29 KiB
Business Logic Model - Unit 2: Display fallbacks
Unit 2 of the feature "prerendering": fade-in, no JavaScript, first photo (FR-PR-09 to 13, FR-PR-21; D6 to D9). It builds on unit 1 (hydration rule BR-HYD-01: the first browser render equals the HTML). Rules: business-rules.md. Data: domain-entities.md. Components: frontend-components.md.
Everything here was checked against web/src, the built web/dist/client, the Lighthouse reports of the measurement at the gate (after and revealvisible, mobile and desktop) and a real Chrome driven by Playwright against the local preview server, on 2026-10-31. No application code was changed.
1. Reveal
1.1 Lifetime of one Reveal
graph TD
Server["Prerender: settled, visible"] --> Html["HTML on screen, content visible"]
Html --> Hydrate["Hydrating render: settled, same markup"]
Hydrate --> Layout["Layout effect before paint: measure top edge"]
Layout --> Test{"Top at or below innerHeight"}
Test -->|"no"| Stay["Stays settled: no change at all"]
Test -->|"yes"| Hide["Set hidden: opacity 0, no transition"]
Hide --> Watch["Observer: threshold 0.12, margin 60 px"]
Watch --> Fire["Intersected once"]
Fire --> Fade["Revealed: fade 0.8 s after delay"]
classDef build fill:#63b3ed,stroke:#2b6cb0,stroke-width:1px,color:#000;
classDef state fill:#9ae6b4,stroke:#2f855a,stroke-width:1px,color:#000;
classDef choice fill:#f6e05e,stroke:#c05621,stroke-width:1px,color:#000;
class Server,Html,Hydrate build;
class Layout,Watch,Fire,Hide state;
class Stay,Fade state;
class Test choice;
Text alternative: the HTML and hydrating render are visible; a layout effect hides only elements wholly below the viewport before the next paint, and an observer reveals each once.
1.2 Why the layout effect and why top >= innerHeight
- Layout effect, not a passive effect. A passive effect runs after the browser painted. An element in the 60 px band at the bottom of the screen would be painted visible from the HTML, then hidden, which is the flash FR-PR-09 forbids. A layout effect runs after React committed the hydrated tree and before the next paint, and a state change inside it is applied synchronously before that paint. No
useLayoutEffectexists insrctoday; React 19.3 does not warn for it while prerendering. - Reads before writes. Every
Revealin the commit reads its rectangle in its own layout effect; the state updates are applied after all effects ran. The browser lays out once, not once perReveal. - The test is the viewport edge, not the 60 px line. The trigger line (
rootMarginbottom -60 px) is how late the fade starts while scrolling. At load the question is "was this painted on screen". Only an element that starts at or below the bottom edge was not. - No transition for the hide. The hidden style sets
transition: none. A transition is taken from the style after the change, so the step from settled to hidden does not animate. The transition properties exist only in the revealed style. - Reduced motion is read directly. Chakra's
useMediaQuery(BR-HYD-07) still returns its server-safefalsein that same layout effect, because its own effect has not run. ReadingmatchMediain the effect avoids a hide followed by an animated reveal.
1.3 Anything that breaks
All 21 call sites keep their props. What each does under the new rule at 412 x 823 and 1350 x 940 (probe of data-reveal and tops, dist build):
| Call site (file, count) | Wraps | Effect |
|---|---|---|
home/IntroGallery (2), home/Pricing (2), home/Testimonials (2), home/ContactSection (2), InstagramGrid (2) |
Headings, photo lists, cards | First Reveal of / is at top 919 px at 412 x 823: all hidden at load, fade on scroll, as now. At 1350 x 940 the first ones are in view and stay. Pricing stagger (delay) only shows on scroll (BR-RVL-06). |
MediaText (2: photo and text) |
Photo column and text column of /over-mij, /soorten-fotoshoots, shoot pages |
In view at the top of these pages: visible at once. Before: 1.0 s after the first paint (observed LCP 977 to 1047 ms at 412 x 823). |
PhotoMosaic (1, as="li" with delay), PhotoDuo (1), PortfolioGrid (1), ProcessTimeline (1), StoryBlocks (1), CallToAction (1), StorySection (1), routes/contact (2) |
Lists, grids, blocks | Same rule. PortfolioGrid is the tall one: see BR-RVL-13 and 1.4. routes/contact has one Reveal around display: none content from lg (zero rectangle, so it counts as in view and is settled). |
data-reveal consumers |
MediaText.test, ShootList.test, Reveal.test use closest('[data-reveal]') only |
Unchanged: the attribute stays, values visible and hidden. |
IntersectionObserver absent (jsdom, old browsers) |
Whole site | Settled and visible (BR-RVL-08); the existing Reveal.test case stays. |
Tests with a fake observer (useReveal.test) |
The hook | Five cases change (BR-DTS-01). No other test asserts hidden content. |
1.4 The /portfolio grid and an existing defect
Measured in Chrome at 412 x 823 on the built page, 2 s after load: the grid wrapper of /portfolio (top 377, height 4537) was still data-reveal="hidden". Its top part is on screen but 12 % of its height (544 px) never fits between the top of the grid and the 60 px line (386 px visible), so the observer never fires. The final screenshots of the Lighthouse reports (master and unit 1) show the same empty area under the filter row at 3 s. This is the behaviour on master too. The new rule makes the grid settled at load (BR-RVL-03), which fixes it for elements that start in view, and is also why /portfolio changes in the measurement (section 6).
