Patterns across the analyses

The headline figures

n = 78 findings in 9 analyses

Analyses published

9

In Retail, Technology, Banks and insurers and Telecom and health, assessed up to 27 October 2026.

Pages examined

85

From 4 to 15 per analysis, at up to three screen sizes each.

Findings

78

78 findings across 9 analyses; none high, 45 medium.

Analyses with a measured axe rule count

7

7 of 9 analyses; the rest were read by hand, with no machine output to count rules from.

Findings per analysis

8

The median of 9 analyses, which found from 6 to 11 each.

Most are medium: 45 of 78. The levels are explained on the method page.

The heuristics that fail most

n = 78 findings in 9 analyses

Findings by heuristic, most first, each bar split by level.

• High
• Medium
• Low
• Best practice

With this rubric high is rare by design, so the ranking is led by WCAG A and AA failures on primary task paths: these analyses are an accessibility-led benchmark with UX commentary. The levels are judged against the rubric on the method page.

1. Accessibility

   34 findings in 9 of 9 analyses. The latest:

   • A 'Get started' button on the chequing page is named for a Starbucks offer
   • Closed FAQ answers on the GIC rates page still take keyboard focus
   • Footnote links '2' and '3' in the mortgage rate table are 10 px tall

2. Mobile and touch

   9 findings in 7 of 9 analyses. The latest:

   • On a phone the privacy notice covers half the screen and the open menu
   • Customer service phone numbers are plain text on a phone, not links to call
   • On phones the Medicare call bar nests a phone link inside a button and covers the search

3. User control and freedom

   7 findings in 5 of 9 analyses. The latest:

   • The privacy preferences open with Performance and Targeting switched on
   • Footer links open a new tab with no icon and nothing in their names
   • Escape does not close the credit cards side menu

4. Visibility of system status

   5 findings in 5 of 9 analyses. The latest:

   • CD rates stay blurred behind a ZIP code prompt that opens at every size
   • Most links tried on Shop and Medicare do not change when the pointer is on them
   • Seen from Toronto, pages show a Seattle warehouse and ZIP 98101 the visitor never chose

5. Content and tone

   4 findings in 4 of 9 analyses. The latest:

   • On narrow screens the card list opens in Grid view, without APR or annual fee
   • iPhone prices carry two or three payment schemes, with tiny footnote links
   • Advanced Image Search reuses the web page’s title, heading and wording

6. Consistency and standards

   3 findings in 3 of 9 analyses. The latest:

   • The price card stays titled ‘2-Day Delivery’ after In-Warehouse is chosen
   • One destination is “Home” on most pages and “TV & Home” on others
   • One data setting has two names, and its card opens a shared browser window

7. Help and documentation

   3 findings in 3 of 9 analyses. The latest:

   • ‘Help & Contact Us’ offers no contact route when signed out
   • Advanced Search gives no example of what to type in its four boxes
   • Desktop help lists about 270 links in 13 cards and offers chat as the only contact

8. Navigation and wayfinding

   3 findings in 2 of 9 analyses. The latest:

   • The top navigation does not show which section you are in
   • Find a Store opens on three US stores for a visitor in Toronto
   • The ‘Household Essentials’ menu item and its page carry three different names

9. Recognition rather than recall

   3 findings in 3 of 9 analyses. The latest:

   • Each footnote on the mortgage rates page opens the same list of 11 notes
   • Size tiles mix rolls, sheets and packs, and print the unit price at 10px
   • On phones the AI entry is an unlabelled sparkle icon with a different name

10. Recovering from errors

    2 findings in 2 of 9 analyses. The latest:

    • The branch locator opens on a "could not find" message before any search
    • Pressing Find with an empty field reports a failed search and clears the list

11. Trust and credibility

    2 findings in 2 of 9 analyses. The latest:

    • The sign-in page's Broker E-Store link has a testing tool's preview token
    • The first row of results is all ads, labelled in 11px type

12. Aesthetic and minimalist design

    1 finding in 1 of 9 analyses. The latest:

    • Seven pages are 1446 px wide at 1440 px and move 6 px sideways

13. Conversion

    1 finding in 1 of 9 analyses. The latest:

    • On a phone, ‘Add to cart’ is 1005px down the product page and does not stick

14. Visual hierarchy

    1 finding in 1 of 9 analyses. The latest:

    • On a phone the product page opens with an ad, and the price needs a scroll

Where on a site the problems are

n = 78 findings in 9 analyses

Each examined page is filed by the role it plays (home, search, product), so product pages are compared with product pages. A cell counts the findings about that kind of page under that heuristic; an empty cell found none. Most are about product: 17 of 78.

Phone or desktop

n = 78 findings in 9 analyses

The screen size each finding was seen at: 50 at desktop, 4 at tablet and 24 at phone. A finding is filed at the size it shows most clearly; not every analysis read a tablet size.

• Desktop
• Tablet
• Phone

WCAG 2.2 criteria cited

n = 36 findings citing WCAG 2.2, of 78

The success criteria the findings cite, most cited first. A finding may cite more than one.

