Datum design systemA design system built like a cockpit
Datum is the system behind this site: one set of decisions that every page, theme and AI part is drawn from. I built it the way I design safety-critical products, so it stays calm when things are normal and gets exact when they are not.
05 Intelligence
04 Patterns
03 Components
02 Foundations
01 Principles
Datum, exploded: the first five layers, each built only from the one below.
My portfolio started as one page and grew into five kinds of page, two themes and more than fifty versions. Each page was built the week I needed it, and each one quietly made its own decisions. I spend my working days on interfaces where a control that moves is a safety issue, and my own site was breaking the rules I design by.
So I wrote it down as one system and called it Datum: in an engineering drawing, the datum is the reference every other measurement is taken from. Dark mode is the flight deck. Light mode is the drafting paper. Same parts, same names, two rooms.
The problem statement
How might one designer keep a growing, motion-rich product consistent, accessible and recognisably his own, without a team to police it?
Borrowed from
Airbus dark cockpitISA-101 control roomsGOV.UK principlesIBM Carbon motionApple semantic colourShopify Polaris voiceNASA standards manual, 1975
Each layer is built only from the ones before it. Access runs through all of them; governance and evolution keep it working over time.
Datum for AI// 8 of 30 parts
AI parts, up close
Datum has thirty parts that keep a person in charge of an AI. Here are eight, in the order a task meets them: set how far it may go, answer its question, check its reasoning, try the batch, stop it, accept its edits, read the receipt, and take over when it cannot go on.
Autonomy · this taskReschedule 40 inspectionsR3High risk
Principles are easy to write and easy to ignore. Each of these comes with a check I run before anything ships. If a design fails the test, the design changes, not the principle.
ClarityP1
Lights out when normal
Calm surfaces, generous space, nothing glows unless it needs you. Straight from the A cockpit philosophy where panel lights stay off while systems are normal, so any light that comes on means something..
Test //Blur a screenshot. You should still find one thing to do next.
ComplexityP2
Tame it, don't hide it
Experts need the dense view; newcomers need a way in. Use Showing the essentials first and revealing detail only when someone asks for it. instead of deleting detail.
Test //A newcomer finishes without the detail. An expert reaches any value in two steps.
ConsistencyP3
Same switch, same place
One action, one name, one position, on every screen. Airbus builds its cockpits so pilots move between aircraft types with minimal retraining.
Test //Two screens side by side. Shared controls match exactly.
TrustP4
Never vague about state
Status has words and a shape, not just a colour. Data says how fresh it is. The system never bluffs.
Test //Switch to grayscale. Nothing becomes ambiguous.
AccessibilityP5
Readable in a hurry, by everyone
12px floor, WCAG's minimum contrast ratio for normal text. 1:1 is no contrast at all; black on white is 21:1., keyboard first, in both themes. Built into the tokens, not added later.
Test //Zero automated issues, then a full keyboard walk.
Human controlP6
The system advises, people decide
Anything automated can be stopped, checked or undone. That goes double for AI.
Test //For every automatic action: where's the stop, and where's the undo?
PurposeP7
Everything earns its place
Motion answers an action. Decoration has a job or it goes.
Test //Remove it. If nothing gets worse, it stays out.
When rules collide//
Order is the tiebreaker
The list is ranked. Clarity beats cleverness. Control beats convenience. Access is never traded away.
System / 02// Foundations
The parts everything is built from
Eight foundations, each cut down to the decisions that matter, plus the grammar that ties them together.
02.1 // Design tokens
Named by job, not by colour
Every value has a name that says what it's for. A theme is just a different set of answers to the same names. Three tiers, each only allowed to use the one beneath it, so a raw value can exist in exactly one place.
--cyan act here. Night #00D4FF, paper #0057B8--amber look here, once per screen--void the surface--alarm failed, or can't be undone--ai made by AI, never status→
Component // where it's used Live
--btn-bg = --cyan, gradient on paper--btn-glow glow at night, ink shadow on paper--elev-paper-1 to 3 paper depth
Components reference semantic tokens, never hex values. That's what let a second theme slot in without rebuilding a single component.
Each page is a self-contained file, so it embeds a copy of tokens.css. The token audit compares every copy with the source and flags any that drift. One exception is declared: the site header is a sealed block shared by every page. It keeps its own raw values, and the audit counts them on a separate line.
02.2 // Colour
A budget, not a palette
Most of the screen is neutral. Cyan means you can act. Amber means look here. Red is kept for failures and for actions that can't be undone. Violet marks anything made by AI. Each colour is re-chosen per room so it keeps its job, and most of those choices are forced by contrast.
NightPaperSurfaceTextActLookAlarmAI
NeutralAct▲
Neutral carries the page. Cyan marks what you can act on. Amber is rationed to one attention point per screen.
How a colour moves between rooms
Action--cyan
Signal cyan on paper is only ✕ 1.6:1. Aero blue reaches ✓ 6.1:1, but fails at night at 2.9:1. Neither works in both rooms, so one name gets two values.
Attention--amber
Bright amber on paper is ✕ 2.5:1. Deepened to signal orange it holds ✓ 5.4:1.
Alarm--alarm
Coral at night 6.6:1, brick on paper 5.9:1. Amber and red sit close in brightness, so they always differ by shape too: triangle for caution, octagon for alarm.
AI--ai
Soft violet at night 8.8:1, deep violet on paper 6.3:1. It always travels with a spark and the word AI, so it never relies on colour alone.
Text--paper
Off-white at night, 17.1:1, not pure white's 20:1. Navy ink on paper, 15.8:1, not black.
Depth
Glow at night becomes a soft ink shadow on paper. Paper has no light source.
Every colour comes with contrast measured in both rooms before it's allowed in.
02.2 // Colour, second cue
Colour is never the only signal
Roughly 1 in 12 men has some form of colour-blindness, and every screen washes out in sunlight. So in Datum, colour adds meaning but never carries it alone. Both panels below show the same six engines, and one needs a check. Switch to the colour-blind view and see which panel still tells you.
A // Colour only
ENG-114ENG-208ENG-301ENG-402ENG-517ENG-623
B // Colour, shape and words
ENG-114ENG-208CheckENG-301ENG-402ENG-517ENG-623
In full colour, both panels point to ENG-208.
Every status gets a Showing the same information in more than one way, like colour plus shape, so it still gets through if one channel is missed.: a shape, a heavier border or a word. It's WCAG 2.2 criterion 1.4.1, and the same rule an ANSI/ISA-101, the standard for designing operator screens in industrial control rooms. It keeps screens neutral and saves colour for abnormal states. control room follows.
02.2 // Colour, human factors
Why the paper feels calmer
The light theme isn't the dark theme turned inside out. Its values were chosen for how eyes respond to them. Three numbers carry most of the story.
Light given off, compared with pure white
Pure white100%
Datum paper89%
Colour intensity
Datum paper4.5%
Signals100%
Text contrast, out of 21
Ink on white17.7
Ink on paper15.8
Calculated from the theme values: How bright a colour appears to the eye, from 0 for black to 1 for white. WCAG contrast ratios are built on it., HSV saturation and WCAG contrast.
Optics
A brighter page makes the pupil contract, which sharpens focus. People proofread better with Dark text on a light background. Light text on dark is negative polarity., most of all at small sizes.
Piepenbrock et al., Ergonomics 2013; Human Factors 2014
Glare
Full white dazzles over a long read. Dyslexia guidance advises cream or soft pastel instead. Datum's paper gives off about 11% less light.
British Dyslexia Association Style Guide
Arousal
Saturation is one of the strongest drivers of arousal. The page is under 5% saturated, so alertness is spent only on signals.
Valdez and Mehrabian 1994; Wilms and Oberfeld 2018
Halation
On dark screens bright letters can This is called halation: a glow or bleed around bright text on a dark background, more noticeable with astigmatism., especially with astigmatism. So night text is off-white, not pure white.
Trimble Modus dark mode guidance
But at night
In a dim room a bright page is the glare. Dark stays the brand and night theme, and the reader picks the room.
02.3 // Typography
Three voices, one job each
AABarlow Condensed // call-out
AaInter // explanation
AaShare Tech Mono // readout
Aa 12 floorAa 13Aa 15Aa 16Aa 17
Nothing under 12px, including labels inside charts. Body text never lighter than 400. Lines of 60 to 75 characters.
02.3 // Typography, hero on phones
On phones, the name gets bigger
The usual move is to shrink a headline until it fits. The home page does the opposite: two short words can carry a poster, so the name takes 24% of a phone's width against 9% on desktop. Frames are drawn to scale.
39024%
Aalok PurohitEnterprise UX
76815%
Aalok PurohitEnterprise UX
14409%
Aalok PurohitEnterprise UX
Phone clamp(4.2rem, 24vw, 7.5rem), desktop clamp(4rem, 9.5vw, 8.5rem). Every fluid size is a A CSS function that lets a size grow with the screen, but never below a minimum or above a maximum. with a floor and a ceiling.
Leading tightens from .88 to .84 as the type grows, so two huge lines read as one block.
Sentence headlines, like this page's, get a gentler rule and balanced wrapping.
02.4 // Spacing
Room to think
96
48
16
An Spacing values in multiples of 8px, so everything lines up on the same invisible grid.. Sections get the most room on desktop and step down on smaller screens.
Remove something before you shrink anything.
02.5 // Layout and breakpoints
One column first
Mobile first: a single column by default, splitting only when there's room. Content tops out at 1200px with fluid margins.
Phone // 1 col
Tablet // 2 col
Desktop // max 1200
320
390
768
1025
1280
1440
Nothing scrolls sideways at 320px, except wide data, inside its own box.
02.6 // Icons
Hairline, and never alone
DoneCautionAlarmInfoSearchFilterAIUndo
Drawn on a 24px grid with 2px of breathing room, shown dashed. Round ends. Stroke is 1.5 at 24px. Small icons are drawn heavier so they keep their weight: 1.6 for the theme switch at 16px, 1.8 at 15px.
Actions get a label or an accessible name. Severity differs by shape, not just colour.
02.7 // Elevation
Height is light, or shadow
Night0 Page1 Card2 Menu3 Dialog
Paper0 Page1 Card2 Menu3 Dialog
At night each level is a slightly lighter navy with a hairline. On paper each level casts a longer shadow, kept tight under the edge so the sheet stays clean.
Glow is reserved for one thing: a primary action under your cursor.
02.8 // Motion
Two modes, nothing for show
Borrowed from IBM Carbon's Carbon's two motion styles: productive for quick, efficient transitions, expressive for moments that deserve attention. motion. Productive handles changes you need but shouldn't notice. Expressive is for moments you trigger, with a small physical overshoot.
Productive // arrive and settle
Reveals, menus, underlines. 0.2 to 0.5s.
--ease-out-expo
Expressive // a little life
Button release, the theme switch marker and its icons. 0.45 to 0.7s.
--ease-spring-soft
Nothing loops. Every effect answers an action, and reduced motion shows every final state.
02.9 // Visual grammar
One signal, one meaning
Every visual signal in Datum means exactly one thing, everywhere it appears. That's what lets people learn the system once. It also caught a real mistake: the first AI cards were drawn in dashed lines, and dashes already meant "unavailable".
Solid lineLive and available: buttons, cards, fields, and AI.
Dashed lineNot available, not yet, or planned: disabled controls, planned items, untested status.
Heavier lineNeeds your attention: errors, the one caution tile.
CyanYou can act: links, buttons, focus, selection.
▲Amber and triangleLook here: caution, low confidence, high risk.
RedStop and read. With an octagon, something failed. On an action, it can't be undone.
Violet and sparkMade by AI. Never used for status.
Before Both read as "off"
AI suggestionSave
After AI is live, the button is off
AISuggestionSave
If two things look alike, people assume they behave alike. So a signal only ever gets one meaning.
System / 03// Components
A curated set, not a library
Thirteen components carry most product screens, and the last five are for AI. Each is documented the way mature systems do it, from anatomy and states to keyboard behaviour. Each one also says where it is live on this site, where it is only specified, and what has been tested. Product examples are illustrative, not client work.
03.0 // Atoms to pages
How the parts build up
Datum follows Brad Frost's model for design systems: atoms combine into molecules, molecules into organisms, organisms into templates, and templates become pages.. Each tier is built only from the one above it, so a change to a token reaches every page without anyone touching a component. Pick a part to follow it all the way down.
The site header at night: the name, three links with Work marked as the current page, the theme switch with the moon pressed, and a Connect link. On touch layouts the links and Connect are replaced by a menu button.
Two buttons, one pressed. The marker stretches across, lands on your choice and locks on. The icon fills in, and the new theme spreads from the point you pressed, or fades in browsers that can't do that. It is the same control as the one in the header, so pressing it changes this page.
Anatomy
AIdentity. The name and a status dot. Always goes home.
BLinks. The current page keeps its brackets and is announced as current, so it is never colour alone.
CTheme switch. Moon and sun, one pressed. The choice is saved under one shared key, so it follows you between pages.
DConnect. The bar's one outward link. An outline, so it never outranks the page's own primary action.
EMenu button. Replaces the links and Connect on touch layouts and opens a full-screen menu.
