{
  "site": "https://wagtail.org",
  "packetSHA256": "e18c88928aba9a129342a6e2de378fab5301a4abe4f82d75054ce725737579fc",
  "reviewer": {
    "harness": "Codex native subagent",
    "model": "gpt-6.1-sol",
    "reasoning": "high"
  },
  "rows": [
    {
      "checkId": "distraction",
      "guideline": "2.5",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "probes.json:journeys:0"
      ],
      "rationale": "The seven public tasks present relevant CMS information, article/search results, policy text and contact routes. Conference/newsletter prompts are static, and initially recorded dialogs and the Copied alert are hidden rather than unsolicited popups. The home video is paused, controls=true, autoplay=false and preload=none; it creates no observed playback interruption. The cookie prompt is a choice prompt rather than an engagement trap, and the opt-out/search journey reaches the policy. This is bounded observation, not every later interaction.",
      "confidence": "medium",
      "nextEvidence": "Review longer sessions and opened menus/embedded-player states if extending the sample."
    },
    {
      "checkId": "choice",
      "guideline": "2.6",
      "outcome": "fail",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13",
        "source.json:files:11",
        "source.json:files:12"
      ],
      "rationale": "All seven initial networks send analytics page-view POSTs to Google Analytics before an explicit cookie choice: /g/collect with en=page_view and /j/collect with t=pageview for two different UA properties. Yet the visible prompt offers Opt in/Opt out specifically to measure website use. In all fourteen staged runs GTM, gtag and analytics.js already transfer in the initial phase; either choice adds zero bytes in the bounded after_consent phase. The measured sequence contradicts a meaningful opt-in to those observed measurement requests. This does not establish cookie storage, all collection semantics, visual parity or legal noncompliance. Candidate cookie code records a preference but is not deployment proof. Both Opt in and Opt out actions completed on each of the seven separate branch pairs; the executed opt-out/search journey also completes. This credits available choices without claiming visual equality, storage verification or a causal difference between independent initial branch totals.",
      "confidence": "high",
      "nextEvidence": "Inspect fresh-profile storage and request payloads across no-choice, accept, reject, reload and withdrawal states. Align optional analytics initialization with the declared choice and verify rendered prominence."
    },
    {
      "checkId": "comprehension",
      "guideline": "2.7",
      "outcome": "fail",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:discovery:0",
        "disclosures.json:sites:0"
      ],
      "rationale": "The sustainability page gives 5,173 tCO2e/year for 6,431 sites together with a 12 tCO2e/year per-site average. The displayed total/count imply approximately 0.8044 tCO2e/site/year, while 12 \u00d7 6,431 is 77,172 tCO2e/year. Without a separately labeled population/statistic, these quantities cannot describe the same mean; this is a demonstrated contradiction for the declared task of understanding the sustainability policy. Other content explains CMS capabilities, next steps, preview status and contact purposes reasonably. Privacy nevertheless describes the Torchbox website/company while its metadata calls it the Wagtail.org policy, leaving service/controller scope unclear. That uncertainty and unrepresented-user comprehension remain unresolved, but do not override the demonstrated numerical failure. Heading inventories truncate before visible primary content, so no broken heading hierarchy is inferred.",
      "confidence": "high",
      "nextEvidence": "Reconcile the mean, unit, population and reporting period against the underlying 2025 dataset; about 0.80 tCO2e (804 kg) per site applies only if the displayed total/count define the intended mean. Clarify the wagtail.org privacy service/controller scope and actual service inventory, then review the corrected public copy and user understanding."
    },
    {
      "checkId": "consistency",
      "guideline": "2.8",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "All seven pages repeat Project, Explore, Developers, Community, Support, search and Get started controls, the same footer groups, Switch theme control and cookie option wording. Content types appropriately add search/pagination or article navigation. This observed naming/navigation consistency supports the sampled interface pattern, without proving a documented design system or consistency of all opened menus.",
      "confidence": "high",
      "nextEvidence": "Compare opened menus, mobile states and additional content types for broader coverage."
    },
    {
      "checkId": "motion",
      "guideline": "2.10",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "No animation is active in the initial desktop snapshots; the home native video is paused and user-controlled. After resize, Home/Features/Sustainability report 5/2/1 running animations, and the same counts remain in reduced_motion with one iteration each. Print transitions create additional transient animation records. The packet records generic targets and current time, but no animated properties, duration, purpose or controls; running entries alone do not establish harmful, decorative or uncontrolled motion. Child-frame playback is not traversed.",
      "confidence": "medium",
      "nextEvidence": "Inspect the actual animated properties, durations and purpose in resize/menu/scroll states, then compare settled and cold reduced-motion behavior and embedded-player controls."
