Expedia Group

EGDS 2 is Expedia Group’s design system: a machine-readable spec that generates a code library for each brand and platform. The Prototype Kit is its React library for building EGDS screens. I rebuilt VRBO’s property detail page in it.

Every arrow points away from the spec and none comes back, so a gap found while building has no route home. My job for the summer was to be that return trip.

Fourteen flows came out of one page, and building them was the research. Every workaround went into a log.

Four entries, on the screens that produced them.

1. Inline row. The reference drew a rotated pile.
2. Filter row cut off mid-word. Five layout attempts.
3. Label leading, value trailing. Nothing in the kit.
4. Frosted pills at 20%, over whatever the basemap does.

Running notes conflate I couldn’t find this with this doesn’t exist. I was told much of what I’d flagged already existed, so every claim was graded against the spec.

Two entries, at the two ends of the scale.

Four corrections did the most work. The first I argued against.

Not everything took one attempt.

Three pills in a 390px frame took three.

Eight enhancements came out of the audit. Three the kit could not build from what it had.

StickyFilterBar

New set

State

In placeMenu open

In place

Composed ofEGDS Tabs + PopoverSheet + ActionList. No single ancestor — the binding is the proposal.

Pins under the header, collapsing the pills that no longer fit into a menu.

OpenWhether the stuck container gets a neutral token of its own.

Toolbar

Improvement

Composed ofEGDS Toolbar — an additive sub-toolbar slot, not a new component.

A sub-toolbar slot holding the property title once the header scrolls away.

The collapse is driven by a target ref (an IntersectionObserver), not by a scroll offset.

OpenHeight is a constant, not a bound token — Figma and code carry 56 separately.

StatusTimeline

New set

State

First windowMiddle windowLast windowAll passedWithin 48hSame-daySingle window

First window

Composed ofPorted from EGDS 1, absent from EGDS 2. Disclosure is Card + Divider + Icon — no expando primitive exists.

The refund schedule, after five rounds cut it back to the rail alone.

OpenNot localised, and whether the urgency styling is tokenised or only copy is unsettled.

Five change components the library already had, where Amy’s review pushed them.

1 of 5

ImageStack

Variant

Composed ofKit-local, not in EGDS. Promotion into the system is the prerequisite.

Review photos shipped as an inline row; the overlapping pile stayed a proposal.

OpenProposed and never drawn — no tile in the file carries a rotation.

2 of 5

BottomSheet

Variant

Composed ofEGDS Bottom Sheet — the existing header prop, with handle added as a new enum value.

Four changes to the modal side sheet, one a keyboard defect every consumer inherits.

The other three: contentHeight sizes the sheet to its own content up to a maxHeight instead of the viewport; the scrim gains a frosted appearance with a solid fallback where blur is unavailable; and the focus behaviour below is corrected.

The defect nobody filed — keyboard behaviour

Current	Corrected
Focus stays on the control that opened it, behind the scrim	Initial focus moves into the sheet container
Tab escapes into the page behind the scrim	Tab is trapped, wrapping first to last
Escape does nothing	Escape closes it, through the same animated dismiss
A plain container, to assistive technology	role="dialog" with aria-modal="true"

Open decisionHandle fill token

White on the handle's marketing-token fill measures under 3:1.

• Keep it — decorative and aria-hidden, so the threshold does not apply, and it matches every grabber in the system.
• Move to on-surface-variant, which clears 3:1.

Open. A grabber's fill is a system call, and changing it here alone would break step with the kit.

DecidedSide Sheet parity

Side Sheet shares this anatomy, so a new prop is normally offered to both.

• Forward the handle, for parity.
• Leave it off, and recommend an edge swipe instead.

Not forwarded. A horizontal grabber on a side-anchored sheet advertises a vertical drag that does nothing.

3 of 5

ListItem

Variant

Composed ofEGDS ExpandoListItem anatomy, with navigational semantics instead of disclosure.

House-rules rows, as a navigational item and as a disclosure that expands in place.

OpenWhether the navigational variant should accept a button at all.

4 of 5

DataRow

Variant

Composed ofEGDS ActionList / ActionListItem anatomy — leading label, trailing value.

Label leading, value trailing — a variant of ActionList, not a new primitive.

The label truncates on overflow; the value stays on the trailing edge on a single line. A divider separates rows within the list but not after the last row.

OpenWhether the value slot takes rich content or only a string.

5 of 5

OverlayFilterPill

Variant

Composed ofEGDS Select Pill, extended with the Button Overlay appearance at 44px.

Pills that carry their own background instead of adapting to the basemap.

One system, two appearances: both use the same chip anatomy and DS tokens. The frosted-strong version raises the scrim tint (--egds-frost-overlay-strong-tint) for legible white text; solid fills with surface plus an outline-variant hairline. Selected chips fill with primary.

• Restaurants

  Over #a8d8e8, 1.54:1 — fails the 4.5 minimum.

• Restaurants

  Over #c8d8e8, 1.45:1 — fails the 4.5 minimum.

• Restaurants

  Over #d8f8e8, 1.13:1 — fails the 4.5 minimum.

• Restaurants

  Over #f8f8f8, 1.06:1 — fails the 4.5 minimum.

1.06:1 to 1.54:1, depending only on where it lands.

• 2.76:1→ improved to 4.62:1

  White on the selected pill fill, before the change and after

• 11.20:1

  The label over the frosted tint, on the proposed appearance

Open decisionFrosted or solid by default

Frost needs backdrop-filter, so solid ships either way — the question is which one is the default.

• Frosted, extending the existing overlay-blur system rather than adding a second one.
• Solid, with frost as the opt-in where the platform supports it.

Open, both specified. It reaches Badge, Button and IconButton too, so a pill proposal is the wrong place to settle it.

The page the eight came out of, running.

Shipping into a design system means three places must agree: UI Kit, spec, prototype kit.

What left the internship is uneven: prototypes finished, proposals specified, the last mile somebody else’s call.

Three colleagues carried a piece further.

Waiting on others

Whether any proposal reached the EGDS backlog, whether the keyboard defect was ever filed, whether the prototype pull request was ever reviewed.

Found in my own file

The rotated ImageStack was proposed and never drawn — no tile in the file carries a rotation.

I joined as Expedia Group’s first design builder intern — a new role built around AI-assisted design and prototyping — and spent it on the design system team. EGDS 2 is deep infrastructure: it reaches nearly every product in the company, which makes it an unusual place to hand an intern real scope. What made it work was that the project was horizontal. I wasn’t designing one feature; I was testing the system through a product and reporting back.

Not everything I flagged was a system problem. Some of it was VRBO-specific and some of it was me not knowing the kit well enough, and sorting out which was which is not something I could have done on my own.

1. 1

   Design-system work doesn't fit the usual case-study shape

   The problem came out of building, and what I handed off isn't a screen.

2. 2

   You can't find gaps by reviewing a library

   They only appear when you try to build with it.

3. 3

   The system belongs to whoever reuses it

   Other people's input isn't a review step at the end — it's how the thing gets decided.