Variants
Links, switch, Connect
Header
Switch and menu button
Touch header
Full screen, numbered links
Phone menu
Sticky list of sections
Side index
Current sheet, progress
Progress bar
States
EnabledMuted text
HoverBrackets slide in, text takes the action colour
CurrentBrackets stay, announced as the current page
FocusRing, never removed
ScrolledThe bar's hairline firms up
Switch pressedIcon fills in and the marker sits under it
Spec // every number is the number in the code
Navigation measurements
Bar height
48px. 52px on touch layouts
Bar padding
32px. 20px on touch
Link
13px mono, tracking 0.08em
Link gap
32px. 24px under 1280px
Switch button
34 by 30px. 36 by 32px on touch, with a 44px tall hit area
Switch icon
16px, drawn on a 24 grid with a 1.6 stroke
Menu button
44 by 44px
Tokens by state
Navigation tokens by state
Link at rest
text --dim2
Hover and current
text --cyan + brackets
Bar
--void at 70% night, 84% paper, blurred
Hairline
10%, 18% once scrolled
Switch marker
--cyan at night, blue gradient on paper
Connect
text --cyan // edge --cyan at 60%
Keyboard and screen reader
Navigation keyboard and screen reader behaviour
Tab
Name, links, switch, then Connect, in reading order
Enter or Space
On the switch, sets the theme. Pressed and not pressed are announced
Escape
Closes the phone menu and returns focus to its button
Arrows, Home, End
Move between the component tabs on this page
Do and don't
✓ DoMark the current page in words and position, not just colour.
✕ Don'tMove navigation items around from page to page.
Agent rules // when an AI builds with it
Navigation agent rules
Allowed on
Nowhere. It is sealed, so an agent never emits it
Never
Emit, edit or restyle it.
Add or reorder its items.
Add a second nav landmark to a page.
Where it's used
LiveThe header above is the home page's, copied byte for byte, and so is the phone menu behind its menu button. A side index runs down case studies and the About page. The progress bar and the tabs you are using are on this page.
AColour bar and index. 3px and an outlined numeral, both in the project's colour.
BMeta. Position, category and where it was made, in 12px mono.
CTitle. The project's name.
DQuestion. What the case study answers, in one line.
EProof. A number, with its unit in words.
FLink. Explicit, and named for screen readers.
Variants
Text and schematic, alternating
Desktop row
Schematic first, colour bar, index
Phone and tablet card
States
EnabledHairline edge
HoverOn the desktop row the title takes the action colour. The link's underline firms up
FocusRing on the link
Spec // every number is the number in the code
Cards measurements
Padding
24px top, 20px sides
Radius
4px
Colour bar
3px
Title
32px condensed, 800. Up to 46px on the desktop row
Question
17px
Link
13px mono, tracking 0.08em
Actions
One
Tokens by state
Cards tokens by state
Surface
--panel, at 55% at night
Edge
hairline at 14%
Bar and link
project colour: amber, cyan, then indigo
Paper depth
--elev-paper-1
Keyboard and screen reader
Cards keyboard and screen reader behaviour
Tab
Reaches the single link
Screen reader
Hears "Read the case study: Datum design system"
Do and don't
✓ DoOne clear link whose name says where it goes.
✕ Don'tWrap the whole card in a link. Every word on it becomes the link's name.
Agent rules // when an AI builds with it
Card agent rules
Allowed on
Controlled and declarative screens
Must set
A title.
One destination.
Two stats at most.
Never
A card inside a card.
More than one link.
Where it's used
LiveCase study cards on the home page. This is the one that leads here, at phone size, with its schematic and its second number left out. On desktop the same content becomes a two-column row.
A chart that looks live but isn't is worse than no chart.
03.9 // AI suggestion
AI suggestion
Proposals from a crew member, never instructions. People should see at a glance what AI made, how sure it is, and how to undo it.
Use forSuggestions someone reviews before anything changes.
Not forChanging records on its own, or hiding that AI was involved.
Night
AIASuggestionBMedium confidence
CLikely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Based on
3 maintenance logs and 2 similar cases
Couldn't check
Sensor 4 was offline for 6 hours
Model
Updated 2 days ago
E
Paper
AISuggestionMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Based on
3 maintenance logs and 2 similar cases
Couldn't check
Sensor 4 was offline for 6 hours
Model
Updated 2 days ago
Anatomy
AAI label. Violet, a spark and the word AI, so it never relies on colour alone. Carbon uses the same idea, its AI label.
BConfidence. Words first, bars second. Low confidence turns amber and removes the one-tap accept.
CContent. The suggestion in plain words, never presented as fact.
DExplanation. What it used, what it couldn't check, and when the model last changed. Carbon calls this the explainability popover.
EActions. Accept, edit or dismiss. Nothing applies itself.
Variants
AISuggestionProposes, with sources
Suggestion card
Bearing wear on ENG-208Tab
Inline suggestion
AIAgent actionAsks before it acts
Agent request
Handed to a personWith what it found
Escalation
Bearing wear on ENG-208.AI-assistedEdited by you Revert
Accepted, with provenance
States
ProcessingA named task, a short progress line and Stop. No fake typing.
SuggestingAI label, confidence and a way into the explanation.
Low confidenceAmber. No one-tap accept; it asks for a review or escalates to a person.
AcceptedBecomes normal content, with a small "AI-assisted" mark that stays.
DismissedGone. Dashed here, because dashes mean "no longer available".
FailedOctagon, and a clear "no changes were made".
Spec // every number is the number in the code
AI suggestion measurements
Padding
14px
Edge
1px solid, violet at 45%
Aura
violet at 10%, fading over the top half
AI label
12px mono, violet on a 14% tint
Confidence
words, plus 3 bars of 5 by 12px
Actions
small buttons, 36px
Tokens by state
AI suggestion tokens by state
Label, edge
--ai
Aura
--ai at 10%
Confidence
--conf-high, --conf-medium
Low confidence
--conf-low
Failed
--alarm
Surface
--panel
Keyboard and screen reader
AI suggestion keyboard and screen reader behaviour
Tab
Explanation, then accept, edit, dismiss
Enter or Space
Opens the explanation, announced as expanded
Escape
Stops a running agent
Screen reader
Hears "AI suggestion, medium confidence" before the content
Do and don't
✓ DoGive AI its own colour and a spark, and keep its lines solid like anything else that's live.
✕ Don'tDraw AI in dashes. Dashes already mean "unavailable", so the card reads as switched off.
Agent rules // when an AI builds with it
AI suggestion card agent rules
Allowed on
Controlled and declarative screens
Must set
The AI label.
Confidence.
A source, or a line saying what could not be checked.
Accept, edit and dismiss.
Never
Apply itself.
Show a bare percentage.
A one-tap accept at low confidence.
Where it's used
SpecifiedNothing on this site is generated while you read it, so nothing here is live. These are drawn for AI-assisted product screens, and the examples are illustrative, not client work.
Marked, explained and reversible is what makes AI output safe enough to act on.
03.10 // Evidence
Evidence
An answer is only as good as what it stands on. Each claim carries a numbered chip, and the chip shows the passage it came from.
Use forAny AI answer or suggestion that states a fact.
Not forDecoration. A chip with nothing behind it is worse than no chip.
Night
AIAnswerMedium confidence
ALikely cause: bearing wear on ENG-208.B The vibration pattern matches two past cases.
CEvidence3 sources
1
DMaintenance log, ENG-208Log entry · 14 MarEVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Past case · ENG-301Wear found on bearing 2 at strip-down.
FCouldn't checkSensor 4 was offline for 6 hours.
Paper
AIAnswerMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Evidence3 sources
1
Maintenance log, ENG-208Log entry · 14 MarVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Past case · ENG-301Wear found on bearing 2 at strip-down.
Couldn't checkSensor 4 was offline for 6 hours.
Anatomy
AClaim. One sentence that can be checked.
BCitation chip. A number in a 24px box, right after the claim it supports. Cyan, because you can act on it.
CEvidence header. Says how many sources there are before anyone opens them.
DSource. Title, kind and date, so people can judge it at a glance.
EPassage. The words the AI relied on, quoted, never paraphrased.
FGap. What it couldn't check. Always shown, even to say nothing was missed.
Variants
1
Selected
2
Resting
3
Source unavailable
Under the answer
Short answers
Beside the answer
Long answers, many sources
States
ClosedOnly the chips show. The sources are one tap away.
OpenThe source list sits under the answer.
Source selectedThe chip fills, its source lights up and shows the passage.
Source unavailableDashed chip. The source can't be reached any more, and the claim says so.
No sourcesAmber. The answer is marked as unverified and offers no action.
Spec // every number is the number in the code
Evidence measurements
Chip
24 by 24px, 12px mono, 2px radius
Chip edge
1px, cyan at 50%
Cited text
line height 1.9, so chips never touch the line above
Source row
8px padding, 24px number column
Passage
13px, 2px rule on the left
Gap line
13px, with a 12px mono label
Tokens by state
Evidence tokens by state
Chip, selected source
--cyan
Selected fill
--cyan at 6%
Unavailable
--dim2, dashed
Gap label, no sources
--amber
Surface
--panel
Keyboard and screen reader
Evidence keyboard and screen reader behaviour
Tab
Each chip, in reading order
Enter or Space
Selects the source, announced as pressed
Screen reader
Each chip reads "Source 1" and the title of the source
Do and don't
✓ DoQuote the passage the claim rests on.
✕ Don'tLink to a whole document and call it a source.
Agent rules // when an AI builds with it
Evidence panel agent rules
Allowed on
Controlled and declarative screens
Must set
Source title, kind and date.
The passage quoted, word for word.
What could not be checked.
Never
Cite a source the AI did not read.
Number a chip with no entry behind it.
Paraphrase inside the quote.
Where it's used
SpecifiedNothing on this site is generated while you read it, so this is not live here. Drawn for AI-assisted product screens. Names, IDs and sources are illustrative, not client work.
A claim people can check in one tap is a claim they can accept or reject quickly.
03.11 // Intent preview
Intent preview
Before an agent acts, it shows the plan. People see every step, what will change and what will not, then choose.
Use forAny agent action at risk tier 2 or above.
Not forReads and drafts. Asking permission for harmless work trains people to tap through.
Night
AIAPlanBR2Risk tier 2: Medium risk
CReassign 3 work orders to Line 2
DMove ENG-208, ENG-114 and ENG-301 to Line 2
Tell the Line 2 supervisor
Update Thursday's shift plan
EChanges
3 work orders, 1 shift plan
Won't touch
Parts orders, sign-offs
FUndo
For 24 hours
G
Paper
AIPlanR2Risk tier 2: Medium risk
Reassign 3 work orders to Line 2
Move ENG-208, ENG-114 and ENG-301 to Line 2
Tell the Line 2 supervisor
Update Thursday's shift plan
Changes
3 work orders, 1 shift plan
Won't touch
Parts orders, sign-offs
Undo
For 24 hours
Anatomy
ALabel and type. AI, and the word Plan. Nothing has happened yet.
BRisk tier. Set by the action, never by the AI.
CIntent. One sentence, verb first, with the count in it.
DSteps. Every one, in order. No step is left out.
EScope. What changes, and what it won't touch.
FUndo. How long the action can be taken back.
GChoices. Proceed, edit the plan, or do it yourself.
Variants
One line, same three choices
Single step
Numbered, in order
Multi-step
R3Risk tier 3: High risk
Step that needs its own gate
States
ProposedWaiting. Nothing runs until Proceed.
EditingSteps become editable. The plan is checked again before it runs.
ProceedingHands over to the Executing state, one step at a time.
Handled by a personThe plan is withdrawn. Dashed, because it is no longer available.
Out of dateSomething changed since the plan was made. It asks to plan again.
Spec // every number is the number in the code
Intent preview measurements
Width
up to 340px
Padding
14px
Intent
21px condensed
Steps
14px, 20px number column, 8px apart
Scope
13px, with 12px mono labels
Actions
small buttons, 36px
Tokens by state
Intent preview tokens by state
Label, edge, step numbers
--ai
Risk tier
--risk-medium to --risk-irreversible
Proceed
--btn-bg
Out of date
--amber
Surface
--panel
Keyboard and screen reader
Intent preview keyboard and screen reader behaviour
Tab
Proceed, then Edit plan, then Do it myself
Enter or Space
Activates the choice in focus
Escape
Closes the preview. Nothing runs
Screen reader
Hears "AI, plan, risk tier 2", then the intent, then the steps as a numbered list
Do and don't
✓ DoSay what the plan won't touch. It answers the question people are afraid to ask.
✕ Don'tFold five steps into "and related updates".
Agent rules // when an AI builds with it
Intent preview agent rules
Allowed on
Controlled screens only
Must set
The intent.
Every step.
What changes and what it will not touch.
The undo window.
The risk tier.
Never
Start before Proceed.
Leave a step out of the list.
Change the plan after it was approved.
Where it's used
SpecifiedNothing on this site is generated while you read it, so this is not live here. Drawn for AI-assisted product screens. Names, IDs and sources are illustrative, not client work.
People approve what they can see. The plan is the consent.
03.12 // Approval gate
Approval gate
The check between an agent and an action. It gets heavier as the action gets harder to take back.
Use forEvery agent action that changes something.
