Platform and Device Support Matrix
beta · Family standing-standards · Phase undefined · Sizes lean · ~7,350 tokens
The standing page that states which browsers, operating systems and devices a product supports, at what level, and which of them the team tests on. It keeps two promises apart, what a user on a given browser gets and what the team actually tests, and it is consulted when a team picks test browsers, triages a bug from an unusual browser, or decides whether a polyfill is owed.
The short card. The reasoning behind every rule here, with the sources, is in
platform-device-support-matrix_companion.md. A fully
worked instance is platform-device-support-matrix_example.md.
Where this card calls something the library’s position, no source read for the bundle publishes it, and
the card says so each time.
When to use
Section titled “When to use”- A team has to choose which browsers to test, and you want a standing written answer rather than a fresh decision for each release. The GDS Way and the ONS both treat the browser list as what a team consults for that choice.
- Someone triages a bug from a browser the team does not list, and you want the rule for what that report is worth written down. The GDS Way, Jenkins and GitHub each state such a rule.
- An engineer decides what a polyfill or a code baseline must cover, and needs the matrix to decide against. GitHub’s browser-support library ties its baseline to its matrix.
- You need to say what a user on a given browser gets, and separately what the team tests. The sources use the word “support” for both, so a matrix that does not define the word inherits the ambiguity.
- Some audiences need less than the rest, such as an internal tool used by staff on one managed browser. The Service Manual lets evidence make a case for an exception, and the ONS allows an evidence-based exemption for internal services. The Purpose and Scope section is where that exemption is stated.
- You want the page to name who may change a grade and what event makes it wrong. The Service Manual says why the page must be standing: “Decisions about compatibility cannot be something you specify at the start of your project and then forget about.”
When NOT to use
Section titled “When NOT to use”- You are recording the browsers one release was tested on. That belongs in a test plan’s environment section. This document is standing, and it states which environments get which promise. The boundary is the family contract’s call, not a sourced finding.
- You are writing one test case’s environment. A test case states its own environment and
configuration. The library’s
test-casetemplate already opens that section with “device or browser matrix”, and its TRAP warns against listing one nobody will run. That the matrix is the shared reference a test case points to is the library’s inference, at medium confidence. - You are reporting one bug. A bug report records the one browser, device and operating system where
the bug was seen, and the
bug-reporttemplate has no field for whether that configuration is supported. The triager compares the report against this matrix, which is the same inference at medium confidence. - You need a table of assistive-technology combinations. GOV.UK keeps its table on a page of its own. The matrix should say where those combinations live and hold none of them. That is the library’s reading, at medium confidence.
- You want a rule about JavaScript switched off, a slow connection or a screen width. The sources place each elsewhere: GOV.UK makes JavaScript-off a build rule, treats bandwidth as a performance budget, and the ONS and USWDS configure breakpoints as layout and not as a support commitment. None is a matrix row.
- You want to decide between version lists and feature detection. MDN and Jeremy Keith argue for capability tests, and the Service Manual keeps a browser list while also saying to use progressive enhancement. No source read argues that a matrix is the wrong tool, so this card does not either. A matrix states commitments, and a capability test is one way of meeting them. That reading is the library’s.
Pick a variant
Section titled “Pick a variant”This bundle ships one size, lean, so there is no second variant to choose. The research looked for the
evidence a heavier size would need: per-row evidence columns such as a usage share or a date last
verified, and a separate polyfill policy. The pages read carry neither. The single size stands at medium
confidence. A named matrix that publishes per-row evidence columns, or a polyfill policy kept apart from
its support grades, would be the evidence for a second size. A feeling that the page needs more is not.
See platform-device-support-matrix_companion.md section 4.
Three choices remain inside the one size. Settle them in this order.
- How many grades. The sources disagree. Yahoo has three grades, GOV.UK Frontend four, Jenkins three levels, and Springer Nature and the AusDTO UI-Kit two. Grades are also optional. Material Components Web has none, and defines its support as the set it tests. A team whose whole promise is the latest browsers, tested, can run one grade, and the cost is that the page has nowhere to say a browser is supported but untested. This card’s advice is to write one grade for each different set of three answers: what the user gets, what the team tests, and what the team does about a bug. Merge two grades whose three answers match. That advice is not a sourced rule.
- A minimum version or a version rule. A minimum names one version, and the UI-Kit gives a reason for its minimum: “Minimum version based on support for CSS Flexible Box layout modules”. A rule states a window. MetaMask states “The last two years of Chrome stable releases”, and the UI-Kit states “All browsers listed are latest stable release, except Internet Explorer.” Fill the mobile table first and give every mobile row one or the other. That ordering is the library’s advice.
- Which figure brings a browser in. Every figure in the sources is one team’s choice for its own traffic, so choose yours and name its analytics source. See the Threshold Rule section.
Quality rubric (self-grade)
Section titled “Quality rubric (self-grade)”Score each row 0, 1 or 2. Under 15 out of 22 and a tester, a triager or an engineer reading the matrix still cannot tell from the page which browsers get which promise, or whom to ask when a row looks wrong. Grade what the author added to the blank template. The template already prints the column headings and the five Review Trigger conditions, so a row that rewards their presence is satisfied before anyone writes a word.
