mr-onion

Labelled Failure Set

The seed for the failure-triage gate Phase 34 deferred (its “Related, deferred” section): every test failure Mr Onion has hand-classified, with its verdict, the evidence, and its source run. Started 2026-09-30 in Phase 36 (TFP-123, Decision 1).

Verdicts

Sources. comparisons/ and runs/ are gitignored, so a source run exists only on the machine that made it; the evidence column carries what a reader needs without it. benchmarks/ and examples/ are committed.

Status says whether a rule has since removed the cause (Phase 34’s “rule first”: a pattern a deterministic rule can recognise is handled by the rule, and a classifier only sees what is left).

Template errors

# Site Test Evidence Source run Status
T1-T7 test-assist 7 “action/form element check” tests: #btnRecord, #btnClearSession, #uploadImage, #btnStopRecord, #textInput, #btnTextOk, #btnTextCancel toBeVisible() failed with Received: hidden on elements hidden by the app’s design until a user action (Video mode, a restored session, the Text tool); #uploadImage is a styled-invisible file input inside a visible label TFP-040, examples/testforge-app/test-assist/run-report.json; analysis in notes.md there Fixed by rule: TFP-053/054 (hidden at inspection means an attached check, not a visibility claim)
T8 comparison fixture reliability: dependency failure injection on /login Blocking the /login form action’s route also blocked the page’s own navigation to /login TFP-099, comparisons/2026-09-24T16-*-fixture (Phase 21 report, “Results by area, fixture”) Open
T9 the-internet TFFU-002 “surfaces a sign-in related control for the described flow” Visited /; read by hand, / is a list of examples with no sign-in control (0 matches), while /login, the page inspected for the request, has 3, all visible TFP-099 (2026-09-24) and comparisons/2026-09-28T20-19-37-582Z-the-internet, runs 0a73a87b, f2fd9c89; reproduced 2026-09-30, runs 15dbf0fd, 04f40ce8 Fixed, TFP-123: the test visits the first inspected page with a sign-in control. Passes on run 03bae90c
T10 the-internet TFFU-007 “navigation link health: Untitled link to https://github.com/tourdedave/the-internet” a[href=…]:visible found nothing. Read by hand: the link is <a><img style="position:absolute…" alt="Fork me on GitHub"></a>, so the anchor’s own box is 0x0 and Playwright calls it hidden, while the 149x149 image is on screen, uncovered, clickable, and the anchor is the first Tab stop as T9 Fixed, TFP-123: a link counts as rendered when it or its content renders; visibility checks and clicks use what renders it. Passes on ae16b41c, 03bae90c
T11 the-internet TFFU-010 “link types: … classified correctly” Same link and cause as T10 as T9 Fixed, TFP-123, as T10
T12 the-internet TFFU-015 “keyboard navigation: links can receive focus and activate with Enter” (30 s timeout) Three layered causes, each found only once the one before was fixed. (1) The T10 locator never matched, so focus() waited out the test. (2) Pressing Enter only starts a navigation: the test’s next waitForLoadState returned at once on the old page, the URL looked unchanged, and the next link was then sought on github.com until the timeout. (3) The Elemental Selenium link is http://…, the site answers 301 to https://, and the test compared URLs including the scheme as T9; (2) and (3) measured 2026-09-30, runs db67113c, 5c1c8351 Fixed, TFP-123: (1) as T10; (2) waits up to 5 s for the URL to change, then reloads the links page; (3) compares host and path, allowing an http-to-https upgrade. Passes on 03bae90c
T13-T15 the-internet Comprehensive security, “responds, with its security headers”: /basic_auth, /digest_auth, /download_secure Each answered HTTP 401 (read with curl: /basic_auth 401). They are behind HTTP authentication; refusing is the server working, and Decision 3’s failures are exposures only TFP-136, benchmarks/2026-10-01-phase35-first-slice/before-fix-security-the-internet.json (run aeb33b15) Fixed, 198dbec: a page answering 400 or more is Not run with the reason (“behind a login, and this run does not sign in”); its version check still reads the headers. Passes on run 12e385fb
T16 the-internet Comprehensive security on /add_remove_elements HTTP 404. The site graph normalises the path; the link’s real href is /add_remove_elements/, which answers 200 TFP-136, benchmarks/2026-10-01-phase35-first-slice/before-fix-security-the-internet.json (run aeb33b15) Fixed, 198dbec: a page known only from the crawl is loaded by the href a link to it uses. Covered as /add_remove_elements/ on run 12e385fb