Not forSlowing down safe work. If everything asks for approval, nothing gets read.
Scale the gate // pick a risk tier
R3, high risk. A dialog. The scope is listed, a reason is recorded, and focus starts on Cancel.
Night
AIAgent requestAR3Risk tier 3: High risk
BReschedule 40 inspections?CAcross 2 sites. 12 are due this week.
Changes
40 inspections, 2 site plans
Undo
For 24 hours, then by hand
DReason, kept in the logSite 2 is closed on Friday
ECancelFReschedule 40
Paper
AIAgent requestR3Risk tier 3: High risk
Reschedule 40 inspections?Across 2 sites. 12 are due this week.
Changes
40 inspections, 2 site plans
Undo
For 24 hours, then by hand
Reason, kept in the logSite 2 is closed on Friday
CancelReschedule 40
Anatomy
ARisk tier. It decides everything below it.
BQuestion. The action and the count, asked as a question.
CConsequence. Who and what it reaches, in plain words.
DReason. Required from R3 up, and kept in the log.
ESafe action. First in the row, and where focus starts.
FConfirm. Named by verb and count. Red only at R4.
Variants
R1Risk tier 1: Low risk
No gate
R2Risk tier 2: Medium risk
Intent preview
R3Risk tier 3: High risk
Dialog and a reason
R4Risk tier 4: Irreversible
Typed confirmation
States
WaitingOpen, with focus on the safe action.
Reason missingConfirm stays off until a reason is given.
ID not matchedR4 only. Confirm is dashed until the typed ID matches.
ApprovedCloses. The log records who, when and why.
DeclinedCloses. Nothing ran, and the log says so.
Timed outThe request is withdrawn. The agent has to ask again.
Spec // every number is the number in the code
Approval gate measurements
Width
up to 340px
Padding
18px
Radius
4px
Question
21px condensed
Reason field
44px tall
Actions
right-aligned, 8px apart, safe action first
Tokens by state
Approval gate tokens by state
R1
--risk-low
R2
--risk-medium
R3
--risk-high
R4, and its confirm button
--risk-irreversible
Surface
--panel
Keyboard and screen reader
Approval gate keyboard and screen reader behaviour
On open
Focus goes to the safe action, never to confirm
Tab
Stays inside the gate
Escape
Declines. Nothing runs
Screen reader
Hears the risk tier first, then the question
Do and don't
✓ DoPut the count in the question: "Reschedule 40 inspections?"
✕ Don'tAsk for approval on reads and drafts. People stop reading gates that always say yes.
Agent rules // when an AI builds with it
Approval gate agent rules
Allowed on
Controlled screens only
Must set
The risk tier, from the action definition.
The scope.
The safe action first.
A reason at R3.
A typed confirmation at R4.
Never
Draw its own gate. An agent can ask for one, never render one.
Lower its own tier.
Put focus on the confirm button.
Put several actions behind one gate.
Where it's used
LiveThe tier selector on this panel, as a working example.
SpecifiedThe gates it shows. Nothing on this site is generated while you read it, so this is not live here. Drawn for AI-assisted product screens. Names, IDs and sources are illustrative, not client work.
Friction should match the cost of being wrong. No more, no less.
03.13 // Activity log
Activity log
Everything an agent did, who approved it, and whether it can still be taken back.
Use forAny product where an agent changes records.
Not forSystem logs for engineers. This one is written for the people doing the work.
Night
AIAgent activityThis week
A14:32
BReassigned 3 work orders to Line 2
CApproved by A. Purohit DR2Risk tier 2: Medium riskE23 h left
11:20
Rescheduled 40 inspections
Approved by A. Purohit R3Risk tier 3: High riskUndone by you at 11:42
Mon 16:40
Reordered the inspection queue
Approved by A. Purohit R2Risk tier 2: Medium riskFUndo closedafter 24 h
Mon 09:15
Released ENG-208 to service
Carried out by A. Purohit R4Risk tier 4: IrreversibleCan't be undone
Paper
AIAgent activityThis week
14:32
Reassigned 3 work orders to Line 2
Approved by A. Purohit R2Risk tier 2: Medium risk23 h left
11:20
Rescheduled 40 inspections
Approved by A. Purohit R3Risk tier 3: High riskUndone by you at 11:42
Mon 16:40
Reordered the inspection queue
Approved by A. Purohit R2Risk tier 2: Medium riskUndo closedafter 24 h
Mon 09:15
Released ENG-208 to service
Carried out by A. Purohit R4Risk tier 4: IrreversibleCan't be undone
Anatomy
ATime. When it happened, newest first.
BAction. Past tense, with the count.
CApprover. A named person, or "No approval needed".
DRisk tier. The same tag the gate showed.
EUndo. With the time left, so nobody has to guess.
FClosed. Dashed, once the window has passed or the action can't be undone.
Variants
A page of its own
Full log
One row, when an agent finishes
Latest action
R3Risk tier 3: High risk
Filtered by risk tier
States
UndoableUndo, and the time left.
UndoneStruck through, with who undid it and when.
Window closedDashed. Undo is gone, and it says after how long.
Can't be undoneR4 actions. Said in the gate first, repeated here.
Undo failedOctagon, and what state things are in now.
Spec // every number is the number in the code
Activity log measurements
Width
up to 360px
Row padding
12px by 16px
Time column
52px
Action
14px
Approver line
13px
Undo
small button, 36px, with the time left in 12px mono
Tokens by state
Activity log tokens by state
AI label
--ai
Undo
--cyan
Risk tag
--risk-low to --risk-irreversible
Closed
--dim2, dashed
Undo failed
--alarm
Surface
--panel
Keyboard and screen reader
Activity log keyboard and screen reader behaviour
Tab
Each Undo, newest first
Enter or Space
Undoes, then says what was restored
Screen reader
Each row reads the time, the action, the approver and the risk tier, then the undo state
Do and don't
✓ DoShow the time left next to Undo.
✕ Don'tHide Undo in a menu, or leave a dead Undo button once the window has passed.
Agent rules // when an AI builds with it
Activity log agent rules
Allowed on
Controlled screens only
Must set
The time.
The action in the past tense.
Who approved it.
The risk tier.
The undo state.
Never
Edit or delete an entry.
Show Undo after the window closes.
Write its own entries. The system writes the log.
Where it's used
SpecifiedNothing on this site is generated while you read it, so this is not live here. Drawn for AI-assisted product screens. Names, IDs and sources are illustrative, not client work.
Autonomy is only safe when people can see what happened and take it back.
System / 04// Patterns
Where systems earn their keep
Components are words. Patterns are sentences. This is where a system saves teams the most time, because the hard decisions have already been made, tested and written down.
✓Plan✓Inspect3Repair4Verify5Release
Step 3 of 5 // saved 1 min ago
Hand overSave and exitContinue
04.1 // Multi-step workflows
Gated steps, never lost
Long, multi-person tasks fail in the middle: people lose their place, skip a check, or can't hand over cleanly.
Gates between steps: nothing is released before it's verified.
Progress and save state always visible.
Any step can be handed over with its context attached.
re-deciding stepper, save and handover rules for every new workflow.
5 normal // 1 needs attention
ENG-114
▲ ENG-208 check
ENG-301
ENG-402
ENG-517
ENG-623
04.2 // Dashboards
Lights-out dashboards
Dashboards become walls of colour where nothing stands out.
Normal tiles stay neutral.
Only exceptions get colour, plus a shape and a word.
A one-line summary first, detail on demand.
arguing about which metric deserves colour.
bearing
Status: check ×Fleet: A320 ×+ Filter
24 resultsClear all
04.3 // Search and filtering
Filters you can see
Filters hide results without saying so, and people stop trusting search.
Active filters always visible and removable.
The result count updates and is announced.
Filters live in the URL, so a view can be shared.
rebuilding filter logic and no-result handling for every list.
2 things need fixing
Inspection dateTorque limit
Part numberInspection date
04.4 // Forms
One mistake, one fix
Long forms punish a single mistake with a wall of red and lost input.
One column, labels above, grouped by task.
Check a field when people leave it, never on every keystroke.
An error summary at the top links straight to each field.
debating validation timing for every form.
2 selectedAssignExport
AssetHours ▼
1,240
980
612
455
04.5 // Tables and dense information
Density is a choice
Experts need to see a lot at once. Everyone else drowns in it. Try the toggle.
Comfortable or compact, remembered per person.
Header and first column stay put.
Bulk actions appear only once something is selected.
reinventing sticky headers, selection and density per table.
SubmittedReviewedDecision
What changed
Amount12,00014,500Supplierunchanged
Return with reasonApprove
04.6 // Approvals
Show the change, not the form
Approvers rubber-stamp because they can't see what changed.
The diff comes first; the full record is one step away.
Returning needs a reason. Approving doesn't need a speech.
Every decision lands in an audit trail.
designing audit trails and diff views from scratch.
Loading state: placeholder shapes show the layout that is coming.
No inspections yet
Inspections you log will appear here.
Log an inspection
Couldn't load inspections
The planning system didn't respond. Nothing you entered was lost.
RetryWork offline
ENG-208
ENG-114
ENG-301
04.7 // Loading, empty and error
The states teams forget
They're the ones users remember. Switch between them above.
Loading shows the shape of what's coming.
Empty invites the first action.
Errors say what happened, what's safe, and what to do next.
every team writing its own "something went wrong".
Person
Brief→✓ Draft A✓ Draft B✕ Draft C→Commit
AI lane
Draft ADraft BDraft CSources attached
Log // drafts by AI, accepted and committed by A. Purohit
04.8 // Human-in-the-loop AI
AI drafts, people commit
AI that writes straight into the record takes control away and makes errors hard to trace.
AI works in its own marked lane.
People accept item by item, never everything by default.
Only a person can commit, and the log says who.
inventing an AI review flow per team, and the trust problems that follow.
System / 05// Intelligence
AI that knows its place
In safety-critical work an AI is a crew member, not the captain. These rules keep a person A person reviews or approves what an automated system proposes before it takes effect.: how the AI speaks, shows doubt, acts, and hands control back.
Human control
Nothing applies itself
People can stop, check or undo anything an agent does.
Transparency
Always marked as AI
In the interface and in the record. It never passes as human work.
Confidence
Doubt changes the UI
Shown in words and bars, so people neither over-trust nor under-trust it. Low confidence changes what's offered, not just a number.
Explainability
Every answer has a "why"
The sources it used, and what it couldn't check.
Grounded in Microsoft's 18 Guidelines for Human-AI Interaction, IBM Carbon for AI and Emily Campbell's Shape of AI. The first guideline pair says it best: make clear what the system can do, and how well it can do it. The agent parts build on Victor Yocco's patterns for agentic AI in Smashing Magazine: intent preview, action audit and undo, and a way to escalate.
05.1 // Signals
Three questions, three signals
Before anyone reads a word of AI output, they should be able to answer three questions. Each answer has its own shape, so the three can sit side by side without blurring.
Who made this?
AI marker
AIAI-assisted
Violet, a spark and a word. AI for what it made, AI-assisted once a person has edited or accepted it. No mark means a person made it.
--ai
How sure is it?
Confidence
High confidence Medium confidence Low confidence
A word first, three bars second. Low turns amber and takes the one-tap accept away.
--conf-high--conf-medium--conf-low
What's at stake?
Risk tier
R1Risk tier 1: Low riskR2Risk tier 2: Medium riskR3Risk tier 3: High riskR4Risk tier 4: Irreversible
A tier code and a word. The action sets the tier, never the AI, and the tier decides the gate.
Confidence is about the AI. Risk is about the action. They never share a shape, so a confident AI can't make a risky action look safe.
05.2 // Lifecycle and parts
One task, eight stages, thirty parts
An agent is always at exactly one of these stages, and each stage has one look, so people can tell at a glance whether it is waiting, working or done. Pick a stage, then a part, and try it: the buttons inside work, and the switch under it steps through its states, 91 across the 25 parts. The five signals every stage carries, the AI label, confidence, risk tier, citation chip and AI header, are in 05.1.
01Pick a stage, then a part8 stages · 25 parts
If it cannot go on
Idle. Available, but silent. Nothing appears until someone asks, or until it has something worth saying.
The UIWhat it can see and how far it may go are settled before anyone types.
Processing. Shows that it is working and on what. No fake typing, no endless spinner.
The UIIts reasoning, folded, every system it touches, and what the run is costing.
Suggesting. Marked as AI, with confidence in words and the sources it used. It proposes; it never applies.
The UILabel, confidence, a citation on each claim, and accept, edit or dismiss.
Awaiting approval. Before an agent acts, it shows the plan: what will change, what it will not touch, and whether it can be undone.
The UIA plan, a gate sized to the risk, and a question when it would otherwise guess.
Executing. Progress item by item, and it can be stopped part-way without leaving a mess.
The UIThe step it is on, and Stop one key away.
Completed, undoable. A plain account of what changed, with undo, and a log entry naming who approved it.
The UIA receipt, the log, a way to say if it helped, and a mark on anything that leaves.
Escalated to a person. When it cannot go on safely, it does not guess. It hands the case to a named person with everything it found.
The UIAn amber edge, the reason it stopped, and a package nobody has to ask about twice.