1. 4.1.2 Name, Role, Value (level A)

   14 findings

   • Closed FAQ answers on the GIC rates page still take keyboard focus
   • On a phone the sign-in page's menu buttons say 'collapsed' while their menus are open
   • Hidden slides in the checking page's article carousel still hold focusable links
   • The ZIP code prompt is not exposed as a dialog, and a page link paints over it
   • The Medicare page's 'How can we help you?' search has invalid ARIA and an unseen focus
   • On phones the Medicare call bar nests a phone link inside a button and covers the search
   • On the plans page the 'Max 12 lines' information button has no name
   • The return policy page's tabs and breadcrumb carry invalid ARIA
   • The phone app banner is on every page, and its close mark is 7px
   • The Trade In panel’s close button is named ‘Learn more, Apple Trade In’
   • Gallery and accordion markup uses roles and list structures that do not exist
   • Hidden controls that still take keyboard focus
   • The drop-downs on both Advanced Search pages do not say whether they are open
   • On the Chrome page, visible controls are hidden from screen readers but take focus

2. 2.5.8 Target Size (Minimum) (level AA)

   5 findings

   • Footnote links '2' and '3' in the mortgage rate table are 10 px tall
   • The sign-in box's Show password button is 21 px tall at desktop size
   • On Shop for a plan the header’s ‘Español’ link is 20 px tall, with a target within 12 px
   • Product-card delivery links and footer links are under the 24 px target size
   • Footer links are 21px tall and tightly stacked, in one ungrouped run

3. 2.5.3 Label in Name (level A)

   4 findings

   • A 'Get started' button on the chequing page is named for a Starbucks offer
   • On Shop for a plan, 'Mental health virtual therapy' is named 'View accident insurance'
   • Plan cards read 'Bring your own phone' but are named 'Keep your own phone'
   • Button and link names do not contain their visible words

4. 1.4.3 Contrast (Minimum) (level AA)

   3 findings

   • The Login Help links on the sign-in page measure 4.27:1
   • The home page's quick links and the plans page's savings miss text contrast
   • Grey body copy on product pages measures 3.62:1

5. 2.4.7 Focus Visible (level AA)

   3 findings

   • The Medicare page's 'How can we help you?' search has invalid ARIA and an unseen focus
   • The store locator's 'Enter ZIP code or city, state' field shows no focus
   • On tablets and phones, the first Tab stops on the home page show no focus

6. 1.1.1 Non-text Content (level A)

   2 findings

   • A 'Get started' button on the chequing page is named for a Starbucks offer
   • On the Chrome page, visible controls are hidden from screen readers but take focus

7. 1.3.1 Info and Relationships (level A)

   2 findings

   • The return policy page's tabs and breadcrumb carry invalid ARIA
   • Gallery and accordion markup uses roles and list structures that do not exist

8. 1.4.4 Resize Text (level AA)

   2 findings

   • The EasyWeb sign-in page turns off pinch zoom
   • The two pages that control your search data block pinch zoom

9. 1.4.10 Reflow (level AA)

   2 findings

   • At 374px the delivery-fee line makes the product page scroll sideways
   • Zoomed to 400%, the desktop home page scrolls sideways and hides ‘Sign in’

10. 1.4.1 Use of Color (level A)

    1 finding

    • Links inside sentences are told apart by colour alone on the home and switch pages

11. 2.1.1 Keyboard (level A)

    1 finding

    • On phones the home page's cash-offer carousel has no usable Previous and Next buttons

12. 2.4.2 Page Titled (level A)

    1 finding

    • Advanced Image Search reuses the web page’s title, heading and wording

13. 3.3.2 Labels or Instructions (level A)

    1 finding

    • The Medicare ZIP field has an empty label and pre-fills a value it cannot accept

What the engine counts

n = per check, in the table’s last column

What a browser counts on every page examined. The first strip is axe, the accessibility engine, as distinct rules failed rather than elements, drawn only for the 7 of 9 analyses whose axe output was saved; each analysis lists its rules by impact. Each dot is one analysis and links to it; the accent mark is the median. A count an analysis does not stand behind is left out, and its page says why; some were counted by hand where a site refused the crawl.

By sector

n = 78 findings in 9 analyses

• Retail

  n = 23 findings in 3 analyses

  17 pages examined; a median of 8 findings per analysis; none high, 10 medium, 11 low and 2 best practice.

• Technology

  n = 21 findings in 2 analyses

  25 pages examined; a median of 10.5 findings per analysis; none high, 14 medium, 6 low and 1 best practice.

• Banks and insurers

  n = 20 findings in 2 analyses

  22 pages examined; a median of 10 findings per analysis; none high, 12 medium, 6 low and 2 best practice.

• Telecom and health

  n = 14 findings in 2 analyses

  21 pages examined; a median of 7 findings per analysis; none high, 9 medium, 3 low and 2 best practice.

By platform

n = 1 of 9 analyses carry a platform reading

The technology the engine detected on each site from what its pages send: headers, the generator tag, script hosts. A site can run several. A technology is ranked by its median findings per analysis once 5 analyses carry it; until then it is listed by name with its n.

Fewer than 5 analyses each, so not ranked

• Adobe Experience Manager (cms): 11 findings per analysis, the median of 1 analysis
• Angular (frontend): 11 findings per analysis, the median of 1 analysis

The data

n = 78 findings in 9 analyses

Every finding counted here, one row each, as JSON and CSV: 78 rows. The columns are report, assessedOn, sector, platform, area, viewport, heuristic, level, wcag; area is the page role, not the site’s own page. patterns.json also carries each analysis’s axe rules by impact and the dataset’s version; every change to a published finding is listed on the revisions page. Rebuilt with every analysis published. Free to use under CC BY 4.0, with attribution to 4th Culture.