    },
    {
      "checkId": "notifications",
      "guideline": "2.12",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "Every recorded page/state shows zero notification permission requests. Initial dialogs are hidden; the only additional home alert is a hidden Copied confirmation tied to a named copy action. Newsletter signup is optional and no notification is sent during the reading tasks. Browser permission is already denied, so this is bounded absence of attempted permission requests, not all external messaging/subscription behavior.",
      "confidence": "medium",
      "nextEvidence": "Use default permission and a deliberately requested signup flow if expanding notification scope."
    },
    {
      "checkId": "documents",
      "guideline": "2.13",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "The sampled visitor tasks are served as HTML and record no download-attributed or PDF/office/text-file links. Print states hide global navigation, footer/choice furniture and unrelated article recommendations while retaining policy/contact/article reading content; the article retains eight of eleven images after three related author images are removed. This passes print visibility and applicable labels in the declared sample only, not optimized pagination, paper savings or document accessibility. No uninspected downloadable document is claimed equivalent.",
      "confidence": "medium",
      "nextEvidence": "Inspect print preview/PDF pagination for long feature and release pages if judging actual paper use."
    },
    {
      "checkId": "unused-code",
      "guideline": "3.2",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "source.json:files:8",
        "source.json:files:9",
        "source.json:files:14"
      ],
      "rationale": "The common deployed main script is about 39.5 kB decoded/11.3 kB gzip and CSS about 125.7 kB decoded/21.1 kB gzip. Initial coverage records many unexecuted JavaScript functions, but they include later menu/search/copy interactions; CSS entries are used-rule observations, not a whole unused-byte measure. Article coverage is explicitly scriptsTruncated and child-frame incomplete. Candidate Webpack production mode and cssnano indicate intended minification, but deployed identity is unverified and do not justify deleting unexecuted code. Much larger third-party analytics/player scripts are concrete investigation targets. The CSS rule-usage list contains observed used ranges rather than every stylesheet rule. The separate GTM/gtag/analytics.js scripts account for 347.162 kB in initial home reject, not part of the first-party bundle; deletion still requires owner measurement requirements and deployed attribution.",
      "confidence": "high",
      "nextEvidence": "Collect journey-wide coverage and deployed bundle/source-map attribution, including menus, search, copying, consent and video activation, before removing code."
    },
    {
      "checkId": "components",
      "guideline": "3.3",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:2",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "source.json:files:10",
        "source.json:files:15"
      ],
      "rationale": "The home native player is correctly user-activated with preload=none and no .mp4 request in staged observations. In contrast, the release article initializes a below-content YouTube iframe before playback is requested, loading embed/player code, fonts, thumbnail and ad/log endpoints. Observed accounted initial bytes are 2,025,844/2,031,558 in reject/accept runs; both captures are partial with unresolved transfers and child-frame gaps, so these are lower bounds, not complete page weights. In initial reject, captured YouTube and associated player-host traffic totals 1,328,308 accounted bytes; this is neither the complete frame footprint nor a validated saving. The article text supports reading without a played video, making a click-to-load facade a strong candidate. However, a usable accessible activation alternative and player task requirements have not been tested, so safe splitting/deferment and exact savings are unresolved.",
      "confidence": "high",
      "nextEvidence": "Prototype a named thumbnail/play facade that loads the existing player on request; validate keyboard activation, captions, transcript/fallback and playback before comparing fresh-load transfer."
    },
    {
      "checkId": "layout",
      "guideline": "3.10",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "probes.json:journeys:0"
      ],
      "rationale": "All seven desktop documents are 1290 px wide in a 1280 px viewport and narrow documents are 370 px in a 360 px viewport: a consistent 10 px horizontal overflow. Task text and image counts remain; images generally shrink from 660/440 px to 300 px and the mobile search/menu controls become visible. Recorded visible link/control/image boxes do not identify the overflow source. Only desktop search was activated; mobile menus and affected controls were not exercised. This is an observed overflow requiring diagnosis, not demonstrated content loss or real-device failure. Fresh no-JavaScript and print states have no horizontal overflow, which helps bound diagnosis but does not identify the cause.",
      "confidence": "high",
      "nextEvidence": "Identify the overflowing deployed element, inspect clipping/scrolling, activate narrow navigation/search, then repeat the declared tasks with keyboard/touch and a real device."