Failed safely. Says what failed, what is safe, and the next step. Nothing half-done is left hidden.
The UIAn octagon, "nothing was changed", then a way to retry or hand over.
02Viewing
AI suggestion cardOrganism · 6 states · press anything in it, or step through its states
1 / 5
Live // press anything
Reads
ENG-208 logsLine 2 ordersParts stock
Can change
Drafts
Never sees
EmailHR records
Press the pill to see the full scope
Reads
ENG-208 logsLine 2 ordersParts stock
Can change
Drafts
Never sees
EmailHR records
The agent asked to read HR records. Only an admin can widen its scope, from the next run.
3 states
Open. Everything the AI can read, may change and never sees, before anyone types.
Atom · AI
Scope indicator
What the AI can read, what it may change, and what it never sees, before anyone types a word.
Use when
Every assistant surface and every agent run, next to the input or the run.
Not for
Permissions management. This shows scope; settings change it.
If an AI builds it, it must set
What it reads.
What it may change.
What it never sees.
Never
Show a narrower scope than the agent really has.
Hide the indicator while the agent is working.
Draw "never sees" with a solid edge.
Live // press anything
3 sources · drafts only
Draft the handover note for Line 2, and flag anything due before Friday.
Enter to send
3 sources · drafts only
Draft the handover note for Line 2, and flag anything due before Friday.
Could not reach the parts system. Your draft is kept.
Enter to send
4 sources · drafts only
Draft the handover note for Line 2, and flag anything due before Friday.
Added: torque spec, PDF. It is read for this message only.
Enter to send
3 states
Ready. The scope sits above the words, and Send says its shortcut.
Molecule · AI
Prompt composer
The scope sits above the words, so people know what the AI will read before they press Send.
Use when
The one place people ask an assistant something.
Not for
Search. A search box finds records; this asks for work.
If an AI builds it, it must set
A label for the field.
The scope indicator.
Send, with its keyboard shortcut written out.
Never
Send on its own.
Clear the draft after an error.
Use a placeholder as the only label.
Live // press anything
Parts agentFinds parts and drafts orders
Active
May use
Parts systemOrders · draft
May not
Sign-off
Ceiling
R2Medium
Hands to
Planner agent →
Parts agentFinds parts and drafts orders
Paused
May use
Parts systemOrders · draft
May not
PaymentsSign-off
Ceiling
R2Medium
Hands to
Planner agent →
Paused by A. Purohit at 14:10. Two drafts are kept.
Parts agentFinds parts and drafts orders
Active
May use
Parts systemOrders · draft
May not
Sign-off
Ceiling
R2Medium
Hands to
Planner agent →
Payments are R4: they cannot be taken back. No agent may make one. A person pays from the finance tool.
3 states
Active. What the agent is for, what it may touch, its risk ceiling and who it hands to.
Molecule · AI
Agent role card
In a team of agents: what each one is for, what it may touch, its risk ceiling and who it hands to.
Use when
Products where several agents share work: planning, parts, scheduling.
Not for
A single assistant. One agent needs a scope indicator, not a roster.
If an AI builds it, it must set
A plain name and one-line role.
What it may use and what it may not.
Its risk ceiling.
Who it hands over to.
Never
Give two agents the same name or role.
Claim a tool it has no access to.
Hide which agent is active.
Live // press anything
Autonomy · this taskReschedule 40 inspectionsR3High risk
Ask firstLevel 2 of 4 · the ceiling
Autonomy · this taskReschedule 40 inspectionsR3High risk
SuggestLevel 1 of 4 · below the ceiling
Autonomy · this taskReschedule 40 inspectionsR3High risk
Ask firstLevel 2 of 4 · the ceiling
Act, then tell is above the ceiling for R3 work. Only an admin can raise the ceiling.
3 states
At the ceiling. Ask first is as far as this task may go: its riskiest step is R3.
Molecule · AI
Autonomy dial
How far an agent may act on its own, set per task. The riskiest step sets the ceiling, so R3 work always asks first.
Use when
Any agent task that runs more than once.
Not for
One-off requests. Those use the intent preview.
If an AI builds it, it must set
The task it applies to.
The current level.
The ceiling, from the highest risk tier in the task.
Never
Let the agent raise its own level.
Offer a level above the ceiling.
Apply a new level to a run already going.
Live // press anything
Assistant memory
Torque values in N·m, never ft·lb
Chat · 2 Oct
Line 2 hands over at 18:00 on weekdays
Shift plan · 28 Sep
Summaries under 100 words
Forgotten by you, todayGone
Only what is listed here is used
Assistant memory
Torque values in N·m, never ft·lb
Forgotten by you, just nowGone
Line 2 hands over at 18:00 on weekdays
Shift plan · 28 Sep
Summaries under 100 words
Forgotten by you, todayGone
Only what is listed here is used
Assistant memory
Torque values in N·m, never ft·lb
Chat · 2 Oct
Line 2 hands over at 18:00 on weekdays
Shift plan · 28 Sep
Summaries under 100 words
Forgotten by you, todayGone
New memories are not saved. The 2 listed are still used until you forget them.
Assistant memory
Torque values in N·m, never ft·lb
Chat · 2 Oct
Line 2 hands over at 18:00 on weekdays
Forgotten by you, just nowGone
Summaries under 100 words
Forgotten by you, todayGone
Only what is listed here is used
Assistant memory
Torque values in N·m, never ft·lb
Forgotten by you, just nowGone
Line 2 hands over at 18:00 on weekdays
Forgotten by you, just nowGone
Summaries under 100 words
Forgotten by you, todayGone
Nothing is remembered now. Answers use only what you type.
5 states
Saving. Each memory in plain words, where it came from and when, with Forget for each.
Organism · AI
Memory panel
Everything the assistant remembers, in plain words, with where it came from. Forget works on one memory at a time.
Use when
Any assistant that carries anything from one session to the next.
Not for
Settings. This lists what it knows about you; settings change how it works.
If an AI builds it, it must set
Each memory in plain words, with where it came from and when.
Forget, for each memory.
Whether new memories are being saved.
Never
Keep or use a memory after Forget.
Remember anything that is not listed here.
Hide the switch that stops saving.
Live // press anything
Not for airworthiness release. Use it to find likely causes. A licensed engineer decides.
AILikely causeHigh
Bearing wear on ENG-208, from 3 logs.
Not for airworthiness release. Use it to find likely causes. A licensed engineer decides.
AILikely causeHigh
Bearing wear on ENG-208, from 3 logs.
Sign off a release, or say a part is airworthy
See parts that are not in stores
Read logs older than 2 years
2 states
Shown. What this AI must not be used for, said where the decision is made. It cannot be dismissed.
Molecule · AI
Caveat banner
What this AI must not be used for, said where the decision is made. Specific, never dismissed.
Use when
Screens where an AI result could be mistaken for a decision: release, diagnosis, approval.
Not for
General disclaimers. If it applies everywhere, it belongs in onboarding.
If an AI builds it, it must set
What it must not be used for.
Who decides instead.
Never
Be dismissed or collapsed.
Say "AI can make mistakes" and nothing else.
Appear on every screen; only where the limit applies.
Live // press anything
AIReasoning5 steps · 1 assumed
Likely bearing wear: 2 of 3 logs show vibration rising for 14 days.
Before you accept: what do you know that this missed?
AIReasoning5 steps · 1 assumed
Likely bearing wear: 2 of 3 logs show vibration rising for 14 days.
Read
3 maintenance logs 1
Compared
Vibration trend, sensors 2 to 5
Assumed
Sensor 4 is calibrated
Ruled out
Loose mount: torque unchanged 2
Checked
Parts stock: 3 in bin B-12
Before you accept: what do you know that this missed?
AIReasoningMedium
Likely bearing wear: 2 of 3 logs show vibration rising for 14 days.
Read
3 maintenance logs 1
Compared
Vibration trend, sensors 2 to 5
Assumed
Sensor 4 is calibrated
Ruled out
Loose mount: torque unchanged 2
Checked
Parts stock: 3 in bin B-12
You added: sensor 4 was replaced on 9 Oct and not yet calibrated. Bearing wear is now medium confidence.
3 states
Folded. Folded by default, so the person thinks first. One line says how many steps and assumptions there are.
Organism · AI
Reasoning trace
What was read, compared, assumed and ruled out. Folded by default, and it asks what you know first.
Use when
Answers someone acts on: a likely cause, a recommendation, a risk call.
Not for
Small talk and lookups. If nobody will check it, fold it away entirely.
If an AI builds it, it must set
A one-line summary.
Each step, with its kind.
Sources for anything read.
Assumptions, marked as assumptions.
Never
Open by default on a decision the person must make.
Show raw model text as the reasoning.
Present an assumption as a fact.
Live // press anything
Parts system0.8 s
Search · read only
Asked
Bearing 6204, Line 2 stores
Got
3 in stock, bin B-12
Work ordersWrites
Draft order · waits for approval
Parts system0.8 s
Search · read only
stock.search · part 6204 · site Line 2 stores · fields: qty, bin
Asked
Bearing 6204, Line 2 stores
Got
3 in stock, bin B-12
Work ordersWrites
Draft order · waits for approval
Parts systemTimed out
Search · read only · 10 s
No answer after 10 seconds. Nothing was changed. The run waits here.
3 states
Done. Every system it touched, by name, with what it asked and what came back. Writes are flagged.
Molecule · AI
Tool call
Every system the agent touched, by name: what it asked, what came back, and whether it could change anything.
Use when
Inside answers and runs, wherever an agent queried or changed a system.
Not for
Developer logs. This is for the people relying on the answer.
If an AI builds it, it must set
The system, by its real name.
Read or write.
What was asked.
What came back, or the error.
Never
Hide a write inside a read.
Show raw payloads by default.
Leave out a failed call.
Live // press anything
Budget · this runResearch: ENG-208 history
Cost1.40 of 5.00 credits
Time17 of 20 minNearing the limit
At this rate: 2.10 credits, 23 min. It will ask before going over.
Budget · this runResearch: ENG-208 history
Cost1.40 of 5.00 credits
Time17 of 20 minNearing the limit
At this rate: 2.10 credits, 23 min. It will ask before going over.
14 of 31 sources read in 17 min. The other 17 are shorter, so about 6 more minutes and 0.70 credits.
Budget · this runResearch: ENG-208 history
Time20 of 20 minPaused at the limit
It needs about 3 more minutes and 0.30 credits to finish 17 sources.
3 states
Nearing the limit. Time and cost against limits a person set, with the estimate at the current rate.
Molecule · AI
Budget meter
Time and money a run may use, what it has used, and a promise to ask before going over.
Use when
Long or costly agent runs: batch updates, research across many sources.
Not for
Plan quotas and billing pages.
If an AI builds it, it must set
Each limit, set by a person.
What is used so far.
What happens at the limit.
Never
Go over a limit without asking.
Raise its own limit.
Show a percentage with no number behind it.
Live // press anything
AISuggestionMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
AISuggestionMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Based on
3 maintenance logs and 2 similar cases
Couldn't check
Sensor 4 was offline for 6 hours
Model
Updated 2 days ago
AI-assistedSuggestionAccepted
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Accepted by you at 14:32. Logged with your name.
AI-assistedSuggestionEdited by you
Likely cause: bearing wear on ENG-208, confirmed by the strip-down on 15 Oct.
AISuggestionDismissed
You dismissed this suggestion. It will not be shown again for ENG-208.
AISuggestionLow confidence
Possible causes: bearing wear, or a loose mount. Not enough to suggest an action.
6 states
Suggested. Marked as AI, with confidence in words. It proposes; nothing applies itself.
Organism · AI
AI suggestion card
Proposals from a crew member, never instructions. People see at a glance what AI made, how sure it is, and how to undo it.
Use when
Suggestions someone reviews before anything changes.
Not for
Changing records on its own, or hiding that AI was involved.
If an AI builds it, it must set
The AI label.
Confidence.
A source, or a line saying what could not be checked.
Accept, edit and dismiss.
Never
Apply itself.
Show a bare percentage.
A one-tap accept at low confidence.
Live // press anything
AIAnswerMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Evidence3 sources
1
Maintenance log, ENG-208Log entry · 14 MarVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Past case · ENG-301Wear found on bearing 2 at strip-down.
Couldn't checkSensor 4 was offline for 6 hours.
AIAnswerMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Evidence3 sources
1
Maintenance log, ENG-208Log entry · 14 MarVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Past case · ENG-301Wear found on bearing 2 at strip-down.
Couldn't checkSensor 4 was offline for 6 hours.
AIAnswerMedium confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Evidence3 sources
1
Maintenance log, ENG-208Log entry · 14 MarVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Past case · ENG-301Wear found on bearing 2 at strip-down.
Couldn't checkSensor 4 was offline for 6 hours.
AIAnswerLow confidence
Likely cause: bearing wear on ENG-208. The vibration pattern matches two past cases.
Evidence3 sources
1
Maintenance log, ENG-208Log entry · 14 MarVibration on bearing 2 has risen at each of the last three checks.
2
Work order WO-1182Past case · ENG-114Same pattern. Bearing replaced, readings back to normal.
3
Work order WO-0967Archived on 2 OctSource unavailable
Couldn't checkSensor 4 was offline for 6 hours.
