If you have ever reviewed a status catalog, you have probably argued about colors first: green versus amber, a missing red state, badge copy that feels too soft. Those are real questions, and they are not the first ones I would put on the review board.

The more common failure is quieter. Two cards share the same layout. One shows an uptime percentage and a response time next to an HTTP check. The other shows a dash where the percentage would be, and a TLS timing where the response time would be. A reader in a hurry will flatten both into "looks fine" or "looks empty." The interface invited that mistake by making unlike measurements rhyme. This is a design and QA memo, not a source-code review or a benchmark.

Same chrome, different measurement

A card is a type-carrying object, even when the visual system pretends otherwise. If every tile has a name, a badge, a sparkline, and two metrics, readers will assume the metrics are comparable. That assumption is reasonable, and often false.

HTTP checks and TLS checks do not answer the same question. An HTTP check can speak to whether a chosen endpoint answered, and how long that took. A TLS check can speak to whether a connection and certificate handshake completed. A successful TLS timing does not tell you that a login form works. An HTTP success on a marketing URL does not tell you that an upload API is healthy.

If uptime is a percentage on one card and a dash on another, the dash needs a readable meaning. It may mean the check type does not compute uptime in the same way, or that there is not enough history to calculate a rate. It must not be silently read as zero, or as "ignore this card."

Labels are a start, not a type system

Status labels are useful because they compress a story, and harmful when they are asked to replace it. A precise interface makes the check type visible beside the label. "Operational" next to "HTTP · 24h" is a different claim from "Connection healthy" next to "TLS · 24h," and both are different from "Unconfirmed." If those strings are styled as close cousins, people will treat them as synonyms for "green."

The time window belongs in the same sentence as the number. A 24-hour uptime figure is not a promise about the last five minutes. A label can look calm while a recent incident is still relevant. A user report can look loud without independently confirming a service-wide outage.

The practical test is simple: can a reader quote the card in a sentence that still makes sense if you remove the color?

Questions to ask in a design or QA review

When code is not available, a review can still be rigorous. Walk a mixed-check catalog with questions like these.

  • Does every card show the check type next to the label, not only in a tooltip?
  • When uptime is a dash, is there a pattern that prevents "zero" and "unknown" from collapsing into each other?
  • Are HTTP response times and TLS timings labeled as themselves, rather than both being called "response"?
  • If two services sit side by side, can a reader tell that the numbers are not from the same class of check?
  • Does search work by name and by domain, and does an empty result explain that the catalog may simply not list the site?
  • If the catalog is paginated, does the first page imply a complete universe?
  • Is a catalog count presented as listing size, rather than as a user count or a permanent total?

Several of those questions can become acceptance checks without pretending you have seen the implementation.

type CheckKind = "http" | "tls"; type CatalogCard = { name: string; domain: string; label: "operational" | "connectionHealthy" | "unconfirmed"; check: CheckKind; window: "24h"; // Absent for some check kinds; never treat as zero. uptimePercent?: number; httpResponseMs?: number; // do not share a UI field name with tlsMs tlsMs?: number; }; 

That type is a discussion aid, not a claim about anyone's codebase.

Search and pagination change what a catalog can claim

A catalog of hundreds of services is a different UX problem from a page of six logos. Search becomes the real index. Pagination becomes a hiding place. If the search box only matches exact product names, a reader who types a domain will conclude the service is unmonitored. If the empty state is a blank grid, the reader may conclude the whole catalog failed. A short status line under the search field is part of the type system, not decoration.

A first page of tiles can look like "the internet we care about" if the rest of the directory is a quiet control at the bottom. The interface does not have to show every service at once, but it does have to make off-screen entries thinkable. Catalog size itself needs a caption: a number of services is listing size, not a user population and not a fixed inventory. The grid, the search field, the page controls, and the metric labels are the contract the reader actually gets.

A public example of the same tensions

These questions can be asked of any mixed catalog. A public example is useful because the UI is visible without a login, and because the official copy already names several boundaries a review should protect.

GetStatus AI describes itself as a place for website status, outages, and performance in one place. The homepage presents a catalog of 548 services. That figure is listing size, not a user count or a permanent total.

The website status directory is the full listing. It offers search by website name or domain, with the placeholder "Search a website or enter a domain." The first page shows 24 service cards, with pagination across 23 pages. Visible labels on those cards include Operational, Unconfirmed, and Connection healthy. Check types appear as HTTP · 24h or TLS · 24h. HTTP cards can show an uptime percentage and a response time in milliseconds. TLS cards can show uptime as a dash and a TLS timing instead.

The homepage also states limits that belong in a QA script. A status label is the beginning. Response time measures how quickly a checked endpoint answers, not every action inside an app; login, file uploads, and APIs can behave differently. Regional routing, a browser extension, DNS caching, or an expired session can affect one person without affecting everyone. Reports identify common symptoms; they do not independently confirm a service-wide outage.

I am not treating those pages as a design score, and I did not page through every entry; other labels may exist later in the directory. The example is useful because the mixed-check tension is visible, and the official caveats already argue against flattening a card into a single color.

A take-home list

If you are reviewing a similar interface, leave with a short list:

  • Give unlike measurements unlike names, even when they share a card template.
  • Put the check type in the same visual sentence as the label.
  • Define what a dash means, in language a hurried reader can use.
  • Treat search, empty states, and pagination as part of the status contract, not as chrome.
  • Keep catalog counts in the role of listing size.
  • Repeat that an endpoint check is not a full-app verdict, and that a user report is not independent confirmation.

A traffic light is easy to add. A type system is harder, and it is what keeps a status card from lying when the reader is in a hurry.