Hosted cards + Studio
Eight public SVG routes with synthetic or live-public previews, plus an HTML dashboard with project actions.
Open Studio00 // One open-source instrument for the whole trail
Your GitHub work,
mapped clearly.
Explore a clearly labelled synthetic example, then generate your own portfolio from public GitHub evidence. CommitAtlas ships hosted SVG cards and scenes, ten static card types, an HTML project dashboard, and a credential-free GitHub Action.
This landing page uses deterministic octocat data. It does not represent your account until you run a Studio preview or generate a static snapshot.
01 // Product boundary · August 2026
What you can use today.
Public-only by design
Unknown is never green
Static portfolio
Ten card types, responsive Atlas companions, a project catalog, and hash manifests from one public snapshot.
View generatorDistribution
The pinned GitHub Action and repository CLI work now. The packages are not published to npm, and the Action is not listed in Marketplace.
View ActionDesign lab
The fascia, six-state rack, and evidence drawer explain product rules. They are not selectable exports or a hosted monitoring service.
See shipped outputs02 // From evidence to embed
One source. Several useful exits.
Preview, generate, embed
No publishing service implied
Start with a preview, then choose the distribution surface that fits the place your work has to live. Each path keeps the source and its limits visible.
- 01
Preview the evidence
Use synthetic data to explore safely, or enter a GitHub handle for a public-only preview.
Open Studio - 02
Take the hosted exit
Use a hosted SVG URL or copy the Studio's HTTPS Markdown after a validated preview.
Open Atlas SVG example - 03
Generate the static exit
The static CLI or pinned Action fetches once, stages the selected ten cards, and writes a byte/SHA-256 manifest.
Static CLI Pinned Action - 04
Embed the committed output
Commit the generated artifacts to a profile README or site. The
See an example profilemanifest.jsonrecords what was actually written.
Hosted: 8 SVG routes and Studio previews. Static: 10 card types, including Cadence and Releases, which are static-only.
The design lab below is optional: it explains evidence, freshness, and state vocabulary on this page; it is not an export, monitoring service, or extra card.
02 // Specimen tray · 8 card routes
See the whole toolkit.
Every preview is a real SVG response
Safe, deterministic octocat data · No GitHub calls are made to render this page
CARD 01 // ATLAS COMPACT480×570 FIXED
The full atlas in a 480×570 portrait plate, for a README column that has no room to run wide.
MOBILE READMEOpen SVG
CARD 02 // PROFILERESPONSIVE
Public repository, follower, contribution, and star signals at a glance.
24 REPOS · 312 FOLLOWERS · 487 STARSOpen SVG
CARD 03 // STREAKRESPONSIVE
Current and longest observed streaks with an honest history boundary.
WINDOW-BOUNDED · NO GUILTOpen SVG
CARD 04 // BREAKDOWNRESPONSIVE
Exact categorized counts when available; otherwise clearly labelled public-profile percentages.
EXACT COUNTS OR LABELLED PROFILE %Open SVG
CARD 05 // RHYTHMRESPONSIVE
Transparent personal consistency based on density and streak — not a GitHub rank.
72/100 · NOT A RANKOpen SVG
CARD 06 // ACTIVITYRESPONSIVE
A compact calendar view of public contribution density over time.
EXACT DATE WINDOWOpen SVG
CARD 07 // LANGUAGESRESPONSIVE
Public repository-language distribution, presented without proficiency claims.
REPOSITORY SHARE · NOT PROFICIENCYOpen SVG
CARD 08 // PROJECTSRESPONSIVE
Declared lifecycle and project signals for Hello-World and Spoon-Knife.
SIX HEALTH STATES · UNKNOWN ≠ FINEOpen SVG
03 // Atlas × 4 card themes — paper ships for light READMEs
Hover: the plate lifts 3px · crop marks brighten · transform-only, 180ms easePlates, not cards: the page frames; the SVG is the untouched specimen
04 // Optional design labSee how CommitAtlas decides what to trust.Landing-page explainers only — these instruments are not separate cards or monitoring services.Open lab
01 // Momentum plotter · 28dM1
vs prior 28d
02 // Rhythm · 0–100M2
Evenness · not a rank
03 // Density survey · 365dM3
active days · /day
04 // Portfolio reticle · CIM4
Unknown is never green
Instrument scaleHinge #FFD166 · lives in the cards
STN A // Contribution signalsDistribution · R0
The trail, counted honestly.
Contributions
Active days
Cmt / PR / Rev / Iss
0/2 CI PASSING · 0 ATTENTION · 2 UNCONFIGURED — shown dark, never green
STN B // Distribution surfacesFour ways out · R12
The same evidence, wherever it has to live.
8SVG routes
4Card themes
0Credentials required
Observed → derived → hypothesis · every number carries its caveat
✓ Public by default ✓ Honest freshness ✓ Live or staticReduced motion: complete static page · nothing enters from invisible
04 // Portfolio signals · CI state rack
A dark gauge is a finding, not a failure of the page.