| # | Criterion | 0 | 1 | 2 |
|---|---|---|---|---|
| 1 | Scope is bounded | No product is named, or the scope reads “all modern browsers” with no record and no reader | The product and the place the record lives are named, but nothing says which audiences are exempt, and nothing says that none is | You can point at the sentence naming the product, the record and who reads it, and at either the audience it exempts with the evidence for the exemption, or a line saying no audience is exempt and why |
| 2 | Grades are commitments | A grade is a letter or an adjective, with no test commitment and no bug commitment | Each grade says what a user gets, but at least one grade omits what the team tests or what it does about a bug | For every grade you can point at what the user gets, what the team tests and what it does about a bug, and a stranger could say what the lowest grade promises without asking anyone |
| 3 | Promise and test apart | One column holds both the promise and the testing, or “supported” is used with no definition | Grade and Tested are filled in, but Tested repeats what the grade says the team should do rather than what it does today | The person who runs the pre-release checks would confirm each Tested value, and wherever Tested falls short of its grade’s test commitment, or a row is undecided, Notes gives the reason |
| 4 | Versions and date | Mobile rows carry no minimum version or rule, and the effective-from date is blank or a placeholder | The mobile rows are versioned but the date is missing, or the date is present and the owner cannot say what it records | A tester can read any mobile row and say which version to install without asking, and the effective-from date is one the owner can explain, such as the review-history line that set it |
| 5 | Threshold has a source | A bare figure, or a figure borrowed from another team’s rule | An analytics source and a decider are named, but nothing says why the figure fits this product’s traffic | You can point at the analytics source, the person who reads it, the person who decides, and the sentence saying why the figure is this product’s own, and a reader could open that report and check the figure |
| 6 | Threshold runs both ways | Only the way a browser enters the matrix is written | An exit rule exists, but it names no figure or event, or the Ownership section names nobody who is told when a browser falls below the line | You can point at the exit rule, at the figure or event that fires it, at what happens to the row, and at the Ownership section naming who is told |
| 7 | Unsupported has an answer | “Other browsers are not supported”, and nothing more | The page says what a user on an unlisted browser gets, or what happens to a report from one, but not both, or it declines a fix without naming grounds | A support agent could say aloud what a person on an unlisted browser gets, the page says what the team does with a report from one, and each ground for declining a fix is written so that a reply can say which ground applied |
| 8 | Owner is reachable | “The team”, or no owner | A role is named without a person, or a person without the place where changes are discussed | You can point at the role, the person in it, the place where changes are discussed, who may change a grade or the threshold, and who is told, and you could message the owner today |
| 9 | Triggers carry owners | The Review Trigger table is blank, or says “review regularly” | Some conditions have a named role and an action, but at least one row has neither, or the annual floor is the only row that does | Every condition has a role as owner and an action, the annual floor sits last and says what the owner does, and the review history holds a dated line for each firing or run, including “no change”, or says the matrix is new |
| 10 | Notice is planned | A raised minimum has no action beside it | An action says to tell users and names no audience, place or timing | The action beside a raised minimum names who is told, whether end users or developers of a library, where they are told, and that it happens before the minimum rises |
| 11 | Assistive technology by reference | The matrix says nothing about assistive technology, or it holds combinations inside the browser table | The scope section names a location for the combinations, but the location does not exist or lists none | The scope section names where the combinations live, you can follow the pointer and find them, and the browser tables hold none |
The test behind every cell above: could someone satisfy it without improving the document? A row that counted browsers, grades or conditions would reward padding the table. Each cell asks instead whether a specific fact is on the page that a second reader, not the author, could point at. Grade with someone who did not write the matrix. The blank template should score 0 on every row, and a new, honest matrix can score 2 on every row without inventing a demotion or a fired trigger.
Named anti-patterns (the usual wrecks)
Section titled “Named anti-patterns (the usual wrecks)”- A grade defined by its letter. The same letter means different things in different schemes, so a letter alone promises nothing. The Support Grades section and companion section 2 fix it.
- One column, two promises. “Support” can mean what the user gets or what the team tests, and one status column cannot say both. Keep Grade and Tested apart, as The Matrix section does. The separation is the library’s position.
- A borrowed figure presented as a standard. Each figure in the sources is one team’s choice for its own traffic. State your own figure and analytics source, as the Threshold Rule section asks.
- Listing a browser nobody tests. A row marked Tested that the team does not run is a false claim. The Tested column and the Notes column in The Matrix section are where the truth goes.
- Every row simply supported. A matrix with no demoted, untested or undecided row shows no judgment. This is the library’s standard for its own example and not a finding from the sources. Companion section 7 explains.
- Writing only the entering side of the threshold. A rule that never says when a browser leaves lets the matrix grow without a decision. The Exit rule field in the Threshold Rule section fixes it.
- A flat refusal. “Not supported” skips the investigation that triage needs first. The Unsupported Browsers section names the grounds for declining a fix, and companion section 3 gives the sources.
- Silent drift, or the annual review as the whole mechanism. A matrix with no owner and no trigger is the ungated case the family contract names. The list changed when usage moved, and periodic reviews checked whether it had. Reading the annual figure as a floor is the library’s inference, at medium confidence. The Review Trigger section holds the conditions, and companion section 3 holds the evidence.
- Raising a minimum with no notice. A Review Trigger row that names no notice action leaves that step to chance, and the audience, end users or developers, has to be named. The Review Trigger section’s action column is the place, and companion section 3 gives the sources.
- Scope creep into other dimensions. JavaScript switched off, a slow connection, a screen width and assistive-technology combinations are not rows in this matrix. The matrix points at where they live.
- Presenting the library’s positions as received practice. The separate Grade and Tested columns have support in the sources all the same. The Notes column has only partial support. The role names in Ownership are a proposal, because no source names a role that may change a grade. The table form of the Review Trigger is also the library’s own. A filled matrix may keep them, and it should not describe them as a standard.
Pairing with your process
Section titled “Pairing with your process”This bundle ships in the standing-standards family as a foundation: a standard that a team’s test
choices and bug triage are judged against, rather than an instrument executed under time pressure. The
reasoning is in
ADR 0088 (the platform and device support matrix joins standing-standards as a foundation).
Three documents in this library point at a matrix that this bundle supplies: the test case, the test
plan and the bug report. That is the library’s inference, at medium confidence. The prd template’s
Non-functional requirements section is the nearest place that precedes the matrix, and the inference
that the matrix gives that section a reference to name is at low confidence.