1.5 Client navigation
Option decided (O1): the same rule as on a first load; no useHydrated branch.
- Why: FR-PR-09's reason is that visible content must never wait for an animation. One rule on both paths has no divergence to test, and the hydration rule (BR-HYD-01) is untouched.
- What changes: today a page reached by client navigation fades in its first screen; after this unit its first screen is visible at once and only content below the fold fades on scroll.
- Alternative: hide in the first render when
useHydrated()is alreadytrueand let the observer reveal what is in view. It keeps the entrance fade, costs a second code path and a second set of tests, and holds back content in view for the length of the fade.
2. Images
2.1 What DeferredImage renders
graph TD
Props["DeferredImage props: image, near, style props"] --> First["img: no src until loaded, loading lazy, data-deferred"]
Props --> Twin["noscript: img with src, alt, width, height, same style props"]
First --> Phase{"Page loaded and near the viewport"}
Phase -->|"yes"| SetSrc["Effect sets src, photo loads"]
Phase -->|"no"| Empty["Stays empty, as in unit 1"]
Twin --> NoJs{"Parser scripting enabled"}
NoJs -->|"yes"| Text["Content is text: nothing fetched"]
NoJs -->|"no"| Shown["Twin is an img: photo loads"]
classDef comp fill:#63b3ed,stroke:#2b6cb0,stroke-width:1px,color:#000;
classDef markup fill:#9ae6b4,stroke:#2f855a,stroke-width:1px,color:#000;
classDef choice fill:#f6e05e,stroke:#c05621,stroke-width:1px,color:#000;
class Props comp;
class First,Twin,SetSrc,Empty,Text,Shown markup;
class Phase,NoJs choice;
Text alternative: DeferredImage renders its deferred image and a noscript twin; without scripting the twin shows the photo and the CSS hides the empty image.
2.2 Facts behind the design
| Question | Answer | Evidence |
|---|---|---|
What does the empty <img> look like without JavaScript |
A sized box (its CSS size) with Chrome's broken-image icon and the alt text inside; complete is true, naturalWidth 0 |
Screenshot of /soorten-fotoshoots with JavaScript off and Reveal forced visible |
| Is the twin double-displayed | Not when the empty image is display: none (BR-IMG-02); both have the same class so the twin has the same box |
Class equality by construction; V-D2 |
| Does the twin download with JavaScript on | No. noscript content is text; React 19.3 client does not render noscript children (shouldSetTextContent in react-dom-client.production.js) |
Source |
| Does the twin add a preload | No. react-dom-server skips the image preload when tagScope & 1 (inside noscript) |
Source, line 2135 ff. of react-dom-server.node.production.js |
| Does the twin cause a hydration mismatch | No: children of noscript are not compared |
Source; V-D1, V-D3 |
| Lazy loading without JavaScript | Not honoured by Chrome; all twins load at once | BR-IMG-05 |
2.3 Every image path that depends on JavaScript
DeferredImage (deferred src) is the only mechanism that withholds a photo from the HTML, besides the hero's later slides. All image-rendering components in src (grep of DeferredImage and <Image):
| Path | Photo in the HTML | Without JavaScript |
|---|---|---|
Hero slide 1 |
src present, fetchPriority="high" |
Shown. Slides 2 and up are JavaScript only (BR-IMG-07). |
IntroGallery (4, near from its own observer) |
No src |
Twin shown. The phone row is a CSS scroll-snap list and keeps working. |
PortfolioGrid tile 1, and tiles 2 and 3 through <source media> (first row, BR-PRE-08) |
Tile 1 has src; tiles 2 and 3 have a <source> and no src |
Tile 1 shown; tiles 2 and 3 shown through the twin (and the source, which loads the same URL). Tiles link to the large photo. |
PortfolioGrid tiles 4 and up (deferred) |
No src |
Twins shown, all load at once. |
InstagramGrid |
No src |
Twins shown; tiles link to Instagram. |
PhotoMosaic, PhotoDuo |
No src |
Twins shown. |
FramedImage without priority (shoot pages, testimonials, MediaText) |
No src |
Twin shown. |
FramedImage / MediaText with priority |
src present |
Shown. |
MediaText detail photo, Pricing decorations |
No src |
Twin shown; the decorations keep aria-hidden. |