T17 the-internet Comprehensive security on /authenticate HTTP 404 to a GET. Known only from a form-submit edge: it is /login’s POST target, not a page TFP-136, benchmarks/2026-10-01-phase35-first-slice/before-fix-security-the-internet.json (run aeb33b15) Fixed, 198dbec: nodes reached only by a form submit or scripted action are not loaded
T18 rizonetech.com TFAC-007 “accessible name and role: Untitled link to https://rizonetech.com/” - a false pass, not a failure The logo link is <a href="https://rizonetech.com/" class="custom-logo-link"><img alt="" title="Rizonetech Logo - Light R2 → Rizonetechcom"></a>: no title or aria-label on the link. HTML-AAM 4.1.10 gives an alt="" image no name and uses title only without an alt attribute, so the link has none, and axe’s link-name flags it. Playwright 1.59.1 names it from the image’s title, so toHaveAccessibleName passed. Chromium 147’s own accessibility tree (CDP, evidence only) gives the link the name "", with no name sources. On 2026-10-01 the check located a[href] first, which is the page’s “Skip to content” link, so the logo was not what it tested. The false pass on the logo itself dates from 0504f92, whose live run passed it Live 2026-10-05: benchmarks/2026-10-05-tfp152-logo-link/live-2026-10-05.log; the 2026-10-01 suite in benchmarks/2026-10-01-rizonetech/standard-suite.spec.ts Fixed, TFP-152: the check follows HTML-AAM 4.1.10 and fails it as “has no accessible name: HTML-AAM 4.1.10 gives an <img alt=""> no name and does not use its title”. Live, the 40 checks on / give 39 passed and this 1 failed (live-checks-2026-10-06.log there)
T19 practicesoftwaretesting.com 15 tests: “navigation link health: Hand Tools” and “link types: … classified correctly” (5 pages each), “keyboard navigation: links can receive focus and activate with Enter” (5, 30 s timeout) “Hand Tools” is <a href="/category/hand-tools"> inside the closed “Categories” dropdown (.dropdown-menu), hidden until the menu opens. The first two require it visible; keyboard navigation waits to focus it until the test times out. Visibility by design, as T1-T7 TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open
T20 practicesoftwaretesting.com TFRM-001 “keeps the core journey visible across common viewport sizes” At 390 px “Home” is in the collapsed navbar, hidden behind a visible .navbar-toggler; visible at 768 and 1440 px TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open
T21 practicesoftwaretesting.com “form element check: Name (A - Z)… [Sort] [form]” on 3 pages getByText of the options’ joined text matches the <select> itself; the check then counts input, textarea, select, button inside it: 0. A <form> around it exists TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open
T22 practicesoftwaretesting.com “form element check” for the /contact and /auth/login forms Located by the whole form’s joined label text (“Email address *Password *”); no single element has that text, so nothing is found TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open
T23 practicesoftwaretesting.com TFAC “accessible name and role: SearchX Search [form]” The located element is a plain <div> (no role, no aria-label): a generic element has no accessible name to give, so asking for one is the check’s mistake TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open
T24 the-internet “orphan page approximation: each scanned page is linked from another scanned page” (3 and 4 tests) The check counts inbound links only among inspected pages. The host timed out on /, so inspection lost the page that links to every example, and /checkboxes, /dropdown, /tables (and /login) each looked unlinked. Read by hand, each is linked from / TFP-150 re-run, benchmarks/2026-10-06-tfp150/the-internet-rerun-* (d90de57 and d202315) Open
T25 Toolshop, both sites the form check (TFP-160) and the form name check (TFP-151): “no form on /#/auth/login with no action, no method” (with-bugs 1; clean 5, every form), “(a form) is not on the page” (clean 2) The generated check looks for the <form> straight after domcontentloaded; the Angular app draws its forms later, so there is none yet. Read by hand: each form is on the settled page with the recorded action and method. TFP-160’s and TFP-151’s fixtures were static HTML, so their tests never met it TFP-154, benchmarks/2026-10-06-toolshop-known-bugs/runs/*-standard-records.json (26fed09) Fixed, TFP-167 (e812795): every visit waits for the page to draw. In the re-score (runs-tfp167/, 7577252) every form check passes on both sites