Update the matrix when a Review Trigger condition fires, not on a calendar. Treat the annual figure as the longest the conditions can go unchecked, and not as a calendar rewrite of the document. That reading is the library’s inference at medium confidence. The GOV.UK history shows the list changing when usage moved and periodic reviews checking whether it had. The quoted review-only entries date from 2020 and 2021, and the annual commitment dates from 2024.
What the research did not settle
Section titled “What the research did not settle”- Whether keeping a matrix helps. No source read shows that a matrix improves outcomes, and none measures what keeping one costs.
- Whether a graded list is overhead for a small team. No source read argues it either way.
- Regulated settings and a solo maintainer. The research did not examine either, and this card makes no claim about them.
- How often a team should review. The quoted GOV.UK review-only entries are about six months apart, and its annual commitment dates from 2024. Treat the cadence of any other team as unknown.
- The role names in Ownership. No source read names a person or role who may change a grade, so the role names are a proposal and not received practice.
The artifacts
Section titled “The artifacts”platform-device-support-matrix_template-lean.md · ~7,350 tokens
---title: "{{product_name}} Platform and Device Support Matrix"doc_type: platform-device-support-matrixsize: leanproduct_name: "{{product_name}}"owner: "{{owner}}"status: draftdoc_version: "{{doc_version}}"created: "{{date}}"updated: "{{date}}"related_links: []source_template: platform-device-support-matrixsource_template_version: 0.1.0---
<!--LEAN PLATFORM AND DEVICE SUPPORT MATRIX. This bundle ships one size, deliberately. The research looked forthe evidence a second, heavier weight would need: per-row evidence columns, such as a usage share or a datelast verified, and a separate polyfill or progressive-enhancement policy section. The pages it read carryneither. The nearest things are a principles list and a dated history kept for the whole page, and theReview Trigger section below already holds that history. A longer variant would model the weight theresearch rejected, so the document stays short on purpose. Confidence in the single size is medium. Seeplatform-device-support-matrix_companion.md section 4 (Variants and sizing).
A MATRIX MAKES TWO PROMISES, AND THIS TEMPLATE KEEPS THEM APART.One is what a user on a given browser gets. The other is what the team tests. The sources use the word"support" for both: Material Components Web defines its supported set as the set it tests, MetaMasksupports browser forks it does not test, and the GOV.UK Service Manual's own list is a list of browsers"to test in". A matrix that does not define its words inherits that ambiguity. So the grades below aredefined by their commitments, and testing is marked in a column of its own. That the senses differ is wellsupported. That a matrix should therefore mark testing separately is this library's position. Seeplatform-device-support-matrix_companion.md section 6 (Debates and contested boundaries).
SOME OF THIS STRUCTURE IS THIS LIBRARY'S OWN POSITION, NOT RECEIVED PRACTICE.No source read for this bundle has separate Grade and Tested columns or a Notes column. None names a personor a role that may change a grade, and none sets its review conditions out as a table. Each of these islabelled where it appears. Nothing here is a standard. Every threshold figure in the sources is one team'schoice for its own traffic. The evidence base is public matrices from government, design-system andopen-source teams plus five commercial pages, and no source shows that keeping a matrix improves outcomesor measures what keeping one costs.
WHAT THIS IS NOTIt is NOT the list of browsers one release was tested on. That belongs in a test plan's environmentsection. This document is standing: it states which environments get which promise, and it is consultedevery time a team picks browsers to test or triages a bug from an unusual one. That boundary is thefamily contract's call, not a sourced finding.It is NOT an assistive-technology table. Those combinations live on a page of their own and this documentpoints at them.It is NOT a rule about JavaScript being switched off, a slow connection or a screen width. The sourcestreat each as something other than a matrix row. See platform-device-support-matrix_companion.md section 3(Anatomy).
All names and figures in the GOOD and WEAK illustrations below are invented.
HOW TO FILL THIS IN1. Read the comment under each heading: WHAT it wants, WHY it matters (with a pointer into platform-device-support-matrix_companion.md for the deep reasoning), guiding questions to ASK, a GOOD and a WEAK example, and the TRAP to avoid. The Support Grades, The Matrix and Review Trigger tables also carry PRIORITY and ROW HINT.2. Replace each {{placeholder}} with your content.3. If a section, or one row inside a table, does not apply, write "N/A" and one line of why, rather than leaving it blank or deleting the section. An unexplained blank and a skipped item look identical to a later reader.4. This is a standing document, revisited on the Review Trigger below, not written once and forgotten. Before you first share it: self-grade against platform-device-support-matrix_guide.md, then DELETE every HTML comment. They are guidance, not content. Keep the source_template and source_template_version frontmatter fields: they are the only record of which template and version produced this document.-->
# {{product_name}} Platform and Device Support Matrix
## Purpose and Scope
<!-- WHAT Which product this matrix covers, who relies on it, whether this file or a tool is the record, which audiences are exempt, and where the assistive-technology combinations live. WHY It says which product the promises apply to. The AusDTO UI-Kit file opens with its purpose: "The UI-Kit team needs to support all of our users, regardless of their device, web browser or other user agent." The GDS Way adds a reason to scope by product: "The exact set of browsers you build for will vary by project". The ONS names the floor under any list: "Services do not have to look perfect in every browser, but users must be able to access and use all the information and features they need." Audience exceptions have two sources. The Service Manual lets evidence change the requirement: "If evidence shows that your users have specific needs or extra support requirements, you can use this to make a case for not meeting these browser requirements." The ONS allows an evidence-based exemption for a class of services: "internal services used exclusively by ONS staff may not need to work on all supported browsers". Assistive technology goes by reference, not inside the matrix: GOV.UK publishes its table of combinations on a page of its own, saying "your service must work with at least the following combinations of assistive technologies and browsers before it goes into public beta". That reading is this library's, at medium confidence. The form of the scope statement is this library's position; no source publishes one. The Service Manual says why the page is standing: "Decisions about compatibility cannot be something you specify at the start of your project and then forget about." Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Purpose and Scope). ASK Which product does this cover, and who reads it? Is this file the record or a tool, and where do the assistive-technology combinations live? Which audiences are exempt, and what evidence says so? GOOD "This matrix covers the Fernhill Library public catalogue site. It does not cover the staff cataloguing tool, an internal service used only by library staff on managed laptops, which is exempt because every staff machine runs one browser. The record is this file in the site repository. Testers and whoever triages bug reports both read it. Assistive-technology combinations are kept in the accessibility test guide, linked below, not here." WEAK "We support all modern browsers." (names no product, no reader and no record, exempts no audience, and says nothing about assistive technology) TRAP Letting the scope grow past browsers and devices. JavaScript switched off is a build rule in the sources, bandwidth is a performance budget and a screen width is layout configuration. None is a matrix row. Folding assistive-technology combinations into the browser table is the same mistake, and it is this library's position that the matrix should point to them instead. -->