Source 3 was archived, so the second claim now rests on one case. Confidence dropped to low.
4 states
Source 1. Each claim carries a numbered chip. Pressing one shows the passage it came from, word for word.
Organism · AI
Evidence panel
An answer is only as good as what it stands on. Each claim carries a numbered chip, and the chip shows the passage it came from.
Use when
Any AI answer or suggestion that states a fact.
Not for
Decoration. A chip with nothing behind it is worse than no chip.
If an AI builds it, it must set
Source title, kind and date.
The passage quoted, word for word.
What could not be checked.
Never
Cite a source the AI did not read.
Number a chip with no entry behind it.
Paraphrase inside the quote.
Live // press anything
AINext stepsHigh confidence
Bearing 2 on ENG-208 needs replacing.
Nothing runs until you pick one
AINext stepsHigh confidence
Bearing 2 on ENG-208 needs replacing.
Based on the likely cause (high confidence) and 3 logs 1.
Booking Bay 3 is R3 because it moves other teams' work. The order is R2: it is a draft until you send it.
AIPickedR1Low risk
Summarise the three logs
Done. The summary is under the answer, and the log has an Undo.
AIPickedR2Medium
Draft a parts order
You will see the draft, with every line, before anything is sent.
AIPickedR3High risk
Book Bay 3 for Thursday
Opens the approval gate: it moves 2 other jobs, so it needs a reason in the log.
5 states
Offered. Three next steps at most, each with its risk tier from its definition. Nothing runs until one is picked.
Molecule · AI
Suggested actions
Three next steps at most, each showing its risk before anyone picks it. Picking opens the right gate.
Use when
The end of an AI answer, where a next step is likely.
Not for
Navigation. Links go somewhere; these do work.
If an AI builds it, it must set
A verb-first label for each action.
The risk tier of each, from its definition.
Three actions at most.
Never
Run an action because it was shown.
Offer actions at low confidence.
Hide an action's risk until it is picked.
Live // press anything
AISuggested editHigh confidence
Handover note · Line 22 changes
Replace bearing 2 on ENG-208 next shift before the next flight1. Torque to 45 N·m. Sign-off by any lead the licensed engineer2.
1Timing
2Sign-off
AISuggested editHigh confidence
Handover note · Line 21 of 2 left
Replace bearing 2 on ENG-208 before the next flight1. Torque to 45 N·m. Sign-off by any lead the licensed engineer2.
1TimingAccepted
2Sign-off
AI-assistedAccepted editHigh confidence
Handover note · Line 2Accepted by you
Replace bearing 2 on ENG-208 before the next flight1. Torque to 45 N·m. Sign-off by the licensed engineer2.
1TimingAccepted
2Sign-offAccepted
AISuggested editHigh confidence
Handover note · Line 22 changes
Replace bearing 2 on ENG-208 next shift before the next flight1. Torque to 45 N·m. Sign-off by any lead the licensed engineer2.
1Timing
2Sign-off
The maintenance manual, section 4.2, says bearing work is signed off by a licensed engineer 1. The note said any lead.
AIRejected editHigh confidence
Handover note · Line 2Rejected by you
Replace bearing 2 on ENG-208 next shift. Torque to 45 N·m. Sign-off by any lead.
2 changes rejected. The note is as you wrote it.
5 states
Two changes. An AI edit shown as numbered changes against what was there, accepted one at a time.
Organism · AI
Suggestion diff
An AI edit as changes, not a rewrite. Each one is numbered and accepted on its own, and removed text is never red.
Use when
Any AI edit to text people own: notes, reports, procedures.
Not for
Version history. This is one proposal, not the record.
If an AI builds it, it must set
Each change, with what was there before.
Accept and reject for each change.
The number of changes.
Never
Apply a change before it is accepted.
Show only the result.
Mark removed text in red. Red means stop.
Live // press anything
AIRepair plan1 of 3
Swap bearing 2 on Monday. 3 h on Line 2. The seal waits for its own check.
DiffersShortest, but a second visit in 6 weeks
AIRepair plan2 of 3
Swap bearing 2 and the seal together on Monday. 5 h on Line 2.
Differs2 h longer than plan 1, no second visit
AIRepair plan3 of 3
Ground ENG-208 tonight and replace both bearings. 9 h, Bay 3.
DiffersMost cautious, but Bay 3 moves 2 other jobs
AI-assistedRepair planChosen by you
Your pick is now the plan for ENG-208.
Plans 1 to 3 are kept with the work order, with how each differed.
AICompare plans3 plans
Plan
Time
Visits
Moves
1
3 h
2
None
2
5 h
1
None
3
9 h
1
2 jobs
5 states
Plan 1 of 3. Alternate answers to one question. Each says how it differs, and nothing is picked for the person.
Molecule · AI
Answer branches
Alternate answers to one question, each with what makes it different. The person chooses; nothing is picked for them.
Use when
Plans and drafts where reasonable people would pick differently.
Not for
Facts. A fact has one answer; show the evidence instead.
If an AI builds it, it must set
How many there are and which one is shown.
What makes this one different.
A way to choose one.
Never
Show branches without saying how they differ.
Keep more than three without being asked.
Choose one for the person.
Live // press anything
AIPlanR2Medium risk
Reassign 3 work orders to Line 2
Move ENG-208, ENG-114 and ENG-301 to Line 2
Tell the Line 2 supervisor
Update Thursday's shift plan
Changes
3 work orders, 1 shift plan
Won't touch
Parts orders, sign-offs
Undo
For 24 hours
AIAgent action2 of 3
Reassign 3 work orders to Line 2
ENG-208 reassignedENG-114 reassignedENG-301
AIPlanR2Medium risk
Reassign 3 work orders to Line 2
Move ENG-208, ENG-114 and ENG-301 to Line 2
Tell the Line 2 supervisor
Update Thursday's shift plan
You removed step 2. The changed plan needs your approval again.
AIPlanYours
Reassign 3 work orders to Line 2
The plan is in your notes. The agent will not act on these work orders.
4 states
Proposed. Before an agent acts, it shows every step, what will change and what will not, then waits.
Organism · AI
Intent preview
Before an agent acts, it shows the plan: every step, what will change and what will not, then people choose.
Use when
Any agent action at risk tier 2 or above.
Not for
Reads and drafts. Asking permission for harmless work trains people to tap through.
If an AI builds it, it must set
The intent.
Every step.
What changes and what it will not touch.
The undo window.
The risk tier.
Never
Start before Proceed.
Leave a step out of the list.
Change the plan after it was approved.
Live // press anything
AIAgent requestR3High risk
Reschedule 40 inspections?Across 2 sites. 12 are due this week.
Changes
40 inspections, 2 site plans
Undo
For 24 hours, then by hand
Reason, kept in the logWhy now?
AIAgent requestR3High risk
Reschedule 40 inspections?Across 2 sites. 12 are due this week.
Changes
40 inspections, 2 site plans
Undo
For 24 hours, then by hand
Reason, kept in the logWhy now?Add a reason. It goes in the log.
AIAgent requestR3High risk
Reschedule 40 inspections?Across 2 sites. 12 are due this week.
Changes
40 inspections, 2 site plans
Undo
For 24 hours, then by hand
Reason, kept in the logSite 2 is closed on Friday
AIAgent requestR3High risk
Rescheduled 40 inspections
Approved by you at 14:32. Reason logged: Site 2 is closed on Friday.
24 h left
AIAgent requestR3High risk
Not rescheduled
Nothing changed. The agent was told no, and will not ask again today.
5 states
Waiting. An R3 action asks as a question, lists the scope and needs a reason. Cancel comes first.
Organism · AI
Approval gate
The check between an agent and an action. It gets heavier as the action gets harder to take back.
Use when
Every agent action that changes something.
Not for
Slowing down safe work. If everything asks for approval, nothing gets read.
If an AI builds it, it must set
The risk tier, from the action definition.
The scope.
The safe action first.
A reason at R3.
A typed confirmation at R4.
Never
Draw its own gate. An agent can ask for one, never render one.
Lower its own tier.
Put focus on the confirm button.
Put several actions behind one gate.
Live // press anything
AIQuestionPaused
Which line takes the 3 work orders?
Waiting before step 3: reassign
AIQuestionAnswered
Which line takes the 3 work orders?
You choseLine 2
Going on with step 3: reassign to Line 2 from 14:00.
AIQuestionAnswered
Which line takes the 3 work orders?
You choseLine 4
Going on with step 3. WO-0963 moves to Monday, and its owner is told.
AIQuestionAnswered
Which line takes the 3 work orders?
You choseLine 3, after the 16:00 check
Line 3 was not an option, so it checks capacity first and asks again if it is full.
4 states
Asked. One question, the decision it blocks, and each option with what it leads to. The run waits.
Molecule · AI
Clarifying question
When two answers would change the result, the agent asks instead of guessing. Each option says what it leads to.
Use when
Gaps only the person can close: a preference, a choice between two real records.
Not for
Facts it can check, and permissions. Those are a lookup and an approval gate.
If an AI builds it, it must set
The decision, named.
Each option with its consequence.
What is waiting on the answer.
Never
Guess when two answers change the result.
Ask what it could look up.
Ask more than one question at a time.
Live // press anything
AIDry runR2Medium
Reschedule 400 work orders
388 would change12 skipped: locked
First 3 of 388
WO-0961
Thu
Mon
WO-0962
Thu
Mon
WO-0964
Fri
Tue
Emails the Line 2 lead, once
Prepared 14:32 · changes need a new preview
AIDoneR2Medium
Rescheduled 386 of 388
Read back 386 of 386.
2 were changed by someone else after the preview. They were not overwritten: WO-1104, WO-1188.
AIDry runR2Medium
Reschedule 400 work orders
388 would change12 skipped: locked
First 3 of 388
WO-0961
Thu
Mon
WO-0962
Thu
Mon
WO-0964
Fri
Tue
The rule changed at 14:35. This preview no longer matches it.
Emails the Line 2 lead, once
Prepared 14:32 · changes need a new preview
AIUndoneR2Medium
Undid 386 changes
Read back 386 of 386. All are on their old dates.
The 2 records others changed were never touched.
4 states
Preview. The batch tried in full first: how many would change, which are skipped and why, and who gets told.
Organism · AI
Dry-run preview
A batch simulated before it runs: how many records would change, which are skipped and why, and who gets told.
Use when
Bulk changes: rescheduling, reassigning, status updates across many records.
Not for
One record. That is an intent preview.
If an AI builds it, it must set
How many records would change, with a sample before and after.
What is excluded, and why.
Side effects: who gets notified, what else changes.
When the preview was made.
Never
Run on more records than were approved.
Approve a plan that hides which records change.
Write over a record that changed since the preview.
Live // press anything
Waiting on you3
Question
Which line takes 3 work orders?
Planner agent · 25 min
Notify
CRS-118 needs a signature. Agents cannot sign.
Records agent · 1 h
ReviewR2Medium
Parts agent · 12 minSend parts order PO-2291
Orders
2 × bearing 6204, 1 × seal
Cost
₹18,400
Undo
Until the supplier confirms
Waiting on you0
Nothing is waiting. PO-2291 was approved by you and is in the activity log.
3 states
Waiting. Everything agents are waiting on, in one queue: reviews first, then questions, then notices.
Organism · AI
Agent inbox
Everything agents are waiting on, in one queue: things to review, questions, and notices. Nothing gets lost in a chat.
Use when
Agents working in the background, across many tasks, for one person or team.
Not for
A single conversation. Questions there stay inline.
If an AI builds it, it must set
Each item's kind: review, question or notify.
Which agent is asking.
How long it has waited.
The risk tier, for reviews.
Never
Bury a review under notices.
Act on an item the person has not answered.
Drop an item when it ages.
Live // press anything
AIWorkingStep 2 of 4
Order parts for ENG-208
Read 3 maintenance logs
Check parts stock
Draft the parts order
Ask you to approve it
Stop ends after this step. Step 1 stays in the log.
AIStoppedAfter step 2
Order parts for ENG-208
Read 3 maintenance logs
Check parts stock
Draft the parts order
Ask you to approve it
Stopped by you at 14:22. Steps 3 and 4 did not run.
AIYours nowFrom step 2
Order parts for ENG-208
Read 3 maintenance logs
Check parts stock
Draft the parts order
Ask you to approve it
You have it. The 3 logs it read are attached.
3 states
Working. The step it is on, the steps left, and Stop one key away.
Molecule · AI
Run control
An agent at work: the step it is on, what comes next, and Stop one keystroke away. Stopping never asks "are you sure?"
Use when
Any agent run longer than a few seconds.
Not for
Page loading. That is a progress indicator, not an agent.
If an AI builds it, it must set
The current step and the total.
Stop, with its shortcut.
Take over.
Never
Hide Stop while anything is running.
Ask "are you sure?" before stopping.
Start the next step after Stop.