T26 with-bugs Toolshop “form element check: 1 [Versatile combination pliers …] [input]” on /#/product/1 The quantity <input type="number"> was recorded with its value “1” as its text and located with getByText('1'); an input’s value is not text, so nothing is found Also on the local with-bugs product page (TFP-180), the quantity field found by its value under a description’s text. as T25 Open
T27 with-bugs Toolshop “api element check: First nameLast nameEmail addressSubject…” on /#/contact The contact <form> was classed api, so TFP-160’s form lookup (category form only) didn’t apply and it was located by all its labels joined, as T22. Its message happens to contain the subject list’s error entries (bug 60) as T25 Open
T28 with-bugs Toolshop passive security “puts no password or session token in a URL”: “a form on /#/auth/login would send its password field in the URL (method GET …)” The check flags any GET <form> holding a password input. Read by hand: the field has no name, so a native submit wouldn’t send it, and the app submits by script. The clean site’s form is the same; its check passed only because the form wasn’t drawn yet (T25). In the re-score (runs-tfp167/) it fails on both sites’ login and register pages as T25 Open
T29 with-bugs Toolshop “link types: … classified correctly” on 5 pages: getByRole('link', { name: 'Categories' }) not found “Categories” is <a href="#" role="button">: its role is button, so a link-role lookup can’t find it. (The clean site’s is a <button>, not a link) as T25 Open
T30 with-bugs Toolshop “orphan page approximation: each scanned page is linked from another scanned page” on all 8 pages: “/#/product/6 is not linked from any other inspected page” Read from the run’s inspection profile: home and /#/category/hand-tools exposed no product link at all, where the re-score’s inspection of the same pages had #/product/1 to #/product/9. Inspection read them before the product cards drew (they load from the API), so the product page looked unlinked. The tests themselves ran on drawn pages (axe names the cards), so no bug’s score depends on it. Not the selection: product 1 would have failed the same way TFP-175, benchmarks/2026-10-07-tfp175-selection/runs/with-bugs-standard-* (7233751) Open: inspection doesn’t wait for a hash route’s data, as generated visits do since TFP-167
T31 practicesoftwaretesting.com “link types: … classified correctly”: getByRole('link', { name: 'Combination Pliers ABCDE$14.15' }) (a product card) and { name: 'Learn Test AutomationHands-on courses for Playwright, Robot Framework, APIs and more' } (a footer card) not found The name is the link’s text as inspection joined it, with no space between its block children (“Automation” + “Hands-on”, the CO2 letters and the price). The browser’s accessible name separates them, so an exact lookup finds nothing, though the link is there. The re-score (5) and TFP-175 (7), clean site only (the with-bugs footer and cards differ) Also on the local clean copy (TFP-180): “Claw Hammer with Shock Reduction Grip ABCDE$13.41”. TFP-178, from benchmarks/2026-10-06-toolshop-known-bugs/runs-tfp167/clean-standard-records.json and benchmarks/2026-10-07-tfp175-selection/runs/clean-standard-records.json Open
T32 Toolshop, both local copies TFP-187’s spelling finding on the product page: "Unsplash" in the link "Unsplash" ... did you mean "splash"? “Unsplash” is the photo library the product image is credited to, a brand spelt right. The footer’s identical credit isn’t read (Emil’s calibrated rule leaves <footer> out); this one sits in the product page’s body, and the dictionaries suggest “splash” within edit distance 2 benchmarks/2026-10-12-phase41-batch2/runs/tfp187-*.json Open: a brand near a dictionary word is reported under the agreed rule
T33 Toolshop, both hosted sites TFP-182’s “navigation links land where their text says” on /: “read-only guard breach, aborted: POST https://…/cdn-cgi/rum?” (several) /cdn-cgi/rum is Cloudflare’s real-user-monitoring beacon, which the hosted sites’ Cloudflare front sends as the page navigates. TFP-181’s guard aborts every non-safe request during an interaction, so following a header link breaches on it, though the test sent nothing itself. The local copies aren’t behind Cloudflare and pass. The risk was named in 7069078’s commit (“a third-party beacon sent during an interaction would also be a breach; measured at TFP-188”). On hosted with-bugs it also hides bug 29 TFP-188, benchmarks/2026-10-13-tfp188/runs-hosted/*-standard-records.json Fixed, TFP-199 (the guard aborts a site’s beacons but notes them, not a breach: Chromium’s ping type, and /cdn-cgi/rum); the hosted re-run is recorded in the run-evidence-sweep backlog