**Product covered:** {{product_covered}}
**Who relies on this matrix:** {{matrix_readers}}
**Where the record lives:** {{record_location}}
**Audiences exempt, and the evidence:** {{audience_exceptions}}
**Assistive-technology combinations live at:** {{assistive_technology_location}}
## Support Grades
<!-- WHAT Each grade this matrix uses, with what a user on it gets, what the team tests, and what the team does about a bug. This section defines the words The Matrix then uses. WHY The sources define grades by what the team commits to do, and the commitments differ. GOV.UK Frontend commits to priority: "treat bugs affecting Grade A browsers as high priority", and "treat bugs affecting Grade B browsers as low priority unless we find evidence they prevent a user from being able to complete their task". For its lowest grade it commits to neither testing nor fixing: "will not regularly test in these browsers" and "We will not fix bugs affecting these browsers." Yahoo's older scheme states its lowest grade the same plain way: "QA does not test, and bugs are not opened against X-grade browsers." The letters do not carry across schemes. In Yahoo's, "C-grade is the base level of support, providing core content and functionality." In GOV.UK's, C is a reduced grade and X gets no fixes. A grade defined only by its letter therefore promises nothing a reader can use. The number of grades is open: Yahoo has three, GOV.UK Frontend four, Jenkins three levels, and Springer Nature and the AusDTO UI-Kit two. Grades are also optional. Material Components Web has none: "We officially support the last two versions of every major browser. Specifically, we test on the following browsers:" The cost of one grade is that the page has nowhere to say a browser is supported but untested. Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Support Grades). ASK For each grade: what does a user on it get? What will the team test on it, and what will it not test? What does the team do about a bug on it: fix with priority, fix only if it blocks a task, or not fix? Could a stranger read the row and say what the team promised? PRIORITY List grades from the strongest promise to the weakest, and write the lowest grade's bug commitment before anything else. Both GOV.UK and Yahoo state it for their lowest grade in plain words. The ordering is this library's advice, not a sourced rule. ROW HINT A good row names a grade, says what the user gets, says what the team tests, and says what it does about a bug, each as something the team actually does. A weak row is a letter and an adjective. GOOD | Supported | The full catalogue, including saved searches and reading lists | Checked by hand on a real device whenever the search or account pages change | Fixed ahead of other work | WEAK | Grade A | Full support | | | (a letter and a phrase; no test commitment and no bug commitment) TRAP Defining a grade by its letter, or putting two promises in one cell. "Supported" that means "we promise it works" in one row and "we test it" in another leaves the reader unable to tell what the team tests. This column set keeps the user's experience, the testing and the bug response apart. -->
| Grade | What a user gets | What the team tests | What the team does about a bug ||---|---|---|---|| {{grade_name}} | {{grade_user_gets}} | {{grade_test_commitment}} | {{grade_bug_commitment}} |
## The Matrix
<!-- WHAT A mobile table and a desktop table, each row a browser, with an effective-from date. The columns are Browser, Platforms, a minimum version or a version rule, Grade, Tested, and Notes. WHY It is the table itself: the browsers, platforms, versions, grades and test status that the other sections define and govern. THE COLUMN SET IS PARTLY THIS LIBRARY'S POSITION. The AusDTO UI-Kit's mobile header row is "Browser Platforms Minimum version Advanced support status". Its desktop table has no minimum version column, and its single status column holds both the level and the test status. No source read has separate Grade and Tested columns, and none has a Notes column. The separation still has support. The GDS Way separates what a team builds for from what it tests, and MetaMask shows a row one status column cannot express: "We try to support browser forks as well (aligned with supported Chrome/Firefox releases), but we don't actively test them." A version rule can replace a minimum version. MetaMask states "The last two years of Chrome stable releases", and the UI-Kit states "All browsers listed are latest stable release, except Internet Explorer." That reading of the column is this library's synthesis, at medium confidence. The Service Manual dates its table: "Browsers to test in (from February 2026)". This library borrows the dated line from that one source, at medium confidence. A row may also be left open on purpose: Jenkins says "Support for mobile browsers (e.g. iOS Safari) has not yet been determined." No source read puts native-app support and web support in one table, and the GOV.UK app's terms publish "iPhones running iOS 16 or later" as a minimum operating system. Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > The Matrix). ASK Which browsers and platforms does the team promise something to, and what is the minimum version or rule for each? Which grade from Support Grades does each get? Is each row actually tested today, and if its Tested value falls short of what its grade promises, why? Which row is undecided, and does the table say so? PRIORITY Fill the mobile table first and give every mobile row a minimum version or a version rule. Mark Tested from what the team does today, not from what its grade says it should do. ROW HINT A good row names the browser and platform, gives a minimum version or a rule, names a grade defined above, marks Tested or Untested, and uses Notes to give the reason whenever Tested falls short of the grade's test commitment or the row is undecided. A weak row is a browser name with the other cells blank or all saying "supported". GOOD | Brave | Windows, macOS | Same rule as Chrome | Supported | Untested | Chromium-based. We fix reported bugs, but nobody runs it before a release. | WEAK | Chrome | All | Latest | Supported | Yes | | (every cell filled and no decision visible) TRAP Listing a browser nobody tests, or marking every row simply supported. A matrix with no demoted, untested or undecided row shows no judgment; that is this library's standard, not a finding from the sources. The library's own test-case template makes the related warning, against listing a matrix nobody will run. -->