TestimonialSlide in the hidden review copies |
No src |
Twins load, nobody sees them (BR-IMG-09). |
ContactAside |
<source srcSet> in <picture>; the <img> has a data: GIF |
Photo shown from lg; no change needed. |
ContactDetails portrait, NavigationBar logos, Footer logo |
src present |
Shown. |
Lightbox |
Not in the HTML | JavaScript only (BR-IMG-08). |
TestimonialCarousel slides |
Not in the HTML | JavaScript only; the placeholder shows the longest review. |
3. The menu without JavaScript
3.1 What the visitor does
graph TD
Closed["Page loaded: panel hidden, open link in the header"] --> Open["Visitor activates open link: address gets #mobile-nav"]
Open --> Target["Panel is the target: clip-path inset 0, visibility visible"]
Target --> Choose["Visitor picks a link: normal page load"]
Target --> Close["Visitor activates close link: address gets #site-header"]
Close --> Closed
classDef state fill:#9ae6b4,stroke:#2f855a,stroke-width:1px,color:#000;
classDef act fill:#63b3ed,stroke:#2b6cb0,stroke-width:1px,color:#000;
class Closed,Target state;
class Open,Choose,Close act;
Text alternative: the open link makes the panel the URL fragment target and visible, a menu link loads a page, and the close link moves the target to the fixed header so the panel hides again.
3.2 The CSS, rule by rule (selectors; declarations in BR-NAV and BR-NJS)
| Group | Selector | Why it wins over Emotion |
|---|---|---|
| Wrapper | noscript |
Element selector, nothing competes |
| Hidden controls | [data-requires-js], [data-email-pending] |
!important (equal weight, Emotion later) |
| Empty photo | img[data-deferred]:not([src]) |
!important |
| Open link | .ns-menu-open and its three bars, hidden from 64rem |
Own class, nothing competes |
| Close link | .ns-menu-close |
Own class |
| Panel open | #mobile-nav:target |
Id: weight 1,1,0 against 0,1,0 (verified computed values) |
| Group expanded | #mobile-nav [id^="mobile-group-"] |
Weight 1,1,0 |
| Desktop submenu | [data-submenu]:focus-within > [data-submenu-panel] |
Weight 0,3,0 against 0,1,0, with transition: none |
| Focus ring | .ns-menu-open:focus-visible, .ns-menu-close:focus-visible |
Own classes |
Emotion's own rules were seen in the built HTML: they are unlayered .css-xxxx single-class rules, written next to the element in the body, so they follow the head CSS in document order. That is why the group rules use ids and why the bare marker rules use !important.
3.3 Verified behaviour (Chrome, JavaScript off, 390 x 800 and 1280 x 800, over-mij with the CSS and links injected through a request route)
| Check | Result |
|---|---|
| Panel while closed | visibility: hidden, links not in the Tab order |
| After clicking the open link | location.hash #mobile-nav, :target is the panel, clip-path inset(0px), visibility visible |
| Scroll jump | Page at scrollY 300 stayed at 300 after opening and after closing through #site-header |
| Tab order after opening | close link, Home, Over mij, Portfolio, Soorten fotoshoots, Newborn, Zwangerschap, Gezin, Loveshoot, Portretten, Contact |
| Group | [id^="mobile-group-"] computed grid-template-rows 1fr, visibility visible |
| Hamburger | Hidden by the marker rule |
| At 1280 px, Tab with 150 ms pauses | Soorten fotoshoots link, then Newborn, Zwangerschap, Gezin ... in the panel; without the pause the first Tab skipped the panel while its visibility transition began, hence transition: none |
3.4 Placement of the CSS (settled)
The <style> goes in <head> inside <noscript>, rendered by the root shell (__root.tsx) after HeadContent. That is the only place the HTML specification allows a style inside noscript; in body it would be invalid although browsers apply it. React 19.3 writes the children of noscript into the HTML in place (it does not hoist inside noscript). The script step does not touch it (raw-text element, no <script>), the CSP allows it (style-src 'unsafe-inline'), and with JavaScript the browser never parses it. An external stylesheet link in noscript was rejected: one extra blocking request for the visitors who need the fallback, plus asset plumbing, for about 1.5 KB.