Model errors

# Site Test Evidence Source run Status
M1 the-internet local accessibility test, exact accessible name “Login” The button’s accessible name is “ Login” (with a leading space); an exact match on “Login” failed TFP-099, comparisons/2026-09-24T16-47-24-029Z-the-internet Open
M2 the-internet; comparison fixture local tests with a guessed form role name The model named a form role the page doesn’t expose; again in the Phase 36 benchmark, C3 functional #2 (getByRole('form', { name: 'Sign in' }), re-run with the base URL) as M1; benchmarks/2026-09-29/C3/rerun-with-base-url.json Open
M3 the-internet gap-fill database test “data entered on the login page is saved” Invented a login-and-reload flow: typed a username and password, submitted, reloaded, expected the fields to keep the values. A sign-in form doesn’t save input. Passed all four generation gates. Repeated in two runs TFP-120, comparisons/2026-09-28T20-08-32-878Z-the-internet, …20-19-37-582Z-the-internet Avoided, not fixed: with page data (TFP-122 inspects automatically) the template reports database not applicable and the model isn’t asked
M4 comparison fixture local functional test “should display/show an error message for invalid credentials” Asserts an error message the fixture’s sign-in page doesn’t show. Checked per suite: 7 clean-run failures (C1 once, C2 twice, C7 twice, C8 once, the confirmation C1 twice) Phase 36 benchmark, benchmarks/2026-09-29*/*/functional-*.spec.ts Open: moves clean-run counts by 1-2 per configuration
M5 comparison fixture local security test “input that looks like a script is not executed” Detects the injected defect only sometimes: it checks for an alert immediately after clicking Save, without waiting for the saved page to render. C1’s three suites were byte-identical and caught the defect once Phase 36 benchmark, benchmarks/2026-09-29/C1/security-*.spec.ts Open: a detection race, not a false failure