**Effective from:** {{effective_from_date}}
**Mobile browsers**
| Browser | Platforms | Minimum version or version rule | Grade | Tested | Notes ||---|---|---|---|---|---|| {{browser_name}} | {{platforms}} | {{minimum_version_or_rule}} | {{grade_name}} | {{tested_status}} | {{row_notes}} |
**Desktop browsers**
| Browser | Platforms | Minimum version or version rule | Grade | Tested | Notes ||---|---|---|---|---|---|| {{browser_name}} | {{platforms}} | {{minimum_version_or_rule}} | {{grade_name}} | {{tested_status}} | {{row_notes}} |
## Threshold Rule
<!-- WHAT The rule by which a browser enters or leaves the matrix: the usage figure, the analytics source it is read from, who decides, and what happens in both directions. WHY It gives a change to the list a stated reason. Each figure below is one team's choice for its own traffic. USWDS: "We officially support any browser above 2% usage as observed by" an analytics service. The ONS: "if your service’s analytics data shows at least 2% of your users are using a different browser or version, then you will need to test against that as well." GOV.UK does not state a floor. It states coverage: "The list in the table represents approximately 98% of the most popular browsers used on GOV.UK." Its change entries cite a percentile: "Removed requirement to test with Safari 11 and earlier versions, as they fall outside of the 95% most popular browsers." The GDS Way leaves the choice to the team: "You should use analytics data and user research to determine whether you should regularly test in any other browsers." Deque's 1% is a rule for assistive-technology combinations, "be used by more than 1% of users", not for browsers. These figures are not norms. Browserslist warns that a usage floor used alone "will in the long run make popular browsers even more popular", and offers a query over "statistics derived from your real-world traffic rather than global marketshare". The Service Manual points at the other end of the decision: "You also have to work out when to stop supporting a product or system if its popularity is declining." It is this library's position that a matrix should name its own analytics source beside any figure. Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Threshold Rule) and section 6 (Debates and contested boundaries). ASK What usage figure brings a browser in, what takes one out, and what does the team do when one crosses the line in either direction? Which analytics source is it read from, and who reads it? Who decides? Why is this figure ours rather than someone else's? GOOD "A browser enters the matrix when the library's own web analytics show it above 4 percent of sessions, and its row is demoted when it falls below 1 percent. Both figures are our own choice for our own traffic and are illustrative here. The web officer reads the analytics report and decides. The head of digital services is told of every demotion." WEAK "We support any browser above 2%." (a figure borrowed from another team's rule, with no analytics source, no decider, and no word on what happens when a browser falls below it) TRAP Presenting a borrowed figure as a standard, or writing only the entering side. A threshold with no named analytics source cannot be checked, and a rule that never says when a browser leaves lets the matrix grow without anyone deciding it should. -->
**Entry figure:** {{entry_threshold}}
**Exit rule:** {{exit_rule}}
**Analytics source, and who reads it:** {{analytics_source}}
**Who decides:** {{threshold_decider}}
**Why this figure is ours:** {{threshold_rationale}}
## Unsupported Browsers
<!-- WHAT What a user on an unlisted or lowest-grade browser gets, and what a bug report from one is worth. WHY The matrix lists what the team supports, so the people who are not on it need an answer too. The sources carry that answer under three names. The AusDTO UI-Kit has a section headed "Unsupported browsers" whose body says "Other combinations may be supported but have not been extensively tested". GOV.UK Frontend carries it as its lowest grade. Jenkins makes its lowest level the default: "This is the default unless a browser/version is listed". The GDS Way sets the posture for a report from outside the policy: "You may receive reported issues on browsers outside of your normal policy. You should still investigate these issues, assess their impact, and consider how to address them." It names grounds for declining a fix after that: "fixing the issue would require disproportionate engineering effort", "a feature of your service inherently relies on an unsupported feature", and "the browser in question cannot be supported without introducing serious security risks". GitHub asks a reporter of a pre-release bug to "please verify that it also exists in the stable version of the same browser". A fallback for the person on an unlisted browser can point at progressive enhancement, in the Service Manual's words: "Use progressive enhancement to give your service the best possible chance of working for the majority of your users." Reading a report from an unlisted browser as a gap in the matrix is this library's inference. Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Unsupported Browsers). ASK What does a person on an unlisted browser get? What does the team do with a bug report from one: investigate, ask the reporter to reproduce it on a listed browser, decline? On what grounds may the team decline a fix? Could a support agent say the answer aloud? GOOD "Anyone on a browser the matrix does not list still gets catalogue search and record pages, because the site is built to work without scripting. We read every report from an unlisted browser and try to reproduce it on a listed one first. We decline a fix when it would cost more than the problem is worth or when the browser cannot be supported safely, and the reply says which." WEAK "Other browsers are not supported." (a decision with no answer for the person on one: it says nothing about what they get or whether their report is read) TRAP Stopping at "not supported". The GDS Way's posture is to investigate a report from outside the policy and then decide. A flat refusal skips the investigation, and on this library's inference it also loses the sign that the matrix may have a gap. -->
**What a user gets:** {{unsupported_user_experience}}
**What happens to a bug report from one:** {{unsupported_bug_triage}}
**Grounds for declining a fix:** {{declining_grounds}}
## Ownership
<!-- WHAT The named owner, who may change a grade or the threshold, and who is told when one changes. WHY The standing-standards family asks every member to name who notices when the document goes wrong, and an unowned matrix has nobody to notice. The GDS Way page names its owner in its own text: "by the page owner #frontend". GitHub's browser-support library names a team that decides whether a polyfill can go. Jenkins records that each of its changes followed a mailing-list thread or a governance meeting, so its owner is a community process rather than a person. THE ROLE NAMES BELOW ARE THIS LIBRARY'S POSITION. No source read names a person or role who may change a grade. The GDS Way states that its content is intended for internal use by the GDS Digital Products community, so it is cited here as one government team's guidance, not as a standard. Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Ownership). ASK Who owns this matrix, by role and by name? Who may change a grade or the threshold figure? Who must be told when one changes? Where does the discussion happen? GOOD "Owner: the web officer, discussed in the #web-team channel, currently Dana Whitlow. A grade or the threshold figure changes only with the owner's agreement after a note in #web-team. The head of digital services and the front-desk lead are told of any change." WEAK "The team owns it." (names no role, no person and no place where a change is discussed) TRAP Naming a team and no person, or letting anyone edit a grade with nobody told. A matrix with no owner and no trigger is the ungated case the family contract describes. -->