    },
    {
      "checkId": "impact-choice",
      "guideline": "5.4",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:2",
        "probes.json:pages:4"
      ],
      "rationale": "Most sampled choices are ordinary reading/navigation and do not need carbon labels. Optional video activation is a consequential resource choice: the home native player defers downloading, whereas the article initializes its third-party player before activation. The packet does not inspect embedded quality/duration/data-use choices or their contextual information. The sustainability page explains image-format and OLED-specific theme tradeoffs, but those describe CMS use rather than an executed upload workflow on this website.",
      "confidence": "medium",
      "nextEvidence": "Inspect user-facing duration/quality/data-use context for optional video and test a deferred player alternative; do not prescribe labels for every control."
    },
    {
      "checkId": "journeys",
      "guideline": "2.4",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:journeys:0",
        "probes.json:pages:3",
        "probes.json:pages:4"
      ],
      "rationale": "The declared task opts out, opens desktop search, enters sustainability, submits and follows the policy result; all five actions complete with no errors. Final URL/title/primary-heading postconditions identify Sustainability, and the after text actually contains the policy, mitigation examples, ecosystem estimate and explicit Wagtail.org scope. The cookie prompt is no longer in the after text. The separate results sample places the policy first among 62 results. This confirms the executed task, not optimal efficiency or whether consent controlled all network behavior.",
      "confidence": "high",
      "nextEvidence": "Repeat with new visitors and narrow navigation; independently test consent enforcement rather than inferring it from banner dismissal."
    },
    {
      "checkId": "media-delivery",
      "guideline": "2.9",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13",
        "source.json:files:10",
        "source.json:files:15",
        "source.json:files:16"
      ],
      "rationale": "Images are task-relevant screenshots, illustration, people/logos and case-study material, with actual WebP delivery for many .png/.jpg URLs. Several page types already lazy-load images; the home video uses preload=none. None of the sampled images provides srcset/sizes. The home rocket illustration is 2234 px natural width, rendered 1280 px desktop/360 px narrow, and its response is 370,360 bytes despite WebP delivery. Feature screenshots are 1300 px natural versus 660/300 px, while article screenshots are 800\u2013900 px versus 660/300 px and eagerly loaded. Some lazy images have naturalWidth=0 because not yet loaded. Candidate dimensions and bytes nominate responsive/deferred-delivery work, but no alternate rendition or text legibility at intended density was visually checked. Actual scroll timing corroborates useful deferral: home adds three image requests (163,394/163,391 additional accounted bytes across reject/accept), Features two (80,608/80,458), and Sustainability one (42,945/42,950). These are cumulative checkpoint deltas, not independent loads or isolated savings. Mixed delivery remains: two article screenshots are actually PNG; four sustainability .webp resources render with dimensions but return application/octet-stream. Their MIME correctness and alternative formats need inspection rather than a broken-image or blanket-conversion claim. The 370,360-byte hero value above is response content length; initial reject accounts for 371,314 received bytes including reported overhead within an incomplete total. All views use DPR 1 and narrow is the same resized document, not a cold mobile measurement.",
      "confidence": "high",
      "nextEvidence": "Compare responsive hero and screenshot variants at 1x/2x with content owners; preserve legibility, art direction and dimensions. Test cold narrow transfer and lazy-loading behavior before claiming savings. Inspect the application/octet-stream responses and the two actual PNG screenshots separately; use response MIME and visual quality, not filename extensions."
    },
    {
      "checkId": "font-delivery",
      "guideline": "2.11",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13"
      ],
      "rationale": "The sampled English body uses Kumbh Sans with a sans-serif fallback. FontFace records declare weights 400/600/700: regular uses display=block, other weights auto; 400 and 700 load on all pages, and 600 additionally loads on the article. Two local WOFF2 files transfer 19,640 and 19,304 encoded bytes, with an extra 20,404-byte face on the article. These variants can serve normal/emphasized content; a web font is not itself a defect. But blocked/constrained font loading and required language coverage are not observed, so fallback readability and potential subsetting/removal cannot be established. Staged received accounting includes transport overhead, so the two ordinary-page local faces account for about 39.5\u201340.8 kB; the article local faces account for 61.172 kB in initial reject. Its additional 40.678 kB player Roboto transfer is separate from main-document typography.",
      "confidence": "high",
      "nextEvidence": "Test slow/blocked font responses with the intended languages and inspect weight use before considering display or subsetting changes."