Real

# Site Test(s) Evidence Source run Status
R1 comparison fixture 9 template SEO test failures The fixture has no canonical link, structured data, robots.txt, sitemap or web-app manifest TFP-099, comparisons/2026-09-24T16-*-fixture Real. TFP-124 makes a missing manifest and missing structured data findings, not failures
R2 the-internet template SEO: meta description, canonical, sitemap reference in robots.txt The page has no meta description or canonical link; robots.txt has no Sitemap: line TFP-099 and every later the-internet run, most recently 03bae90c (2026-09-30) Real, and stay failures (Phase 36 Decision 2)
R3 the-internet template SEO: structured data, web-app manifest Neither exists as R2 Real, but optional: findings from TFP-124 (Decision 2)
R4 the-internet Comprehensive accessibility, color-contrast (serious, WCAG 1.4.3) on 17 of 20 pages Measured: every flagged element is the site’s rgb(43,166,203) with white - the footer’s “Elemental Selenium” link (16px, on white), the buttons on /add_remove_elements/, /dynamic_controls, /challenging_dom, and the 32px “Login” heading - at 2.83:1, under 4.5:1 for normal text and 3:1 for large TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R5 the-internet image-alt (critical, 1.1.1) on /broken_images and /dynamic_content The flagged <img> elements have no alt TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R6 the-internet label (critical, 4.1.2) and TFP-121’s field-name check on /checkboxes <input type="checkbox"> checkbox 1<br> - text beside the box, no label TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R7 the-internet select-name (critical, 4.1.2) and TFP-121’s check on /dropdown <select id="dropdown"> with no label anywhere on the page (0 <label> elements) TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R8 the-internet label (critical, 4.1.2) and TFP-121’s check on /dynamic_controls The checkbox and the text input have no label TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R9 the-internet list (serious, 1.3.1) on /disappearing_elements The <ul>’s direct children are LI,LI,LI,LI,LI,UL - a list inside a list without an <li> TFP-136, benchmarks/2026-10-01-phase35-first-slice/accessibility-none-the-internet.herokuapp.com.json (run 84094801) Open (the site’s)
R10 practicesoftwaretesting.com axe button-name (critical, 4.1.2) on /auth/login (and /auth/register, TFP-178) The show-password toggle is <button class="btn btn-outline-secondary"><fa-icon><svg aria-hidden="true">…</svg></fa-icon></button>: no text, no label; its aria snapshot is - button. New with Phase 38’s scan (TFP-148) TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json Open (the site’s)
R11 practicesoftwaretesting.com “storage failure injection: localStorage” on all 5 pages With localStorage throwing, the app throws it uncaught and renders an empty body (0 characters, no products) TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Open (the site’s)
R12 practicesoftwaretesting.com template SEO: meta description and canonical (5 pages each), robots.txt Sitemap:, sitemap.xml No meta description or canonical link after render; /robots.txt and /sitemap.xml answer 200 with the app’s HTML shell TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Real, and stay failures (Phase 36 Decision 2)
R13 Toolshop, both sites template SEO: meta description and canonical on the 5 Standard routes, robots.txt Sitemap:, sitemap.xml No meta description or canonical link; /robots.txt is a real robots file with no Sitemap: line, and /sitemap.xml answers 404. Not a listed bug TFP-154, benchmarks/2026-10-06-toolshop-known-bugs/runs/with-bugs-standard-records.json Real, and stay failures (Phase 36 Decision 2)