**Owner (role and person):** {{owner_role_and_person}}
**Where changes are discussed:** {{change_channel}}
**May change a grade or the threshold:** {{who_may_change}}
**Told when a grade changes:** {{who_is_told}}
## Review Trigger
<!-- WHAT A table of conditions, owners and actions: what event makes this matrix wrong, who notices, and what they do about it. The conditions are things an owner can observe, not dates, with the annual floor last. WHY The standing-standards family requires every member to carry this mechanism. THE TABLE FORM IS THIS LIBRARY'S POSITION: the sources show the content in prose and in dated change histories. The conditions have these sources. A browser crosses the threshold: the Service Manual records "Updated the requirement to test in Safari for iOS from 9.3 to 10.3 to reflect usage statistics." A vendor ends support: USWDS records "the Design System no longer supports Internet Explorer 11", and Jenkins "2022-02-01 - Remove support for Internet Explorer, Add Edge". The sources record the drop, not the cause, so reading each drop as a vendor ending support is the library's inference. A new platform is adopted: this condition is the library's synthesis, and no source states it. A bug is triaged against an unlisted browser: the GDS Way describes the report, and reading it as a gap in the matrix is the library's inference. The annual floor: the Service Manual records "18 September 2024 List of browsers to test in updated. Commitment to update browser list updated to annual.", and the GDS Way page says "It needs to be reviewed again on 13 July 2027". The Service Manual's history also records reviews with no change, "Reviewed list of browsers to make sure it's still current, based on recent usage figures for GOV.UK. No change from the June 2020 version of the list.", and three of its changes cite usage. So the list changed when usage moved, and periodic reviews checked whether it had. Those review-only entries date from 2020 and 2021, before the annual commitment of 2024. That is high confidence for GOV.UK and says nothing about any other team. Reading the annual figure as a floor is the library's inference, at medium confidence. Telling users about a raised minimum belongs in the action column. MetaMask: "before bumping our minimum supported browser versions, we should always notify users in the application first." AMP requires "Allow at least 2 weeks of open discussion after the I2D is created before proceeding with the deprecation of the feature", for an audience of library developers. Cockpit asked for "a friendlier message that communicates that their browser does not work and message which browsers do work". Deep dive: platform-device-support-matrix_companion.md section 3 (Anatomy > Review Trigger). ASK What could the owner observe that shows this matrix is wrong? Who notices each condition, by role? What do they do about it, and when a minimum is raised, who is told, how, and is the audience end users or developers? Who runs the annual floor? PRIORITY List the event conditions first and the annual floor last. The floor is the longest the conditions can go unchecked, not a calendar rewrite of the document. ROW HINT A good row states a condition the owner could observe, names a role as owner, and states an action, which for a raised minimum includes the notice to users. A weak row says "review regularly" and names no one. GOOD | Analytics show Safari on iOS has fallen below the demotion figure | Web officer | Demote the row, record the analytics date and the reason in the review history, tell the head of digital services, and if a minimum is being raised, put a notice on the search page first that names the browsers that still work. | WEAK | Review regularly | The team | Update the matrix. | TRAP Naming one condition, usually the usage figure, and treating the annual review as the whole mechanism. That the conditions are several is this library's position. A related slip is a raised minimum with no notice action, which leaves that step to chance. -->
| Condition | Owner | What They Do About It ||---|---|---|| Analytics show a browser or version crossing the threshold, in either direction | {{crossing_owner}} | {{crossing_action}} || A browser vendor ends support for a version a row depends on | {{vendor_end_owner}} | {{vendor_end_action}} || The team adopts a platform this matrix does not yet cover | {{new_platform_owner}} | {{new_platform_action}} || A bug is triaged against a browser the matrix does not list | {{unlisted_bug_owner}} | {{unlisted_bug_action}} || The annual floor: at least once a year, whatever else has fired, the owner reconfirms every row | {{annual_floor_owner}} | {{annual_floor_action}} |
**Review history (one dated line each time a condition fired or the annual floor ran, with theoutcome, including "no change"):** {{review_history}}platform-device-support-matrix_example.md
---title: "Harbourline Bookings Platform and Device Support Matrix"doc_type: platform-device-support-matrixsize: leanproduct_name: "Harbourline Bookings"owner: "Imogen Falk (Web and Booking Lead)"status: activedoc_version: "1.1.0"created: "2026-04-20"updated: "2026-10-06"related_links: []source_template: platform-device-support-matrixsource_template_version: 0.1.0---
> **Worked example.** A filled `platform-device-support-matrix`, the one size this bundle ships, for> Harbourline Bookings, the booking site of a small fictional passenger-ferry operator. It is not on> the library's Acme Analytics thread. The `standing-standards` family contract weakens that> chaining on purpose, because a standing standard belongs to a team rather than to a moment in a> story (see [section 4 of the contract](../../docs/internal/contracts/standing-standards.md)). So> the body below cites no sibling example. It is dated 2026-10-06, the day of the last line in its> review history.>> Read it alongside> [`platform-device-support-matrix_guide.md`](platform-device-support-matrix_guide.md), the rubric> it was graded against. Four decisions are visible in it. One row promises more than the team> tests. One row has been demoted. Two Review Trigger conditions have fired, so the table is shown> working. The assistive-technology combinations are named by reference and held nowhere in the> tables. The separate Grade and Tested columns, the Notes column, the role names and the table form> of the Review Trigger are this library's own position, not received practice. Every figure, date> and name below is illustrative, and the threshold figures are Harbourline's own choice for its own> traffic, not norms for anyone else.