    },
    {
      "checkId": "semantic-forms",
      "guideline": "3.6",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "All seven documents report lang=en, one main and one H1. Task controls have meaningful text, explicit names or label relationships: both search forms, newsletter Email, cookie choices, theme and contact links. The search-page input has a Search label even though keyboard trace text is empty. However, heading lists are truncated at 100 entries by duplicated hidden navigation headings, before any visible H1; count presence cannot establish its meaningful text or the main content hierarchy. Dialog naming/focus activation is also untested. No missing accessible name is inferred from textless controls alone.",
      "confidence": "high",
      "nextEvidence": "Capture visible main-region headings and resolved naming relationships directly; execute modal and form tasks before expanding to semantic/WCAG claims."
    },
    {
      "checkId": "metadata",
      "guideline": "3.8",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "The six content pages have descriptive task-specific titles, HTTPS canonicals matching their URLs and relevant descriptions. Search uses a query-specific results title and robots=noindex; canonical/description absence is appropriate for this dynamic search page. Home and the release article declare JSON-LD objects that parse, including WebPage/Organization or BlogPosting. The other pages declare none; optional structured data is judged by purpose rather than a mandatory-field checklist. This is parseability/useful metadata, not schema validation or search eligibility.",
      "confidence": "high",
      "nextEvidence": "Validate declared schema meaning only if expanding beyond the sampled metadata check."
    },
    {
      "checkId": "preference-behavior",
      "guideline": "3.9",
      "outcome": "fail",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "Every page starts with dark text rgb(38,38,38) on white. Changing the system preference in-page makes dark=true but leaves those colors and light-theme text unchanged on all seven pages. Fresh cold_dark contexts instead render white text on rgb(38,38,38), retaining task text. Initial preference support therefore works while live preference change fails the declared expectation. Reduced-motion states also retain transient animation records on three pages, but their purpose/properties are insufficient to independently judge motion respect. No energy saving is inferred from color alone.",
      "confidence": "high",
      "nextEvidence": "Verify unset manual overrides and update live system-theme listeners while preserving explicit choices; test both directions, persisted preferences and reduced-motion transitions."
    },
    {
      "checkId": "no-js-content",
      "guideline": "2.17",
      "outcome": "fail",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "Core article/features/policy/search reading and navigation remain present without JavaScript; most text differences are omitted theme controls and cookie prompt. Contact is a concrete exception: hello@wagtail.org (general/conduct) and security@wagtail.org become [email protected] links to /cdn-cgi/l/email-protection instead of readable mailto addresses. The declared task is finding the right contact route, including security; generic FAQ/Slack links do not provide an explicit equivalent security-email fallback. This fails the sampled core content expectation, not all no-JavaScript behavior.",
      "confidence": "high",
      "nextEvidence": "Keep readable contact/security addresses and working mailto links in baseline HTML or provide a tested explicit equivalent that needs no script; verify with scripts disabled."
    },
    {
      "checkId": "api-requests",
      "guideline": "3.11",
      "outcome": "cantTell",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13"
      ],
      "rationale": "Each initial page makes one small /sitewide_alert/ JSON request plus three analytics page-view requests. The two /j/collect calls use different UA property IDs, and /g/collect uses a GA4 ID, so URL/path similarity alone is not proof of duplicate events safe to remove. The article also shows repeated YouTube/ad/log requests, some repeated scripts with zero subsequent received bytes; these may involve cache/frame initialization rather than repeated transfer. No task freshness requirements, destination purposes or backend cost evidence justify all observed analytics/player work. Repeated player resource entries with zero accounted received bytes are not assigned missing traffic or treated as evidence of repeated backend work. No long-duration polling test was performed.",
      "confidence": "high",
      "nextEvidence": "Inventory the three analytics destinations and owners, inspect needed events and consent state, and trace repeated player initialization. Remove obsolete/redundant work only after preserving required measurements and playback."
    },
    {
      "checkId": "public-cache-policy",
      "guideline": "4.2",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13"
      ],
      "rationale": "Sampled hashed JS/CSS/WOFF2 assets expose max-age=315360000, public, immutable and validators. Remote media exposes max-age=31536000 plus ETag/Last-Modified; /sitewide_alert/ has explicit short client/longer shared caching. The contact decoder has public caching with Expires. Dynamic search and analytics/log responses legitimately use no-cache/no-store and are not reusable static resources; data URLs, zero-byte diagnostic endpoints and errors do not require the same policy. This passes observed public policy only, without asserting warm reuse or offline availability.",
      "confidence": "high",
      "nextEvidence": "Compare repeat visits and cache invalidation through content changes before claiming measured reuse or offline support."