R14 with-bugs Toolshop “anchor links point to existing in-page targets”: “Anchor target missing for #filters” (2 pages) <a href="#filters" data-bs-toggle="collapse"> (the small-screen filters toggle); no element has id filters, after the page settles too. The clean site uses a button with an existing target. Not a listed bug as R13 Open (the site’s)
R15 with-bugs Toolshop axe button-name (critical, 4.1.2) on /#/auth/login and /#/auth/register The show-password toggle, <button class="btn btn-outline-secondary"><i class="fa fa-eye"></i></button>: no text, no label. The same defect as R10 on the clean site, which this run’s clean scans didn’t reach (its form wasn’t drawn yet, T25). Not a listed bug as R13 Open (the site’s)
R16 Toolshop, both sites axe list (serious, 1.3.1) on home and the category (.checkbox:nth-child(n) > ul), and on the clean register page (#passwordHelp > ul) A <ul> that directly holds something other than <li>: the filters’ nested category lists. Seen once TFP-167 let the filters draw. Not a listed bug TFP-154 re-score, benchmarks/2026-10-06-toolshop-known-bugs/runs-tfp167/*-accessibility-records.json (7577252) Open (the site’s)
R17 Toolshop, both sites axe target-size (serious, 2.5.8) on /#/product/1 The related products’ <a large href="#"> links are smaller than 24 px with no spacing. Not on the re-score’s clean product; TFP-175 found it on all 9 clean product pages its Comprehensive run covered (.card-body > a[href="#"][large=""]), so it is shared behaviour, not a planted bug. Not a listed bug as R16 Open (the site’s)
R18 with-bugs Toolshop axe color-contrast (serious, 1.4.3) on home’s .float-start Read by hand: <span data-test="out-of-stock" class="float-start text-danger">Out of stock</span>, red rgb(220, 53, 69), below 4.5:1. The clean site passes. Not a listed bug as R16 Open (the site’s)
R19 Toolshop, both local copies TFP-183’s grouping check on the category page: “5 checkboxes sharing the name "category_id" … are not in a fieldset with a legend or a named group” (with-bugs; clean 8), and the same for “brand_id” (with-bugs 10, clean 2) Read in the source: the category page’s category and brand checkboxes sit in no <fieldset> in either version (category.component.html; clean’s only fieldset there wraps the eco-friendly box). The home page’s are grouped in clean (bug 94 is home’s) benchmarks/2026-10-12-phase41-batch1/runs/tfp183-*.json Real; shared by both versions, so not a find under TFP-154’s control rule. TFP-188 (2026-10-13): the same on every category page Comprehensive accessibility covered (with-bugs power-tools and undefined; clean other, power-tools, special-tools), benchmarks/2026-10-13-tfp188/runs/*-accessibility-records.json
R21 with-bugs Toolshop, hosted “anchor links point to existing in-page targets” on the product page, a finding: link "More information" has href="#", which points at no in-page target (a placeholder link) The related products’ “More information” links are <a href="#"> placeholders (the same elements as R17). TFP-141 reports a placeholder as a finding by design; seen on the hosted run’s product page in TFP-188, not on the local one TFP-188, benchmarks/2026-10-13-tfp188/runs-hosted/with-bugs-standard-records.json Real, a finding
R20 with-bugs Toolshop TFP-183’s tab-order check on /#/contact (positive tabindex 1 to 7) and /#/auth/login (1 and 3, and Tab going from the submit back to the password) Read in the source: contact.component.html has 7 tabindex attributes and login.component.html 2; clean has none on either page. Not in Testsmith’s bug list (bug 53 is the registration form’s) benchmarks/2026-10-12-phase41-batch1/runs/tfp183-with-bugs-*.json Real; with-bugs only, outside the answer key