Live // press anything
Receipt · 14:32R2Medium
Reassigned 3 work orders
WO-0961Line 4Line 2
WO-0962Line 4Line 2
WO-0964Line 3Line 2
Approved
A. Purohit
Done by
Planner agent
Verified
Read back 3 of 3
23 h left
Receipt · 14:40R2Medium
Undid: reassigned 3 work orders
WO-0961Line 2Line 4
WO-0962Line 2Line 4
WO-0964Line 2Line 3
Undone by
A. Purohit
Done by
Planner agent
Verified
Read back 3 of 3
Receipt · 14:32R2Medium
Reassigned 3 work orders
WO-0961Line 4Line 2
WO-0962Line 4Line 2
WO-0964Line 3Line 2
Approved
A. Purohit
Done by
Planner agent
Verified
Read back 3 of 3
The undo window closed 24 hours after the action. To change it back, ask the planner.
3 states
Done. One action, fully accounted for: each change, who approved it, proof it worked, and Undo.
Organism · AI
Action receipt
One action, fully accounted for: each change before and after, who approved it, proof it worked, and Undo.
Use when
Behind every row of the activity log, and wherever someone asks "what exactly did it do?"
Not for
Summaries for managers. This is the record, line by line.
If an AI builds it, it must set
Each change, before and after.
Who approved it, and which agent did it.
How it was verified.
The undo window.
Never
Report success without reading the result back.
Leave out a change because it was small.
Offer Undo after the window closes.
Live // press anything
AIAgent activityThis week
14:32
Reassigned 3 work orders to Line 2
Approved by A. Purohit R2Medium risk23 h left
Mon 16:40
Reordered the inspection queue
Approved by A. Purohit R2Medium riskUndo closedafter 24 h
Mon 09:15
Released ENG-208 to service
Carried out by A. Purohit R4IrreversibleCan't be undone
AIAgent activityThis week
14:32
Reassigned 3 work orders to Line 2
Approved by A. Purohit R2Medium riskUndone by you, just now
Mon 16:40
Reordered the inspection queue
Approved by A. Purohit R2Medium riskUndo closedafter 24 h
Mon 09:15
Released ENG-208 to service
Carried out by A. Purohit R4IrreversibleCan't be undone
AIAgent activityThis week
14:32
Reassigned 3 work orders to Line 2
Approved by A. Purohit R2Medium risk
Mon 16:40
Reordered the inspection queue
Approved by A. Purohit R2Medium riskUndo closedafter 24 h
Mon 09:15
Released ENG-208 to service
Carried out by A. Purohit R4IrreversibleCan't be undone
Could not undo. WO-0962 was changed by R. Iyer at 15:10. Nothing was undone.
3 states
Log. Everything an agent did, who approved it, and whether it can still be taken back.
Organism · AI
Activity log
Everything an agent did, who approved it, and whether it can still be taken back.
Use when
Any product where an agent changes records.
Not for
System logs for engineers. This one is written for the people doing the work.
If an AI builds it, it must set
The time.
The action in the past tense.
Who approved it.
The risk tier.
The undo state.
Never
Edit or delete an entry.
Show Undo after the window closes.
Write its own entries. The system writes the log.
Live // press anything
AILikely cause: bearing wear on ENG-208.
Was this useful?
AILikely cause: bearing wear on ENG-208.
Was this useful?
What was off?
Wrong sourceOut of dateMissed a stepToo long
The 14 Mar log was replaced
Sends this answer and its sources
AILikely cause: bearing wear on ENG-208.
Sent: this answer, its sources and your note. Thank you.
3 states
Asked. Two buttons under the answer. It never blocks the answer and asks once.
Molecule · AI
Feedback capture
Useful or not, then why, with the reasons people actually give. It says exactly what gets sent.
Use when
Under AI answers and suggestions.
Not for
Surveys about the product. This rates one answer.
If an AI builds it, it must set
The answer it belongs to.
Useful and not useful, as toggle buttons.
What is sent with it.
Never
Block the answer until feedback is given.
Ask more than once per answer.
Send anything beyond what it says it sends.
Live // press anything
Drafted by
Datum assistant
Edited by
A. Purohit
When
14 Oct, 09:12
Sources
3 logs, 1 manual
Full history in the record
Inspection report · PDFNo badge
Written by R. Iyer, with no AI drafting. Nothing to declare.
3 states
Badge. A small badge on anything that leaves the product: who made it and how much was AI.
Atom · AI
Provenance badge
Who made it, how much was AI, and who edited it since. It travels with anything exported or shared.
Use when
Reports, notes and images that leave the product: PDFs, emails, exports.
Not for
Labelling AI on screen while people work. That is the AI label.
If an AI builds it, it must set
Who or what made it.
Whether a person edited it, and who.
When.
Never
Mark something human-made when AI drafted any of it.
Sit on top of the content it describes.
Drop the badge when content is exported.
Live // press anything
Handed over14:40
Two logs disagree on the last torque check, so it stopped before ordering parts.
Read 3 logs
Order parts
RI
R. IyerLine 2 lead · notified
Nothing was changed.
Handover · ENG-20814:40
14:31Read 3 maintenance logs
14:3612 Oct log: torque 45 N·m. 13 Oct log: 40 N·m.
14:40Stopped before ordering parts
Question for R. Iyer: which torque check is right?
Yours now14:44
R. Iyer was told you took it. The agent will not touch ENG-208 until you hand it back.
3 states
Handed over. Why it stopped, what is done and not, and the named person it went to.
Molecule · AI
Handoff
When the agent cannot go on safely, it says why, what is done and what is not, and hands over to a named person.
Use when
Sources conflict, confidence is low, access is missing, or a person asks.
Not for
Errors the agent can retry. That is run control.
If an AI builds it, it must set
Why it stopped.
What is done and what is not.
Who it is handed to.
Whether anything changed.
Never
Make the person repeat what the agent already knows.
Retry quietly after handing over.
Hand over without saying whether anything changed.
The five core parts are specified in full, with anatomy and keyboard rules, under Components, 09 to 13. Props, variants, states, tokens and code for all thirty are in the open Datum system (opens in a new tab)↗. Names, IDs and readings are illustrative, not client work.
05.3 // Doubt and risk
The signal changes the screen
A signal that only decorates is noise. In Datum, confidence decides what the AI may offer, and risk decides how hard it is to say yes.
High
Offer a one-tap accept.
Still logged, still undoable.
Medium
Reasons first, then ask for a review.
Sources open by default.
Low
Don't suggest an action.
Hand the case to a person with what was found.
a bare percentage. "87%" tells people nothing about what to do next.
Risk tierWhat countsThe gateUndo
R1Risk tier 1: Low risk
What countsReads and drafts. Nothing outside the draft changes.Example: Summarise 3 maintenance logs
The gateNo gate. It runs, then shows in the activity log.
UndoNothing to undo.
R2Risk tier 2: Medium risk
What countsChanges your own work, and can be fully undone.Example: Reassign 3 work orders to Line 2
The gateIntent preview. One tap to proceed.
UndoUndo for a stated time.
R3Risk tier 3: High risk
What countsChanges other people's work, or many records at once.Example: Reschedule 40 inspections across 2 sites
The gateApproval dialog. Scope listed, focus on Cancel, reason recorded.
UndoUndo for a stated time, then by hand.
R4Risk tier 4: Irreversible
What countsCan't be taken back: delete, send, release.Example: Release ENG-208 to service
The gateThe agent prepares. A named person carries it out, with a typed confirmation.
UndoNone, and it says so first.
Medium confidence moves an action up one tier. Low confidence offers no action at all.
Scope shown before actingReversible or not, said up frontIrreversible needs explicit confirmationEvery action logged with who approved itStop is one keystroke away
05.4 // Generated interfaces
How much the AI may draw
When an AI assembles the screen itself, the question is how much freedom it gets. Datum sets that by surface, so the answer never depends on who built the feature.
LevelWhat the AI doesWhat it sendsAllowed on
Controlled
What the AI doesThe AI fills in values. The layout is fixed and tested.
What it sendsText and numbers for named slots.
Allowed onAlarm and control screens, approval gates, anything at R3 or R4.
Declarative
What the AI doesThe AI picks and arranges parts from the catalog.
What it sendsA description that names parts from catalog.json. Anything else is refused.
Allowed onWork screens: dashboards, tables, forms and the assistant panel.
Open-ended
What the AI doesThe AI writes its own markup.
What it sendsMarkup of its own, outside the catalog.
Allowed onSandboxes and prototypes only, in a labelled frame with no access to real records.
An approval gate is always controlled. An agent can ask for one. It can never draw one.
The three levels follow CopilotKit's split of generative UI into static, declarative and fully generated. This is policy, not a shipped renderer. Nothing on this site is generated while you read it, so here none of the three is in use.
Every AI part, by the stage of a task it belongs to, is in 05.2 Lifecycle and parts above. Eight of them are drawn large near the top of this page.
On the bench // dashed means planned, not built
Embedded app frameGenerated region statesAI-filled form fieldsAI-filtered tablesThinking and streaming motion
System / 06// Data visualisation
Charts on the same colour budget
A chart is a small screen with the same rules as the big one. Neutrals carry it, hue is spent only on meaning, and every hue has a second cue, so no reading depends on telling two colours apart.
Neutrals first
Grey does the work
Gridlines, axes and labels stay neutral. Hue is spent only where it means something.
Identity
Four series at most
In a fixed order, never cycled. A fifth folds into Other, or the chart becomes small multiples.
Attention
Amber is spent once
With several charts on a screen, limits and targets are neutral. Amber marks the one reading that needs a person.
Second cue
Colour is never the key
A direct label, a legend shape or a printed value goes with every hue, and every chart has a one-sentence summary.
06.1 // The colour budget
Six colours, each with a job
Charts don't get a palette of their own. Every chart colour is a token the rest of the screen already uses, so a chart can never out-shout the alarm next to it. Four series carry identity, in a fixed order. Two colours are locked to a meaning and only appear when the data means it.
Identity // fixed order, never cycled
Series 1--cyanA direct label where the line endsNight10.6 to 1Paper6.5 to 1
Series 2--cyan at 65%A direct labelNight5.0 to 1Paper3.2 to 1
Series 3--paperA direct labelNight16.0 to 1Paper16.8 to 1
Series 4--dim2A direct labelNight5.1 to 1Paper5.9 to 1
Locked to a meaning
Made by AI--aiThe spark in the legendNight8.2 to 1Paper6.7 to 1
Failed--alarmThe octagon in the legendNight6.2 to 1Paper6.2 to 1
Contrast against the chart panel, measured in both themes. Every series mark clears 3:1 and every label clears 4.5:1.
Sequential scale // one hue, five steps
0Close to nothing
1Low
2Middle
3High
4Peak
For magnitude, as in the heatmap. Always drawn with its numbers. The lightest step on Paper is only 2.15:1, which is why cells keep a 2px gap and the scale is always labelled.
Never a series
AmberThe screen's one attention point. If every limit were amber, nothing would stand out.
IndigoOn Paper it is nearly the same blue as the first series. Two colours nobody can tell apart are one colour.
Checked for colour blindness too: under simulated protan and deutan vision, neighbouring series stay well apart in OKLab, at least 17 on Paper and 19 at Night against a floor of 8.
06.2 // Anatomy
The same card, every chart
Every chart sits in the same card, in the same order: what it is, how fresh it is, and one sentence that says what it shows. People read the takeaway first and the drawing second, so the drawing only has to confirm it.
Bar chart of success rate by agent over 7 days: Triage-02 98.4%, Records-01 97.6%, Diag-01 96.9%, Triage-05 95.3%, Diag-04 92.7%, Plan-03 88.9%. The target is 95%.
The bar chart, pinned. Readings are illustrative.
ATitle. SECTION // DETAIL in the label style, so every card reads the same way.
BFreshness. Top right. A chart that looks live but isn't is worse than no chart.
CTakeaway. One line at 15px, weight 500, that says what the chart shows.
DReference line. 1px dashed, 5 on and 4 off, always labelled. Neutral when charts share a screen.
EAxis labels. Real text outside the drawing, never under 12px.
FLegend. A swatch and a word, kept even when lines are labelled where they end.
+Tooltip. Specified for hover and keyboard focus. It adds, it never gates: every value is also in a label, the legend or the summary.
Marks // every number is the number in the code
Chart marks
Line
2px, round joins
Bar
8px thick, from zero, 1px radius at the data end
Marker
8px, inside a 2px panel ring
Touching fills
A 2px panel gap, never a border
Ring and arc
12px ring on the donut, 8px arc on the gauge, square ends
Numbers
Share Tech Mono, tabular, 12px floor
Confidence // the word comes first
High confidence 86 of 100
Medium confidence 71 of 100
Low confidence 52 of 100
Never "86%". Band lines mark where Medium and High start, and the product sets them. The Datum AI console uses 60 and 80. Confidence never borrows the risk tag's shape.
States // what a chart says when it can't draw
LiveFreshness shown.
StaleGreys out to the quietest grey and says how old.
LoadingThe shape of what is coming, a named step and one progress line. No spinner, no shimmer.
EmptyA dashed frame, the reason in plain words, one next step. Dashed means not available, so empty is never amber.
ErrorOctagon, what is safe, retry.
06.3 // Chart parts
Ten parts, one question each
Pick a chart by the question it answers, not by how it looks. Stacked area for how a total moves and who carries it. Gauge for where one score sits against its bands. Heatmap for when it gets busy. Each part below comes with its anatomy, tokens, states and the rules an AI agent has to follow when it builds with it. Night and Paper are drawn one above the other, and the readings are illustrative.