    },
    {
      "checkId": "text-compression",
      "guideline": "4.3",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1",
        "staged.json:runs:2",
        "staged.json:runs:3",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "staged.json:runs:6",
        "staged.json:runs:7",
        "staged.json:runs:8",
        "staged.json:runs:9",
        "staged.json:runs:10",
        "staged.json:runs:11",
        "staged.json:runs:12",
        "staged.json:runs:13"
      ],
      "rationale": "Across all seven sampled page responses, HTML has Brotli and common CSS/JS gzip; material third-party text assets likewise carry compression. Main CSS is 21,060 encoded bytes versus roughly 125.7 kB decoded, and main JS 11,338 versus roughly 39.5 kB decoded. Unencoded examples are tiny 2-byte JSON alerts, 3-byte analytics acknowledgments, 29-byte helper scripts or empty responses, not material compression omissions. Error responses and already-compressed WOFF2/images are not required to be compressed again. Partial accounting limits totals, not observed encoding headers.",
      "confidence": "high",
      "nextEvidence": "No further evidence needed for sampled material text encoding."
    },
    {
      "checkId": "keyboard-path",
      "guideline": "3.6",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "rationale": "All seven bounded Tab traces reach the event link, visible skip link, named home link, meaningful primary-navigation buttons, named search and Get started, then page task controls/links. Every recorded focus target is visible and focusVisible=true with a 1 px automatic dark outline on the light sampled background. The textless search input has a Search label in control evidence. This supports the initial path and indicator only, not activated menus, skip-link destination, reverse traversal, traps or full accessibility.",
      "confidence": "medium",
      "nextEvidence": "Activate skip/search/menu controls and inspect full forward/reverse paths and focus in narrow/dark states."
    },
    {
      "checkId": "disclosure-scope",
      "guideline": "2.1",
      "outcome": "pass",
      "evidenceIds": [
        "staged.json:discovery:0",
        "disclosures.json:sites:0",
        "probes.json:pages:4"
      ],
      "rationale": "The retrieved policy explicitly names Wagtail.org, docs.wagtail.org and guide.wagtail.org as projects whose impacts/emissions it intends to reduce, alongside CMS/ecosystem and community-event actions. The linked Digital emissions statement specifically scopes homepage/website estimates, states SWD v4, measurement method/period/grid factors, separates site from ecosystem and describes monitoring alerts. Thus service relationship, impacts and mitigation are disclosed. This does not verify execution, footprint accuracy or the policy arithmetic; the 12 t/site mismatch remains a separate claim-quality concern. The retrieved 2025 ecosystem article warns that estimated pageview changes drive much of the headline reduction. Its 2024 baseline is unclear: 6,327 tons appears in opening copy versus 6,260 in the year list; median page weight also differs from the earlier report. Different sample, method or revision may explain some differences, so these are unresolved comparability questions rather than an asserted arithmetic error. Digital emissions is an attributed April 2026 mobile/Lighthouse homepage scenario with different inputs from the present partial desktop CDP evidence; its estimate/rating and annual range are not independently reproduced or endorsed here.",
      "confidence": "high",
      "nextEvidence": "Correct the displayed total/average relationship, document the comparable 2024 baseline/sample/method revisions and provide execution records if evaluating actual reductions. To reproduce the separate site/homepage estimate, capture first/repeat mix, dated traffic counts, page weighting, infrastructure and complete model inputs; keep the 30\u2013300 kg/year range a methodological scenario range, not statistical bounds."
    },
    {
      "checkId": "carbon-discovery",
      "guideline": "3.13",
      "outcome": "pass",
      "evidenceIds": [
        "staged.json:discovery:0",
        "disclosures.json:sites:0"
      ],
      "rationale": "Root /carbon.txt returns 200 text/plain and parses with declared version 0.4, exact-host disclosure relationships and usable URLs; there is no redirect/delegation ambiguity. All four declared disclosure URLs are retrieved with 200. Review reports no structural issues under its explicit v0.x common-field profile. schemaValidated=false means this is not full 0.4 schema or IETF 1.0-RC1 conformance. The 2023/2024 reports are marked past declared validity and remain historical evidence; they are not silently promoted to current certifications. Provider entries are self-declared relationships only. The last_updated value is 2026-01-07. A 2025 report is retrieved through linked content, rather than declared directly in carbon.txt; exact-host labels do not automatically extend its declared domain relationships to subdomains.",
      "confidence": "high",
      "nextEvidence": "Consider adding the current 2025 report and refreshing last_updated; use a version-specific schema validator for a formal syntax-conformance claim."