Unclear

# Site Test Evidence Source run Status
U1 test-assist TFFU-036 :visited link styling The stylesheet defines no :visited rule; the page has two links, both external. A real observation, but whether it matters is a judgement call TFP-040, examples/testforge-app/test-assist/ Recorded as a real but unclear finding
U2 practicesoftwaretesting.com “visited style configuration” on 5 pages No :visited rule in 1,826 inspectable same-origin rules, as U1 TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Recorded as a real but unclear finding
U3 practicesoftwaretesting.com “visual representation: links are distinguishable from surrounding text” on /category/power-tools The footer’s “explore” card links (.ptt-explore-link) have their parent’s colour and no underline, but are block links with no surrounding text. Timing-dependent: the run failed this page and passed 4, a later probe found them on 3 others and not this one TFP-150, benchmarks/2026-10-06-tfp150/toolshop-after-report.json (and -before-, d90de57) Unclear: the finding, and the check’s race with the footer’s render. TFP-154 (2026-10-06): the same check failed on 4 of the with-bugs Toolshop’s 5 Standard pages; its message names no link, so which links were not read
U4 with-bugs Toolshop “form element check: Your password * [Login] [input]” on /#/auth/login: “not editable … Element is not an ” Two elements carry id="password": the app-password-input wrapper and the input inside it. [id="password"] lands on the wrapper. The duplicate id is real (axe lists duplicate-id-aria for manual review on register); the check’s message blames the field, which is editable TFP-154, benchmarks/2026-10-06-toolshop-known-bugs/runs/with-bugs-standard-records.json Also on /#/auth/register (“form element check: Your password * [Password] [input]”, TFP-175, benchmarks/2026-10-07-tfp175-selection/runs/), the same two elements Unclear: a real duplicate id, reported as the wrong fault
U5 practicesoftwaretesting.com “link types: … classified correctly” on the clean product page: getByRole('link', { name: 'Hammer More information' }) not found A product card link named as inspection saw it, a few minutes before the test, wasn’t there when the test ran. Not read back: an hour later the clean site’s catalogue had been reset and the run’s product (/product/01M4B76Y1E21FPWYSS4K07YSMB) drew no product at all, so its cards can’t be compared. The same check failed the same way in the re-score on “Combination Pliers ABCDE$14.15” TFP-175, benchmarks/2026-10-07-tfp175-selection/runs/clean-standard-* Unclear: the cards’ contents shift between inspection and test
U6 practicesoftwaretesting.com “link health: internal, external, and download links do not return broken responses”: the footer’s GitHub link answered HTTP 500 Once, on one of 8 pages; the same link passed on the other 7 and in the re-score. GitHub’s answer, not the Toolshop’s TFP-175, benchmarks/2026-10-07-tfp175-selection/runs/clean-standard-* Unclear: an external host’s passing error
U7 practicesoftwaretesting.com “orphan page approximation” on all 5 pages of the re-score: “/product/01M48AGCWN26TFYQ1FQ6S7FNJF is not linked from any other inspected page” Read from the run’s profile: home and /category/hand-tools each listed 9 products, none of them this one. It is TFP-154’s hand-picked product, fixed by id on 2026-10-06; the clean catalogue is reset (the benchmark’s known limit), so by the re-score its page was reached only by its address. The check is right about the data it had TFP-178, from benchmarks/2026-10-06-toolshop-known-bugs/runs-tfp167/clean-standard-* Unclear: a benchmark artifact, not the site’s or the template’s
U8 Toolshop, local copies “page weight stays under 5 MB” (6.8 MB with-bugs, 5.7 MB clean) and “publishes robots.txt” / “sitemap.xml” (HTTP 404), on the local copies only The local copies run Angular’s development server (ng serve): unminified bundles (the largest, FontAwesome’s icons, 1,013 KB) and no robots.txt or sitemap.xml. The hosted sites are production builds, so neither shows there. Not the site’s behaviour TFP-180, benchmarks/2026-10-07-phase41-local/runs/ Unclear: the local environment, not the site
U9 with-bugs Toolshop, local Standard’s “link health: linked routes draw a page, not the app’s "not found" view” on three pages, as a finding: “5 linked routes not checked: the check stops after 14 s (/#/product/5, …)” The route check’s own time budget: it names the hash routes it had no time to check. A note about the check’s coverage, not a claim about the site; with-bugs only because clean routes by path. Not seen at Gate A, where the same pages drew faster TFP-188, benchmarks/2026-10-13-tfp188/runs/with-bugs-standard-records.json Open: a coverage note, reported as a finding. Also on hosted with-bugs in TFP-188 (runs-hosted/with-bugs-standard-records.json, three pages)
U10 with-bugs Toolshop, local Standard’s “cumulative layout shift during load stays under 0.1” on /#/, as a finding: “CLS 0.100 … needs improvement” Measured once, in a lab, on a development build, exactly at the band’s edge; the shift is the navbar settling. Not at Gate A on the same page, and not on clean TFP-188, benchmarks/2026-10-13-tfp188/runs/with-bugs-standard-records.json Open: a lab reading at a band edge
U11 with-bugs Toolshop, hosted Five older Standard checks timed out (30 to 42 s each): “keeps page title stable for /#/”, “keeps meta description stable for /#/”, “session cookies set by the site expire and stay scoped to it”, “storage failure injection: sessionStorage on /#/chainsaws”, “accessible name and role: 3 [input]” The hosted site answered slowly during TFP-188’s run: the same checks passed on the local copies and in TFP-175’s hosted run, and together they cost 198 s, which took hosted with-bugs Standard to 1,816 s TFP-188, benchmarks/2026-10-13-tfp188/runs-hosted/with-bugs-standard-report.json Open: host speed, not the site’s behaviour
U12 practicesoftwaretesting.com “usability: primary actions are reachable without scrolling at desktop viewport”: “SearchX Search requires scrolling to reach on first paint” (905.3 px against 900) Five pixels past the check’s limit on the hosted clean home; it passed in TFP-175’s hosted run and on the local copy. A judgement at the edge of a threshold TFP-188, benchmarks/2026-10-13-tfp188/runs-hosted/clean-standard-records.json Open: at a threshold’s edge

What this set is, and isn’t

The clean Toolshop’s failures, labelled (TFP-178, 2026-10-07)

Decision 9’s condition 4 (Phase 40) could be read only against new failures, because the first score labelled with-bugs only. Every clean-site failure in TFP-154’s re-score (runs-tfp167/) and in TFP-175 (benchmarks/2026-10-07-tfp175-selection/runs/) was read by hand from the committed records, then given a label by label-clean.mjs (in TFP-175’s folder), which fails if any failure is left without one. No new runs.

Every one of the 122 clean-site failures has a label. By verdict:

R10+R16 is one scan failing on both rules, on register.