

The idea

From metric to score in four steps

1

Measure

Sistum reads Figma, your code, your docs and your surveys and computes the metrics.

2

Rate

Each metric is rated against its target: Healthy, Needs attention or At risk.

3

Average

A rating is worth 100, 60 or 20 points. A vital is the average of its metrics.

4

Combine

The Sistum score is the average of the vitals.

A worked example

The arithmetic, with nothing hidden

Step 1 to 3 · The efficiency vital

Detach rateAt risk20
Time to merge a contributionNeeds attention60
ReuseHealthy100
Release healthHealthy100
Average of 4 metrics70

Step 4 · The seven vitals

Code
90
Quality
80
Efficiency
70
Docs
100
People
80
AI
60
Design
100

(90 + 80 + 70 + 100 + 80 + 60 + 100) / 7 = 83

Based on 16 of 16 metrics

Illustrative numbers, to show the arithmetic.

The seven vitals

What each vital looks at

Code

Is the system being used?

Share of product code that uses the system, coverage, outdated usage, reuse.

Design

Do designers use the library?

Share of Figma files that use the library, detached instances, and how well tokens and components match the code, when your plan provides the data.

AI

Can AI tools use it?

Prop descriptions, examples and deprecation notes that make a component clear to an agent.

Quality

Is it used well?

Style overrides, design and code consistency, token usage (a token reference only counts when it is defined) and sync, accessibility checks.

Efficiency

Does it save time?

Detach rate, contribution speed, release health, time saved reported by your teams.

Docs

Can people find out how?

Documentation of components and your documentation site, ownership, lifecycle.

People

Do teams like it?

Anonymous survey results: satisfaction, NPS, hours saved. Hidden under 3 answers.

Each metric has its own "How is this calculated?" page in the app, with its target and its source.

What Sistum reads

Four sources, and nothing else

Sources Sistum reads
FigmaLibrary files, styles, variables. Library analytics where your plan provides them.
GitHub, GitLab and BitbucketYour design system repository and your product repositories: usage, props, overrides, releases, issues.
Zeroheight and NotionPages and coverage of your documentation, in Zeroheight or Notion. Optional.
SurveysAnonymous answers collected with Sistum. Results hidden under 3 responses.

What Sistum sends

Two ways out, and only when you ask

Sistum measures by reading. The only things it writes are the ones you set up yourself.

Tickets

When you choose Create ticket on an item in Do this next, Sistum opens one issue in GitHub, GitLab, Bitbucket, Jira or Linear, with the evidence in the description and a link back. Nothing is created without that click.

Alerts and digests

Sistum can post the digest and new alerts to a Slack or Teams channel through a webhook you create, or to a signed webhook of your own. Each message is sent once and nothing is read back from your chat.

Both are optional and off until you connect them.

What you add

Two inputs that come from people, not code

Some things no repository can tell you. Sistum asks for them, and labels them as answers and estimates.

Team surveys

Anonymous surveys from a template, shared by a public link. Totals show once 3 people have answered, and comparisons between roles or rounds need 5 per group. Satisfaction, NPS and hours saved feed the dashboard.

ROI estimate

People using the design system, times hours saved per person, times 12 months, times hourly cost, gives the value per year. ROI is value minus cost, divided by cost. Every figure is typed by you or prefilled from your surveys, so it is an estimate, not something Sistum measures.

Coverage

A partial reading says it is partial

Next to every score Sistum shows how many metrics it is based on. A reading made from a public repository alone is labelled partial and is never ranked against full readings.

Full reading

Based on all the metrics your connections can provide.

Partial reading

Based on 5 of 16 metrics. Never compared with a full reading.

Honest limits

What the score does not tell you

It measures signals, not design quality.

A system can score well and still be hard to use. Use the score to start a conversation, not to end one.

Targets can be gamed.

Wrapping everything in a component raises adoption without raising quality. Read the vitals together, and read the sentence, not only the number.

Missing data is not zero.

When a plan or a connection does not provide a metric, or a source cannot be read, Sistum leaves it out, says why, and shows a gap in the trend instead of a drop to zero.

Versions

Every change is written down

If the method changes, the version number changes, and every score keeps the version it was computed with. Trend charts mark the point where the method changed, because values before and after may not be comparable. Disagree with a choice? Tell us at [email protected].

Method versions
VersionDateWhat changed
v.917 November 2026The AI-readiness score card is removed. It averaged the other AI cards and also counted the Figma description coverage from the Docs tab, so the AI group counted the same signals twice. The AI score is now the average of the AI cards that have data, each rated Healthy, Needs attention or At risk like every other group, and Figma descriptions count only in Docs. The AI score and its trend can move because of this change, not because of your design system.
v.816 November 2026AI-readiness grows. Two new signals: importable components (the share of components exported from the package entry point or exports map) and deprecations with a replacement (the share of @deprecated notices that say what to use instead). AI context and tooling counts llms.txt and agent instructions as half when they were last changed more than six months before the newest change to the components (GitHub and GitLab only). Documented props now also counts a stated default for every optional prop, so that score can fall for the same code. The AI-readiness score averages the new signals too.
v.712 November 2026Visual match is removed. Comparing a picture of the Figma component with a render of its Storybook story gave too many false differences (stories that wrap a component in a trigger button, changed copy, sample data), so the Quality score no longer includes it. Older readings that have a value for it are scored without it. Design-code consistency still measures whether each Figma component has a coded counterpart.
v.69 November 2026One rule to match component names between Figma, code and documentation: the exact name, then the same name ignoring case, emoji, hyphens, spaces and punctuation (letters of any alphabet are kept), then the last or first part of a path name such as "Forms/Text field". Two components that have the same name under this rule are no longer matched by order: neither is matched, and it is reported. Names that matched before still match.
v.59 November 2026The default variant is read from the default values of the variant properties, the values Figma puts on a new instance, instead of the first variant of the set. For most components these are the same. Where a designer set other defaults, the picture used by Visual match changes.
v.49 November 2026Visual match compares the default variant. The Figma picture is now the variant a designer gets when inserting the component, instead of the first variant in the order Figma lists them, which could be a Hover or Expanded variant. Scores of components whose picture changes can differ from this version on.
v.38 November 2026Token references must be defined. A token reference in a product repository now counts as valid only when the design system, or the repository itself, declares it. Otherwise it is an undefined token and counts against the share. The share is valid references divided by valid, hardcoded and undefined ones. Before, every reference counted as valid, even a misspelled one.
v.28 November 2026A failed fetch is no longer a measurement. When a source cannot be read (network error, timeout, access refused, rate limit), every metric that depends on it is stored as missing, with the reason, and never as 0 or a default. The score averages only the metrics that have data. Earlier readings that carry a source failure are left out of trend lines.
v.18 November 2026Versioning starts. Every reading now records the method it was computed with, and trend charts mark where the method changes. Readings made before this have no version and show as "Before versioning". For example, the AI tokens metric moved from 100 to 38.5 on 5 November because its formula changed, before versions existed.
Before versioning4 November 2026Initial public description of the method, before versioning. Seven vitals, ratings worth 100, 60 and 20 points, averaged.

Get your own Sistum score

Join the waitlist, or try an instant reading on a public repository.

Free plan. Early access for design system teams. No spam. We use your email only to tell you about the beta. See our privacy page.