01Pick a chart part7 charts · 3 around them
ChartsAround a chart
Testing keyTestedchecked on this pageSpecifiedwritten down, shown as a static exampleNot yetscreen reader testing is next
02Viewing
Stacked area chartOrganism · Specified · 1 of 10
1 / 10
06.3 // Stacked area chart
Stacked area chart
How a total moves over time, and who carries it.
Use forA total over time, split into four series at most.
Not forSeries that cross or need comparing one to one. A stack hides that. Use lines.
Night
Agent activity // By agent type
Updated 2 min ago
Volume peaked at 11:00 with 1,045 tasks an hour. Triage carries 44% of the load.
Triage
Diagnosis
Planning
Records
Tasks per hour
1,2009006003000
Triage 44%Diagnosis 27%Planning 18%Records 11%
16:0020:0000:0004:0008:0012:00
Hour, IST
Stacked area chart of tasks per hour by agent type over the last 24 hours. Volume fell to about 200 an hour overnight and peaked at 1,045 at 11:00. Triage 44%, Diagnosis 27%, Planning 18%, Records 11%.
Paper
Agent activity // By agent type
Updated 2 min ago
Volume peaked at 11:00 with 1,045 tasks an hour. Triage carries 44% of the load.
Triage
Diagnosis
Planning
Records
Tasks per hour
1,2009006003000
Triage 44%Diagnosis 27%Planning 18%Records 11%
16:0020:0000:0004:0008:0012:00
Hour, IST
Stacked area chart of tasks per hour by agent type over the last 24 hours. Volume fell to about 200 an hour overnight and peaked at 1,045 at 11:00. Triage 44%, Diagnosis 27%, Planning 18%, Records 11%.
Anatomy
ATitle and freshness. SECTION // DETAIL on the left, the last update on the right.
BTakeaway. One line that says what the chart shows.
CLegend. One swatch per series, in stack order.
DPlot. A wash per band, a 2px line on top of each.
EEnd labels. The series name and its share, where each band ends.
FHover. A crosshair, a marker on each line and a tooltip with every value and the total.
Tokens used
Stacked area chart tokens
Series 1 line and wash
--cyan, --cy-rgb at 26%
Series 2 line and wash
--cy-rgb at 65%, --cy-rgb at 12%
Series 3 line and wash
--paper, --paper at 14%
Series 4 line and wash
--dim2, --dim2 at 22%
Gridline
--hl-rgb at 12%
Baseline
--hl-rgb at 35%
Crosshair
--hl-rgb at 45%
Tooltip
--panel2, --hl-rgb at 30%
Spec // every number is the number in the code
Stacked area chart measurements
Line
2px, round joins
Marker
8px, in a 2px panel ring
Plot height
300px at least
Axis column
--space-500
End label column
--space-1400
Gaps
--space-100
Labels
12px mono, real text, never inside the drawing
Variants
Four series
Fewer series, same order
States
LiveFreshness shown
HoverCrosshair, markers, tooltip
StaleLines grey out to --dim2, says how old
Loading, emptySee ChartState
Keyboard and screen reader
Stacked area chart keyboard and screen reader behaviour
Arrow keys
Move the crosshair one point at a time
Screen reader
A one-sentence summary of the trend and the shares
Data
The values are available as a table
Do and don't
✓ DoPut the largest or steadiest series at the bottom. It gets the flat baseline.
✓ DoLabel each band where it ends, and keep the legend.
✕ Don'tA second y-axis. Two scales invent a link the data does not have.
✕ Don'tA fifth series. Fold the tail into "Other".
Agent rules // when an AI builds with it
Stacked area chart agent rules
Allowed on
Controlled and declarative screens.
Must set
A title.
A unit.
A takeaway.
A legend and end labels.
The time of the last update.
Never
More than four series.
Colour as the only key to a series.
A second y-axis.
Where it's used
SpecifiedAgent activity by agent type on the Datum AI console. The readings are illustrative.
Bar chart of success rate by agent over 7 days: Triage-02 98.4%, Records-01 97.6%, Diag-01 96.9%, Triage-05 95.3%, Diag-04 92.7%, Plan-03 88.9%. The target is 95%.
Bar chart of success rate by agent over 7 days: Triage-02 98.4%, Records-01 97.6%, Diag-01 96.9%, Triage-05 95.3%, Diag-04 92.7%, Plan-03 88.9%. The target is 95%.
Anatomy
ALabel. The thing being ranked, in mono.
BBar. From zero, on a faint track.
CValue. At the end of the row, in tabular figures.
DTarget. A dashed line with its label.
EAxis. Zero, the middle and the top, with the unit.
Tokens used
Bar chart tokens
Bar
--cyan
Track
--hl-rgb at 10%
Target
--paper, dashed
Labels and values
--paper
Axis numbers
--dim2
Spec // every number is the number in the code
Bar chart measurements
Bar
--space-100 thick, 1px radius at the data end
Rows
--space-150 apart
Label column
--space-1100
Value column
--space-600
Target
1px dashed
Variants
Ranked, sorted high to low
With a target
Without
States
LiveFreshness, or the period
HoverTooltip with the value and the gap to the target
Loading, emptySee ChartState
Keyboard and screen reader
Bar chart keyboard and screen reader behaviour
Tab
Each row in turn
Screen reader
Hears each name with its value, then the target
Do and don't
✓ DoStart every bar at zero and print its value.
✓ DoKeep the target neutral on a screen with other charts. Amber is spent once.
✕ Don'tA colour per bar when the bars are one series. Bars that start above zero.
Agent rules // when an AI builds with it
Bar chart agent rules
Allowed on
Controlled and declarative screens.
Must set
A unit.
A zero baseline.
A value at each bar.
Never
Start the axis above zero.
A colour per bar when the bars are one series.
Where it's used
SpecifiedTop agents by success rate on the Datum AI console. The readings are illustrative.
A cut axis makes a small gap look large. The printed value keeps a full axis readable.
06.3 // Sankey flow
Sankey flow
Where things go between stages, and how many.
Use forVolume moving through three or four stages, such as intake, AI decision, review and done.
Not forA process with loops or many branches. Use a table of counts.
Night
Task flow // Intake to done
24 h
One in four tasks still passes through a person.
IntakeAI decisionReviewDone
8,986 auto-resolved3,1202,68312,480 in
Auto-resolved8,986
Approved2,683
Rejected437
Failed374
Flow diagram of 12,480 tasks. The AI auto-resolved 8,986, sent 3,120 to human review and failed on 374. Reviewers approved 2,683 and rejected 437. 11,669 tasks were completed.
Paper
Task flow // Intake to done
24 h
One in four tasks still passes through a person.
IntakeAI decisionReviewDone
8,986 auto-resolved3,1202,68312,480 in
Auto-resolved8,986
Approved2,683
Rejected437
Failed374
Flow diagram of 12,480 tasks. The AI auto-resolved 8,986, sent 3,120 to human review and failed on 374. Reviewers approved 2,683 and rejected 437. 11,669 tasks were completed.
Anatomy
AStage labels. One per column, above the drawing.
BNodes. A solid bar per stage outcome, as tall as its count.
CFlows. A tinted band from node to node, in the colour of where it ends up.
DLabels. The count on the widest flows, in --paper.
ELegend. Every outcome with its count, including the thin ones.
Tokens used
Sankey flow tokens
Closed by AI
node --ai, flow --ai at 26%
Through a person
node --paper, flow --paper at 14%
Rejected
node --dim2, flow --dim2 at 30%
Failed
node --alarm, flow --alarm at 40%
Labels on a flow
--paper
Spec // every number is the number in the code
Sankey flow measurements
Node
2.4% of the width
Node gap
--space-150
Plot height
176px
Stage labels
12px mono, --space-200 row
Variants
Four stages
Three stages
States
LiveFreshness, or the period
HoverTooltip with the count and its share of intake
Loading, emptySee ChartState
Keyboard and screen reader
Sankey flow keyboard and screen reader behaviour
Tab
Each flow, in stage order
Screen reader
Hears the flows as sentences: "The AI auto-resolved 8,986, sent 3,120 to review and failed on 374"
Data
The counts are available as a table
Do and don't
✓ DoMake every count add up from stage to stage.
✓ DoPut every outcome in the legend with its count. Thin flows have no room for a label.
✕ Don'tFlows that cross when a different order would avoid it. More than four stages on one card.
Agent rules // when an AI builds with it
Sankey flow agent rules
Allowed on
Controlled and declarative screens.
Must set
Stage labels.
A count for every flow.
Totals that add up.
Never
Flows that cross where order could avoid it.
More than four stages on a card.
Violet for a flow the AI did not close.
Where it's used
SpecifiedTask flow from intake to completion on the Datum AI console. The readings are illustrative.
The flows are tints, under 3:1 by design. The solid nodes and the printed counts carry the reading.
06.3 // Scatter plot
Scatter plot
Do two measures agree?
Use forTwo measures of the same things, such as stated confidence against measured accuracy.
Not forA trend over time. Fewer than about ten points. Use a table.
Night
Calibration // 24 agents
14 days
Planning agents sit about 9 points under the line.
Measured accuracy, %
1007550
Under-confidentOver-confident
5075100
Stated confidence, score
Planning, 5
Other agents, 19
Perfect calibration
Scatter plot of stated confidence against measured accuracy for 24 agents. Nineteen sit close to the calibration line. The five Planning agents sit about 9 points below it.
Paper
Calibration // 24 agents
14 days
Planning agents sit about 9 points under the line.
Measured accuracy, %
1007550
Under-confidentOver-confident
5075100
Stated confidence, score
Planning, 5
Other agents, 19
Perfect calibration
Scatter plot of stated confidence against measured accuracy for 24 agents. Nineteen sit close to the calibration line. The five Planning agents sit about 9 points below it.
Anatomy
APlot. One marker per thing.
BReference line. Dashed, here the line of perfect calibration.
CRegion labels. What each side of the line means.
DAxis titles. Both, with units.
ELegend. Each group by shape and colour, and the line.
Tokens used
Scatter plot tokens
The group in question
--paper, diamond
Everything else
--dim2, dot
Reference line
--dim, dashed
Gridline
--hl-rgb at 12%
Axes
--hl-rgb at 35%
Marker ring
--panel
Spec // every number is the number in the code
Scatter plot measurements
Marker
8px, in a 2px panel ring
Reference line
1px dashed, 5 on, 4 off
Plot height
180px
Axis column
--space-350
Variants
One group picked out
All one group
States
LiveThe period it covers
HoverTooltip with both values and the name
Loading, emptySee ChartState
Keyboard and screen reader
Scatter plot keyboard and screen reader behaviour
Tab
The picked-out group first, then the rest
Screen reader
Hears how many sit on the line and how far the picked-out group sits from it
Data
The values are available as a table
Do and don't
✓ DoChange the shape as well as the colour when a group is picked out.
✓ DoUse the same range on both axes when the line is "equal".
✕ Don'tA trend line the data does not support. More than two groups.
Agent rules // when an AI builds with it
Scatter plot agent rules
Allowed on
Controlled and declarative screens.
Must set
Both axis titles with units.
A legend with shape and colour.
A label for the reference line.
Never
Colour as the only key to a group.
A trend line the data does not support.
Where it's used
SpecifiedCalibration of 24 agents inside the AI insight panel on the Datum AI console. The readings are illustrative.
Empty is not a warning. A chart that looks alarmed when nothing is wrong teaches people to ignore amber.
06.4 // In use
One screen, one amber
The Datum AI operations console puts nine of the chart parts on one screen with the AI parts from System / 05. It is the test of the colour budget: six charts, a review queue and an audit trail, and exactly one reading in amber, the low-confidence task that needs a person. Every limit and target around it stays neutral.
Datum AI consoleAgent operations // Last 24 hours
Static example. Readings are illustrative.
A static, illustrative operations console for AI agents. Four headline numbers: 12,480 tasks processed, most of them auto-resolved by AI, and 37 waiting on review. Below them, activity by agent type over 24 hours, a confidence gauge reading high, a donut of task outcomes, a heatmap of busy hours, a bar chart of agent success rates against a 95% target, and a flow from intake to done. Then a review queue of four tasks, one of them low confidence, an audit trail with undo, and an AI insight that Planning agents sit about 9 points below their stated confidence.
Task, agent, asset
Range
Live // 14:32 IST
Overview // Last 24 hours
Agent operations
24 agents ran 12,480 tasks. 37 are waiting on you.
Tasks processed
12,480
▲Up 8.2%vs prior 24 h
Auto-resolved
72.0 %
▲Up 1.4 ptsvs prior 24 h
Median time to decision
1.8 s
▼Down 0.3 svs prior 24 h
Awaiting review
37 tasks
▲Up 9Oldest 14 min
Agent activity // By agent type
Updated 2 min ago
Volume peaked at 11:00 with 1,045 tasks an hour. Triage carries 44% of the load.