    },
    {
      "checkId": "hosting-evidence",
      "guideline": "4.1",
      "outcome": "cantTell",
      "evidenceIds": [
        "staged.json:discovery:0",
        "disclosures.json:sites:0",
        "source.json:files:4",
        "source.json:files:7"
      ],
      "rationale": "carbon.txt self-declares Cloudflare CDN, Heroku platform/database/cache, AWS object storage and GitHub CI/CD. Digital emissions uses an Ireland data-center carbon factor; candidate README says Heroku hosts the site. These identify claimed relationships but do not independently attest account/region/service deployment or renewable-electricity/operational metrics for this origin. The source revision is clean but deploymentVerified=false. No independently corroborated sustainable-hosting judgment or SWDM green flag follows.",
      "confidence": "high",
      "nextEvidence": "Obtain current origin/provider/region attestations and independent provider environmental evidence with matching operational scope."
    },
    {
      "checkId": "source-reuse",
      "guideline": "3.4",
      "outcome": "pass",
      "evidenceIds": [
        "source.json:files:9",
        "source.json:files:10",
        "source.json:files:11",
        "source.json:files:12",
        "source.json:files:13",
        "source.json:files:14"
      ],
      "rationale": "In clean candidate revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, home_page.html calls shared hero/button patterns and renders CTA items through one included button component; main.js uses one initComponent routine for selector-based CookieMessage, SiteSearchDesktop and other instances. Cookie/search behavior lives in dedicated modules rather than repeated handlers in the selected page. This supports selected-source reuse, not whole-project efficiency. Home extends unselected base_page.html, so no complete inheritance/deployed chain is claimed; deploymentVerified=false.",
      "confidence": "medium",
      "nextEvidence": "Inspect additional concrete template call sites only if expanding source-wide duplication review."
    },
    {
      "checkId": "source-dependencies",
      "guideline": "3.12",
      "outcome": "cantTell",
      "evidenceIds": [
        "source.json:files:0",
        "source.json:files:1",
        "source.json:files:2",
        "source.json:files:3",
        "source.json:files:4",
        "source.json:files:6"
      ],
      "rationale": "Candidate pyproject.toml/package.json inventory direct dependencies. Docker/CI use npm ci and Poetry install, with lockfiles explicitly selected. However, poetry.lock and package-lock.json expose metadata_only identity/size, not resolved contents or manifest consistency. Intended repeatable installation is visible but valid resolution cannot be established from filenames. Necessity, vulnerabilities, installed/deployed versions and ongoing updates remain outside scope; clean candidate revision is not deployment-verified. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "high",
      "nextEvidence": "Capture bounded lockfile/manifest consistency or clean-install receipts at this revision, then separately verify deployed versions where needed."
    },
    {
      "checkId": "source-runtime",
      "guideline": "3.15",
      "outcome": "pass",
      "evidenceIds": [
        "source.json:files:0",
        "source.json:files:4",
        "source.json:files:5",
        "source.json:files:6",
        "source.json:files:7"
      ],
      "rationale": "Selected candidate Python declarations agree on 3.14: pyproject ^3.14, Docker python:3.14-trixie and both CI setup-python paths 3.14. Node is 24 in .nvmrc, frontend Docker and development NodeSource setup; CI reads .nvmrc. Docker and CI also agree on Poetry 2.4.0. This passes selected declaration consistency only. README references additional unselected Vagrant paths, deploymentVerified=false, and no authoritative release reference establishes latest-stable/deployed versions. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "high",
      "nextEvidence": "Capture any other supported deployment/runtime paths and deployed inventory if extending the consistency scope."
    },
    {
      "checkId": "source-automation",
      "guideline": "4.6",
      "outcome": "pass",
      "evidenceIds": [
        "source.json:files:6"
      ],
      "rationale": "The selected clean candidate CI declares push triggers only for main/next and pull_request; workflow-level concurrency groups by workflow/ref and cancel-in-progress=true to stop superseded same-ref runs. Node installation is conditionally reused with lock/runtime-keyed caches and tests depend on lint. These are concrete filtering/cancellation controls, with no demonstrated duplicate same-context path in the frozen packet. This does not establish every job is necessary, rule out cross-event overlap in a specific next-branch PR, measure CI energy or attest current execution/deployment. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "medium",
      "nextEvidence": "Inspect run history and pre-commit contents if investigating actual redundant validation; preserve necessary lint/build/backend checks."