# Harbourline Bookings Platform and Device Support Matrix
## Purpose and Scope
**Product covered:** The public booking website, where visitors search ferry sailings, choosefoot-passenger or vehicle places, pay by card, and show an e-ticket at the quay. It does not coverany native app, because Harbourline has none.
**Who relies on this matrix:** The contractor who runs the pre-release checks, the harbour officesupervisor who answers customer complaints, and the developer who decides what the seat map and thedelay ticker must run on.
**Where the record lives:** This file, in the booking site's repository. No tool holds a secondcopy, so a disagreement between this file and anything else is settled in favour of this file.
**Audiences exempt, and the evidence:** None. The harbour office staff take phone bookings on thesame public pages that visitors use, on their own office PCs. A gap for a visitor is therefore alsoa gap for the person helping that visitor by phone, and the staff have no separate tool that couldjustify a lower promise.
**Assistive-technology combinations live at:** The accessibility test guide,`docs/accessibility-test-guide.md` in the same repository. It lists the screen reader and browserpairs the contractor runs. This matrix holds none of them.
## Support Grades
The grades run from the strongest promise to the weakest. In this document, "supported" means theSupported grade and nothing wider. Testing is marked in its own column in The Matrix.
| Grade | What a user gets | What the team tests | What the team does about a bug ||---|---|---|---|| Supported | The whole booking path, including the animated seat map and the live delay ticker | The full path, run manually on a physical device or a current desktop install before every release, ending in a card-provider test payment | Fixed in the next release, and a bug that blocks a booking holds that release || Core | Sailing search, booking, payment and the e-ticket. The seat map may appear as a plain list and the delay ticker may be missing | The full path, run by hand only when the search, booking or payment pages change | Fixed when it stops a booking or a payment. Otherwise it waits behind bugs on Supported browsers || Not listed | Whatever the booking pages do without their enhancements, plus the harbour office booking line for help | Nothing, on any schedule | The report is read and reproduced on a listed browser first. A fix is declined only on the grounds written under Unsupported Browsers |
## The Matrix
**Effective from:** 2026-09-03, the date of the Review History line that last changed a row.
**Mobile browsers**
| Browser | Platforms | Minimum version or version rule | Grade | Tested | Notes ||---|---|---|---|---|---|| Safari | iPhone and iPad (iOS and iPadOS) | The current and the previous major release | Supported | Tested | One iPhone and one iPad live in the harbour office test drawer. Both run the full path before each release || Chrome | Android | The latest stable release and the one before it | Supported | Tested | Two Android handsets run the full path. One is a deliberately low-memory model, so the seat map is judged on a slow device || Samsung Internet | Android | The latest stable release | Supported | Untested | The promise is higher than the testing. The test drawer has no Samsung handset and the budget has no device-rental line. Rustam Beketov accepted this on 2026-04-20 until a handset is bought. A report from this browser is reproduced on Chrome first || Firefox | Android | The latest stable release | Core | Tested | Demoted from Supported on 2026-09-03. Its share of booking sessions was 1.2 percent in the July read and 1.1 percent in the August read, both under the exit figure. Checked by hand when search, booking or payment pages change. Last checked on 2026-09-17, after the payment page changed |
**Desktop browsers**
| Browser | Platforms | Minimum version or version rule | Grade | Tested | Notes ||---|---|---|---|---|---|| Chrome | Windows, macOS | The latest stable release and the one before it | Supported | Tested | Run on a Windows PC and the office MacBook before each release || Edge | Windows | The latest stable release and the one before it | Supported | Tested | Run on the harbour office PCs before each release. Staff use these same PCs to take phone bookings || Safari | macOS | The current and the previous major release | Core | Tested | Checked by hand on the office MacBook when search, booking or payment pages change || Firefox | Windows, macOS | The latest stable release | Core | Tested | Checked by hand when search, booking or payment pages change. Last checked on 2026-09-17, with the Android row |
## Threshold Rule
**Entry figure:** A browser and platform pair joins the matrix when it makes up at least 3 percentof booking sessions over the last 90 days, and it joins as Supported. A pair between 1.5 and 3percent joins at Core. These numbers are illustrative.
**Exit rule:** A Supported row falls to Core after two consecutive monthly reads below 1.5 percent.A Core row leaves the matrix, and becomes Not listed, after two consecutive monthly reads below 0.5percent. Both numbers are illustrative.
**Analytics source, and who reads it:** Harbourline's own booking-site analytics, in the reportnamed "Sessions by browser and platform". Imogen Falk reads it on the first working day of eachmonth and writes the read date beside any change.