Triage
Diagnosis
Planning
Records
Tasks per hour
1,2009006003000
Wed 11:00 // Tasks per hourRecords116Planning179Diagnosis262Triage488Total1,045
Triage 44%Diagnosis 27%Planning 18%Records 11%
16:0020:0000:0004:0008:0012:00
Hour, IST
Stacked area chart of tasks per hour by agent type over the last 24 hours. Volume fell to about 200 an hour overnight and peaked at 1,045 at 11:00. Triage 44%, Diagnosis 27%, Planning 18%, Records 11%.
Confidence // All agents
2 min ago
Confidence sits in the High band, 6 points clear of the line.
60800100
86 OF 100High confidence
Confidence score
Band lines: Medium from 60, High from 80
Gauge: overall model confidence is high, 86 of 100. Medium starts at 60 and High at 80.
Task outcomes // 24 h
2 min ago
AI closed 72% of tasks without a person.
12,480tasks
Auto-resolved by AI72.0%
Human-approved21.5%
Rejected3.5%
Failed3.0%
Donut chart of 12,480 task outcomes: auto-resolved by AI 72.0%, human-approved 21.5%, rejected 3.5%, failed 3.0%.
Workload // Hour by day
4-wk mean
Load peaks midweek, between 10:00 and 16:00.
MonTueWedThuFriSatSun
00:0006:0012:0018:00
Tasks per hour
Under 150150350550750+
Heatmap of mean tasks per hour by weekday and hour. Busiest cell: Wednesday 12:00 with 990 tasks an hour. Weekends stay under 350.
Bar chart of success rate by agent over 7 days: Triage-02 98.4%, Records-01 97.6%, Diag-01 96.9%, Triage-05 95.3%, Diag-04 92.7%, Plan-03 88.9%. The target is 95%.
Task flow // Intake to done
24 h
One in four tasks still passes through a person.
IntakeAI decisionReviewDone
8,986 auto-resolved3,1202,68312,480 in
Auto-resolved8,986
Approved2,683
Rejected437
Failed374
Flow diagram of 12,480 tasks. The AI auto-resolved 8,986, sent 3,120 to human review and failed on 374. Reviewers approved 2,683 and rejected 437. 11,669 tasks were completed.
Four tasks have waited longest. One is low confidence, so it has no one-tap approval.
Tasks waiting for your review, sorted by time waiting
Proposed action
Confidence
Accuracy, 14 d
Risk
Waiting ▼
Reassign 3 work orders to Line 2TSK-48213 // Plan-01
High confidence 91/100
Up +2.1
R2Risk tier 2: Medium risk
14 min
AI
Based on
Line 1 is at 112% load until 18:00 . Line 2 has two free bays with the same tooling .
Couldn't check
Technician leave for the evening shift
Model
Planner 4.2, updated 2 days ago
If approved
Runs at once. Undo for 24 hours. The entry lands in the audit trail.
Flag ENG-208 for a bearing checkTSK-48209 // Diag-04
Medium confidence 72/100
Down −0.2
R2Risk tier 2: Medium risk
11 min
AIRaised from R1
Defer 12 inspections to nightsTSK-48201 // Plan-03
Low confidence 48/100
Down −4.6
R2Risk tier 2: Medium risk
9 min
AINo one-tap approval
Release ENG-311 to serviceTSK-48196 // Records-01
High confidence 88/100
Up +1.2
R4Risk tier 4: Irreversible
6 min
AINeeds a named person
AIAudit trail // Agent actionsToday
14:18
Rescheduled 6 borescope inspections
Approved by A. Purohit R2Risk tier 2: Medium risk23 h left
13:52
Closed 112 duplicate sensor alerts
No approval needed R1Risk tier 1: Low riskUndone by you at 13:58
11:40
Released ENG-298 to service
Carried out by R. Iyer R4Risk tier 4: IrreversibleCan't be undone
AISuggestionHigh confidence
Planning agents are over-confident
Above a score of 80, the five Planning agents state confidence about 9 points ahead of their measured accuracy . The other 19 agents stay within 5 points of the line .
Calibration // 24 agents
14 days
Measured accuracy, %
1007550
Under-confidentOver-confident
5075100
Stated confidence, score
Planning, 5
Other agents, 19
Perfect calibration
Scatter plot of stated confidence against measured accuracy for 24 agents. Nineteen sit close to the calibration line. The five Planning agents sit about 9 points below it.
Sources2 read
1Calibration run, Planning agents1,140 tasks // 14 days
2Calibration run, all other agents11,340 tasks // 14 days
Suggested next stepR2Risk tier 2: Medium risk
Raise Planning's auto-approve line from 80 to 90
About 40 more Planning tasks a day would come to review. Nothing already approved changes.
Undo for 24 hours
Illustrative data // Not client workBuilt on Datum // Night and Paper
Not done yet, and said plainly: hover is specified but not built, there is no table view behind each chart yet, and none of this has been through a screen reader. The summaries are written; the testing is next.
Thread// Accessibility
Not a checklist. A layer.
Accessibility runs through every sheet, so building without it takes more effort than building with it. Then every release is checked anyway with axe-core, an open-source automated accessibility checker made by Deque. and a keyboard walk, because automated tools only catch part of the problem.
2.2 AAWCAG target, both themes
0Automated issues, desktop and phone, dark and light, normal and high contrast
4.5:1Lowest text contrast anywhere
24pxSmallest tap target
01 Principles
Access is rule five, and it's never traded away.
When rules collide, the order decides, but access isn't on the table.
02 Foundations
Contrast is baked into token pairs, in both rooms.
Text at least 4.5:1, focus and selected states at least 3:1, 12px floor, reflow to 320px.
03 Components
Names, focus and targets come built in.
Visible focus ring, 24px minimum targets, pressed and selected states announced.
04 Patterns
Failures are designed, not apologised for.
Error summaries link to fields, result counts are announced, nothing typed is lost.
05 Intelligence
AI is marked by colour, a spark and the word AI, and doubt is in words.
Confidence never relies on colour, and stop is reachable from the keyboard.
06 Data visualisation
Every series mark clears 3:1, and every hue has a second cue.
Direct labels, a legend shape for AI and failed, printed values and a one-sentence summary for screen readers.
Contrast
Turn up contrast in your system settings and the page follows.
Secondary text, labels, hairlines, field edges and the paper blue get stronger. Surfaces and signals stay as they are.
Motion
A device setting for people who find animation dizzying or distracting. Sites can read it and calm down. is honoured everywhere.
Nothing loops. Short demo animations stop within five seconds, and every final state shows without them.
All 13 WCAG 2.2 criteria, and how they're met
WCAG 2.2 success criteria and how Datum meets them
Criterion
How Datum meets it
Level
1.3.1Info and Relationships
Real headings in order, tables with headers and captions, lists marked as lists.
A
1.4.1Use of Color
Every status has a second cue: shape, border weight or a word. See the side-by-side test in 02.2.
A
1.4.3Contrast (Minimum)
Every text pair at least 4.5:1 in both themes. Body text 17.1:1 at night, 15.8:1 on paper.
AA
1.4.10Reflow
One column down to 320px. Only wide data scrolls, inside its own box.
AA
1.4.11Non-text Contrast
Focus rings and selected states clear 3:1: cyan 10.6:1 at night, blue 6.1:1 on paper.
AA
2.1.1Keyboard
Every control works from the keyboard, including the component tabs, which use arrow keys.
A
2.2.2Pause, Stop, Hide
Nothing loops. Demo animations and the one ambient effect stop within five seconds.
A
2.4.1Bypass Blocks
A skip link jumps straight to the content.
A
2.4.7Focus Visible
A 2px ring with a 4px background-coloured halo, on everything focusable.
AA
2.4.11Focus Not Obscured (Minimum)
Focus and jumps land below the sticky header and chapter bar. Checked by tabbing the whole page.
AA
2.5.8Target Size (Minimum)
Every target is at least 24 by 24px.
AA
4.1.2Name, Role, Value
Toggles announce pressed, tabs announce selected, the menu reports open or closed.
A
2.3.3Animation from Interactions
With reduced motion on, decorative motion is removed and every final state still shows.
AAA bonus
Being honest about itA clean automated audit is the floor, not the finish line. Screen reader testing with VoiceOver and NVDA is next on the bench.
System / 07// Governance
How it stays one system
A system without governance is a style guide with a short life. This is the lightest process that still works, for one designer, a whole team, or an AI agent.
Find
Search the system first. Most needs already have an answer.
Anyone
Propose
A need seen three times, with screenshots and the problem it solves.
Designer
Build together
Design and build use the same token names, so nothing gets renamed in handoff.
Designer and engineer
Review
Design, engineering and accessibility sign off against the pre-flight checklist.
Design, build, access
Release
Versioned and logged. Pages are replaced whole, never patched.
System owner
How it's used feeds the next proposal
On a team, review is three people. On this site it's three passes I make myself: design, build and accessibility, each against the checklist.
Before building something new
Does it already exist?Use it
Is it nearly right?Add a variant
Needed in three or more places?Propose it
Truly a one-off?Build locally, log it
One name, one job
Name // --cyan⇄Job // you can act here⇄Values // #00D4FF night, #0057B8 paper
Pre-flight checklist // nothing ships unticked
Accessibility auditZero issues, both themes, desktop and phone.
Keyboard walkEvery control reachable, focus never hidden.
Six widths, real fontsNo overflow or clipping from 320 to 1440.
This page was built over many AI-assisted sessions, and an AI forgets everything between sessions. Without a written system it guessed, and the guesses drifted. These are real problems from earlier versions of this page:
Name collisionsA new label reused a class from another section and drew a stray box around it.
Leftover rulesOld CSS squeezed the colour panel into one narrow column.
Silent lossThe red and amber styles on state chips disappeared during a clean-up, and nobody noticed for two versions.
Broken floorChart labels shrank to 5 or 6px on phones, under the 12px rule.
68 markdown files every session reads first: principles, foundations, guidelines, then atoms to templates.
Fixes: no memory between sessions
Closed tokens
Three tiers. Datum's raw colours exist in one block only; everything else picks a name. The sealed site header is the one declared exception.
Fixes: invented values
Audit
A script flags every raw value with the token to use instead, and fails on any error.
Fixes: drift inside a session
Sync check
Each page embeds a copy of the tokens; the audit flags any copy that differs from the source.
Fixes: pages falling out of step
Token export
tokens.json is generated from tokens.css in the Design Tokens Community Group format, version 2025.10, so tools that can't read CSS get the same values.
Fixes: tools keeping their own copy
Catalog
catalog.json lists all 44 parts with their props, states, tokens and agent rules. A script checks it against its schema, the specs and the tokens.
When the tokens went in at v9, the page was compared pixel by pixel in both themes: nothing you could see had changed. Still open, as warnings: 224 off-grid spacing values and 92 raw durations. The header is the site's sealed block, copied byte for byte from the home page, so the audit reports it on its own line and leaves it alone: 12 raw colours, 32 raw spacing values and 7 raw text sizes that have to be fixed at the source.
Taste goes in once, as a written rule. After that, every session, human or AI, starts from the same place.
06.2 // Open system
Take it, build on it
Datum is open to anyone with the link: the tokens, the foundations, and all 44 parts with their guidelines, agent rules and a live preview. Read it, hand it to your AI agent, and build your own screens on it.
A system costs more at the start and less on every screen after. The value isn't the library. It's the decisions nobody has to make twice.
EffortBreak-evenScreens built →
One-off designWith a system
Illustrative, not measured. It's the shape every system follows: pay once, save on every screen after.
Reuse
Built once, used everywhere
The header here is the home page's, copied byte for byte. Buttons, footer and focus ring carry the same values.
Consistency
One header, one memory
The same controls in the same place, and one saved theme choice that follows you between pages.
Less duplication
Defined once, by job
Every raw colour Datum owns lives in one block of tokens.css. The audit fails if one appears anywhere else, and counts the sealed header's on its own line.
Speed
No blank pages
This case study started from the Writing page's shell and tokens, and took its header from the home page.
Accessibility
Inherited, not retrofitted
Contrast and focus live in tokens and components, so new pages start from passing values.
Scalability
One source, many outputs
Two themes and six screen widths from one set of decisions.
Evolution
Replaced, not patched
53 versions of the home page, each rebuilt whole on the last.
Shared language
Arguments about problems
Tokens are named by job, so a review can say "action" instead of a hex code.
Flight log
v5
The dark instrument-panel look, built from scratch.
v37
The drafting-paper theme, and a theme switch on every page.
v38
Motion tokens. Every effect answers an action.
v41
Each case study gets its own phone card.
v47
The switch becomes two icons. Connect becomes an outline.
v53
Current reference. One sealed header, one paper palette.
Principles without tests are posters
Once each rule had a check, arguments got shorter. It passes or it doesn't.
Constraints made it more expressive
Rationing colour made the moments that use it land harder. Quiet gives a signal its volume.
Patterns matter more than parts
Teams rarely struggle to draw a button. They struggle with approvals, errors and AI. That's where the system pays back.
Next on the benchScreen reader testing with VoiceOver and NVDA, moving the last 224 off-grid spacing values onto the grid, moving the sealed header onto these tokens at its source, and running the token audit on every page of the site, not just this one.
End of case study
Building where signals matter? Let's connect.
Open to safety-critical products, design systems and AI that people have to trust. Or read why every design system is also a safety system.