    },
    {
      "checkId": "source-maintenance",
      "guideline": "5.10",
      "outcome": "pass",
      "evidenceIds": [
        "source.json:files:7",
        "source.json:files:6"
      ],
      "rationale": "Selected candidate README explains Docker/Vagrant setup, migrations, cache-table setup, access-controlled data/media retrieval, frontend watch/lint/format commands and production/staging deployment. CI gives explicit system/migration/test commands. This supports the narrow documentation disclosure for maintaining the service; it does not attest staffing, cadence, security/sustainability execution or correctness of every unselected auxiliary path. Source deployment remains unverified. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "high",
      "nextEvidence": "Obtain owner maintenance cadence and operational records for the wider strategy."
    },
    {
      "checkId": "source-change-docs",
      "guideline": "5.12",
      "outcome": "pass",
      "evidenceIds": [
        "probes.json:pages:2"
      ],
      "rationale": "The live dated Wagtail 8.1 release article identifies the version and user-facing API, workflow, metadata and multilingual-sitemap changes, distinguishes prototype/preview status, explains upgrade/deprecation implications and links full release notes. It therefore communicates the declared visitor task\u2019s product evolution, independently of candidate source deployment. This is public Wagtail product change documentation, not a comprehensive history of the wagtail.org implementation or proof every affected user receives it. Any candidate-source corroboration remains limited to clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, deploymentVerified=false; this judgment rests on the live product announcement.",
      "confidence": "medium",
      "nextEvidence": "Capture a site-specific changelog if the question is changes to the wagtail.org application itself."
    },
    {
      "checkId": "source-budgets",
      "guideline": "5.24",
      "outcome": "cantTell",
      "evidenceIds": [
        "source.json:files:1",
        "source.json:files:6",
        "source.json:files:7",
        "source.json:files:8",
        "staged.json:discovery:0",
        "disclosures.json:sites:0"
      ],
      "rationale": "Selected Webpack/build/CI/docs contain no explicit relevant asset/performance ceiling plus exceedance action. A development performance stats display option and linters are not a concrete budget. The public Digital emissions page says weekly DebugBear monitoring/alerts keep page weight within a target, but supplies no target value or captured alert configuration. Thus relevant enforcement is claimed but not reviewable; absence from the selected files is not organization-wide absence. Candidate deploymentVerified=false. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "high",
      "nextEvidence": "Capture the owner-approved numeric budget, scope/device protocol, alert or build configuration and action when exceeded. Include the concrete DebugBear target, response owner and alert/history evidence; do not infer operational execution solely from the public statement."
    },
    {
      "checkId": "source-open-source",
      "guideline": "5.25",
      "outcome": "cantTell",
      "evidenceIds": [
        "source.json:files:0",
        "source.json:files:1",
        "source.json:files:7",
        "probes.json:pages:5"
      ],
      "rationale": "Selected manifests declare BSD-3-Clause for the Python project and ISC for package metadata, and README links the public repository and explains development. No actual license or contribution-policy document is included, so reuse terms and the contribution process for this codebase cannot both be inspected. The live Privacy page restricts site content reuse; describing Wagtail CMS as open source does not automatically license this website\u2019s code/content. This is a selection gap, not proof those documents do not exist; candidate deploymentVerified=false. Selected source identity: clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609, local/wagtail.org; deployed artifact identity is not verified.",
      "confidence": "high",
      "nextEvidence": "Capture the codebase LICENSE and contribution instructions, reconcile package-metadata scope, and distinguish website content terms from software licensing."