**Who decides:** Entries and demotions are Imogen Falk's call, made from the monthly read. RustamBeketov, the operations director, is told of each one, because a change moves which handsets thetest drawer must cover.
**Why this figure is ours:** Most bookings come from phones, and a small visitor base lets onebrowser carry a large share. The entry figure was set so that the test drawer, which holds fourhandsets and tablets, can cover the Supported rows. Samsung Internet is the one row it does notcover, which is why that row says Untested. Neither figure is taken from another team's rule, and afigure that fits Harbourline's traffic may not fit anyone else's.
## Unsupported Browsers
**What a user gets:** The booking pages arrive as plain forms first and are enhanced afterwards. Avisitor on a browser outside the matrix can usually reach the booking form and the payment step. Theseat map and the delay ticker may not appear. If the form fails, the page shows the harbour officebooking line.
**What happens to a bug report from one:** Imogen Falk takes each report that names an unlistedbrowser, gets its version from the visitor, and replays the steps on a listed browser. A problemthat reproduces there is fixed at that browser's grade. Whatever the replay shows, she then checksthe unlisted browser's share in the analytics report. If it sits under the entry figure, no row isadded, and the reply says so.
**Grounds for declining a fix:** The reply names which ground applied, by number.
1. The fix would change the payment step in a way that puts bookings from a Supported browser at risk.2. The browser cannot open the secure payment form that Harbourline's card provider requires.3. The problem appears only with an extension, a content blocker or a modified browser build that the reporter installed, and it goes away without it.
## Ownership
**Owner (role and person):** Web and Booking Lead, currently Imogen Falk.
**Where changes are discussed:** The `#booking-site` chat channel, and the fortnightly web stand-up.
**May change a grade or the threshold:** Imogen Falk may change a grade once she has posted theanalytics read behind it in `#booking-site`. Moving the entry figure or the exit figure needsRustam Beketov's sign-off, because it changes how many handsets the test drawer must cover.
**Told when a grade changes:** Rustam Beketov, operations director, and Mirela Dussault, harbouroffice supervisor. Pavol Reiter, the contractor who runs the pre-release checks, is told so hischecklist matches the table.
## Review Trigger
| Condition | Owner | What They Do About It ||---|---|---|| Analytics show a browser or version crossing the threshold, in either direction | Web and Booking Lead | Applies the entry or exit rule and moves the row. Writes a review-history line giving the date of the read and the browser's share. Tells the operations director, the harbour office supervisor and Pavol Reiter, so his checklist changes with the row. If a minimum is being raised, adds a line to the booking confirmation e-mail and the e-ticket page at least four weeks before the raise takes effect, because visitors book weeks ahead and open the e-ticket on the day they sail || A browser vendor ends support for a version a row depends on | Web and Booking Lead | Raises the minimum on the affected row and records the vendor's date in the review history. Puts the same line in the confirmation e-mail and on the e-ticket page before the raise. Warns the harbour office supervisor, so staff can answer calls from visitors who see the notice || The team adopts a platform this matrix does not yet cover | Operations Director | Before launch, asks the Web and Booking Lead to add rows with a minimum version, or to record that the platform sits outside this matrix and say where its support is written down || A bug is triaged against a browser the matrix does not list | Web and Booking Lead | Follows the steps under Unsupported Browsers. Checks that browser's share in the analytics report. Adds a row if the share is above the entry figure. Otherwise records "no change" in the review history with the share found || The annual floor: at least once a year, whatever else has fired, the owner reconfirms every row | Web and Booking Lead | Reads the latest analytics report against every row, and asks Pavol Reiter to confirm each Tested value from what he runs. Records the result in the review history, including "no change" |
**Review history (one dated line each time a condition fired or the annual floor ran, with theoutcome, including "no change"):**
- **2026-04-20.** First version written from the March analytics read, with every row set on this date. Samsung Internet on Android was marked Supported and Untested, and Rustam Beketov accepted that until a handset is bought. Safari on macOS and Firefox on desktop entered at Core, from the band between 1.5 and 3 percent. No earlier history exists.- **2026-09-03.** The condition "a browser crossing the threshold" fired. The read dated 2026-09-01 showed Firefox on Android below 1.5 percent for the second month running. Imogen Falk demoted its row from Supported to Core. Rustam Beketov and Mirela Dussault were told in `#booking-site` the same day, and Pavol Reiter was told so his pre-release checklist dropped the Firefox on Android row. A demotion raises no minimum, so no notice for visitors was needed. The effective-from date above is this date.- **2026-10-06.** The condition "a bug triaged against an unlisted browser" fired. A visitor on Vivaldi for Windows reported a seat map that stayed blank. Imogen Falk reproduced it on Chrome with a content-blocking extension installed, so the fault was in the seat map and not in the unlisted browser. The seat map will show its plain list when its script is blocked, in the next release. Ground 3 for declining a fix was not applied, because the seat map should fall back to its plain list whenever its script is blocked, which the Unsupported Browsers promise already covers. The analytics check ran, as it does whenever a report names an unlisted browser. It showed Vivaldi at 0.3 percent of booking sessions, under the entry figure, so no row was added. Outcome: no change to the matrix.- **Annual floor.** Not yet run, because the matrix is under a year old. The floor does not restart when another condition fires, so the lines of 2026-09-03 and 2026-10-06 did not move it. It falls due on 2027-04-20, a year after the first version.Provenance
Section titled “Provenance”The reasoning, the history and every source, in the repository:
- Companion - the long-form argument: why these sections, where the sources disagree, and what the bundle refuses to claim
- History - what changed in this bundle, and when
- Research log - every source consulted, with what each one actually supports
- Catalog metadata - the machine-readable record this page is generated from
Catalog record: 7 sections across 1 format(s), methodology generic, typically owned by Frontend or web lead.