Live synthetic reading: CI passing · 0 attention · 2 unknown
72h freshness window · calculateCiState()
Design-lab explainer — not a separate monitoring service. Every state gets a distinct lamp shape, a distinct trace, and a printed word — colour is never the only encoding. The three “we do not know” states are deliberately the most visually interesting bays in the rack.
UNAVAILABLE—
NO SIGNAL
PROVIDER RETURNED NOTHING.
NOT ZERO — UNOBSERVED.
LAMP UNLIT · STEADY · NO PULSE
UNCONFIGURED×2
NO PROBE FITTED.
NO NAMED WORKFLOW TO WATCH.
EMPTY SOCKET · DASHED TRIM
STALE—
LAST OBSERVED >72H AGO.
WAS PASSING · NO LONGER CLAIMED.
NEEDLE FROZEN · DOTTED TRACE
PASSING—
NAMED WORKFLOW GREEN,
OBSERVED INSIDE 72H.
LAMP LIT · SOLID TRACE · ✓
FAILING—
RED RUN, FRESH EVIDENCE.
A FACT, NOT A SHAME.
DIAMOND LAMP · SPIKED TRACE
PENDING—
RUN IN FLIGHT.
ACQUIRING — THE ONLY STATE
ALLOWED TO PULSE. HALF LAMP
M7 // Acquisition sequence — how a gauge fails honestlyM7
On load, an unavailable gauge does not snap to an error badge — the needle hunts: it rises toward a reading, stutters twice, falls back to the −90° rest stop, and the lamp never lights. The plate reads NO SIGNAL from frame zero, so the motion adds theatre, never information.
5.5s cycle · cubic-bezier(.4,0,.3,1) · transform-only · reduced motion: needle rests at −90°, plate unchanged
Six shapes: plate · socket · frozen clock · disc · diamond · half-disc — legible in greyscaleAn unknown, missing, or stale signal is never displayed as healthy
05 // Evidence layer · observed → derived → hypothesis
Facts become rates. Rates become guesses.
10 readings explained on this page
Every dotted number opens the drawer
Design-lab interaction — not an exported card. Every reading sits on one of three rungs, and every dotted number on screen answers “how do you know that”. CommitAtlas already refuses to paint an unknown signal green; this is the same rule one level up, applied to the difference between a number that was counted and a number that was inferred.
Order 1 · OBSERVED ●THIS IS IN THE DATA
contributions
Rule: contribution-total-observed
Order 2 · DERIVED ◍THIS FOLLOWS FROM THE DATA
active-day density
Rule: density-derived
Order 3 · HYPOTHESIS ○THIS IS A GUESS — LABELLED AS ONE
rhythm level
Rule: rhythm-band-hypothesis
Rung is encoded three ways: word · dot (solid / half / dashed) · border style.
Every reading sorts onto these three rungs — never a fourth.
Closed — every dotted number is a button: rhythm scorePress any dotted number: “how do you know that?”
What the drawer shows
One shared drawer rather than a popover per number: there is only ever one question being asked, and a single dialog is the only version of this that stays navigable by keyboard.
- BASIS
- Where the number came from, naming the window and the source.
- FORMULA
- Printed in mono for every derived and hypothesis reading. An observed value has none to show, and shows none.
- CAVEAT
- What the number cannot see. Never omitted, never softened.
M8 — drawer: slides up 12px and settles · 260ms · cubic-bezier(.2,.9,.3,1) · ESC or click outside to close · reduced motion: appears in place, fully drawn.
COVERAGE: 17 DIMENSIONS · 365-DAY PUBLIC WINDOW · "EVERY NUMBER SHOULD KNOW WHAT IT CANNOT SEE"Source: SYNTHETIC OCTOCAT
05 // Research bridge · C0 invented evidence
The baseline won. That is the finding.
An offline method trial compared Gaussian BOCPD with the retained Rolling median and MAD baseline. The evidence is an invented C0 weekly system series — not GitHub profile evidence and not production monitoring.
Detection · baseline / candidate75% / 75%Equal measured detection
False alerts / year · baseline / candidate2.97 / 4.2Lower is better; candidate is higher
Threshold viabilityBoth nonviableNeither selection cleared its gate
The candidate is rejected because both selections are nonviable and false alerts are higher. Rolling median and MAD stays as the retained fallback.
Open the public method trialRead this as research, not a product signal. It does not establish real-repository validity, support person-level inference, promote a model, or establish online PELT performance.
Pinned from Developer Lens producer commit 425708e03e7b · run wbc1_demo · no runtime fetch
06 // Configure without guesswork
One toolkit. Every signal.
Choose live public data or a clearly marked synthetic preview, declare your own project lifecycles, then copy only the cards you want as README Markdown.