    }
  ],
  "topFindings": [
    {
      "checkId": "choice",
      "priority": "high",
      "title": "Make the measurement choice effective before analytics initializes",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6",
        "staged.json:runs:0",
        "staged.json:runs:1"
      ],
      "summary": "All seven initial page networks send GA4 and two UA page-view requests while the Opt in/Opt out measurement prompt is still present; analytics scripts already transfer before either branch action.",
      "recommendedAction": "Audit current tag/container configuration and fresh consent states, gate optional measurement as declared, then verify no-choice/reject/accept/reload/withdrawal behavior. This is an observed request-sequence finding, not a legal compliance verdict.",
      "status": "observed defect"
    },
    {
      "checkId": "components",
      "priority": "high",
      "title": "Test a click-to-load facade for the release article player",
      "evidenceIds": [
        "probes.json:pages:2",
        "staged.json:runs:4",
        "staged.json:runs:5",
        "source.json:files:15"
      ],
      "summary": "Before a play action, the release article initializes YouTube with player scripts, font, thumbnail and ad/log requests. Initial reject captures 1.328 MB of related accounted traffic within 2.026 MB of page traffic; both are partial observations with child-frame limits, not complete footprints or validated savings.",
      "recommendedAction": "Prototype an accessible named preview that loads the player on activation, preserve captions/transcript and playback, and measure actual savings after validation.",
      "status": "investigation"
    },
    {
      "checkId": "no-js-content",
      "priority": "medium",
      "title": "Keep contact and security email routes available without scripts",
      "evidenceIds": [
        "probes.json:pages:6"
      ],
      "summary": "The fresh no-JavaScript Contact state replaces readable general/conduct/security email addresses and mailto links with [email protected] and Cloudflare decoder links.",
      "recommendedAction": "Render the addresses and mailto targets in baseline HTML, or provide an explicit equivalent security/contact fallback that works without JavaScript.",
      "status": "observed defect"
    },
    {
      "checkId": "comprehension",
      "priority": "medium",
      "title": "Correct the footprint summary and clarify privacy scope",
      "evidenceIds": [
        "probes.json:pages:4",
        "probes.json:pages:5",
        "staged.json:discovery:0"
      ],
      "summary": "The policy pairs 5,173 tCO2e/year and 6,431 sites with a 12 tCO2e/site/year average; the total/count instead imply about 0.8044. The privacy page also describes a Torchbox-only scope despite being presented as the wagtail.org policy.",
      "recommendedAction": "Verify the intended 2025 mean, units and population; correct the summary or label any different statistic. Clarify the wagtail.org privacy/controller and actual service inventory. Explain the differing 2024 baseline values as a sample/method revision if supported, rather than assuming they are directly comparable errors.",
      "status": "observed numerical contradiction; privacy/baseline scope requires clarification"
    },
    {
      "checkId": "preference-behavior",
      "priority": "medium",
      "title": "Respond when the system theme changes on an open page",
      "evidenceIds": [
        "probes.json:pages:0",
        "probes.json:pages:1",
        "probes.json:pages:2",
        "probes.json:pages:3",
        "probes.json:pages:4",
        "probes.json:pages:5",
        "probes.json:pages:6"
      ],
      "summary": "Dark preference works on fresh loads but all seven open pages stay light after a live operating-system preference change.",
      "recommendedAction": "Handle live media-query changes while respecting explicit overrides, then test both directions and persisted choices.",
      "status": "observed defect"
    }
  ],
  "reviewNotes": {
    "reconciliation": "All 16 evidence and 14 claims issues have explicit resolutions. Comprehension changes from cantTell to fail for the demonstrated displayed quantity contradiction; unresolved privacy scope and unrepresented-user needs remain unknown. Disclosure-scope stays pass for service-specific public disclosure only. The responsive-image investigation remains in its full row but is replaced in the top five by the more immediately actionable numerical/privacy clarity work.",
    "measurementLimits": "All 14 staged visits and all 42 cumulative phases are partial. Ratings remain not_calculated; no new SWDM output, DCR grade, annualized scenario or measured energy saving is assigned. Null byte accounting is never zero-filled or replaced by Content-Length. Aborted Plausible 404, aborted GA4 204 and unresolved YouTube traffic are distinct failure/accounting observations. Repairing analytics is not a prerequisite for sustainability work.",
    "sourceLimits": "Only 18 selected candidate files at clean tracked revision 9fa65385f27c15b83c86a4f6e0e3ac18d4784609; deploymentVerified=false; lockfile bodies are unavailable. Local Plausible markup and live GTM/GA behavior are not assumed identical. No technology evidence ID is invented, PWA fingerprint is not offline proof, and provider names do not set green-hosting model inputs.",
    "sourceChangeDocumentation": "The release announcement pass is for the public Wagtail product changes relevant to the sampled task, not a complete wagtail.org implementation changelog.",
    "disclosureLimits": "Four long discovery excerpts cap at 3,000 characters. Unexposed later text is unknown. April 2026 mobile/Lighthouse estimates and ecosystem totals stay distinct from current partial desktop CDP traffic and this origin\u2019s annual footprint."
  }
}