4. What a visitor without JavaScript sees
Common to all pages: skip link; header with the logo and, below lg, the menu link instead of the hamburger, from lg the five links (the shoot submenu opens on keyboard focus); all text; all photos (twins load at once); footer with the Socials column, the footer menu and the copyright with the build year, but no Contact column; no e-mail address or mailto link anywhere (FR-PR-22); no hidden control that does nothing.
| Page | Also sees | Missing or different |
|---|---|---|
/ |
Hero slide 1 with h1 and call to action; intro gallery (phones swipe the CSS row); about block; price cards with "Boek" links; one review; contact section heading with the coming-soon notice; Instagram tiles | Other hero slides, arrows, pause, progress bar; other reviews (BR-NJS-04); form and mail button; "Boek" does not apply a package (BR-NJS-07) |
/over-mij |
All text and photos | Nothing |
/portfolio |
Intro, all photos (links to the large photo), Instagram tiles | Filter buttons, lightbox; ?shoot= has no effect |
/soorten-fotoshoots |
Five shoot bands with their links | Nothing |
| Five shoot pages | Story blocks, mosaic or duo photos, call to action | Nothing |
/contact |
Intro, contact card with the social channels, notice, aside photo from lg |
Form (hidden streamed segment), e-mail address, mail button |
/privacyverklaring |
Text | Nothing |
| 404 page | Text and "Terug naar home" | The two extra buttons under /soorten-fotoshoots/ (unit 1, BR-404-04) |
Imperfect and accepted: BR-IMG-05 (everything loads at once), BR-NAV-11 (history), BR-NAV-12 (JavaScript on but not loading), BR-NJS-03 (a carousel with one slide for screen readers), BR-NJS-04, BR-NJS-05 (no way to send a message once the form is enabled), BR-NJS-07, and the missing address.
5. The first photo
5.1 Largest paint per page and screen size
Chrome 2026-10-31, PerformanceObserver largest-contentful-paint, last entry after 2.5 s, unthrottled, Reveal forced visible to stand for this unit (the current build gives the same elements at 1 s, except /portfolio at 412 px, where the grid is hidden today):
| Page | 412 x 823 | 768 x 1024 | 1350 x 940 | Preload today |
|---|---|---|---|---|
/ |
h1 text | h1 text | h1 text | Hero photo (React), logos |
/over-mij |
andrea-portret |
same | same | andrea-portret (React) |
/portfolio |
f41a1916 (first tile) |
f41a1922 |
f41a7883 |
Today none. After this unit: tile 1 (React), tiles 2 and 3 with media 48em and 64em (BR-PRE-08 to 11) |
/soorten-fotoshoots |
newborn |
same | same | newborn (React) |
newborn, zwangerschap, gezin, loveshoot, portret |
own photo | same | same | Own photo (React) |
/contact |
intro text | intro text | contact (aside photo) |
None for the aside; the 72 px portrait is preloaded at every width |
/privacyverklaring |
footer logo | h1 | h1 | Logos only |
The Lighthouse reports name the same elements (mobile: / the h1 span; over-mij, soorten and shoot pages their photo; portfolio the intro p in the current build and the first grid photo in the "Reveal visible" run; contact the intro p; desktop contact the aside photo, desktop portfolio a grid photo).
5.2 Decision flow
See flowchart 1.3 in the business rules. Result: nothing to add for seven of the eleven pages (React already preloads), one explicit link for /contact from lg, for /portfolio the first row of the grid: tile 1 by React, tiles 2 and 3 by two links bound to 48em and 64em (the largest paint is a different tile per width, always one of these), nothing for /privacyverklaring. Two preloads are avoided by making their images lazy (the light logo off /, the contact portrait), because they are display: none where they are not used.
6. Largest paint: what the reports show
6.1 Two numbers per URL
Lighthouse here is in simulate mode (mobile: 150 ms RTT, 1.6 Mbps, CPU x4). The page is loaded once without throttling, and the simulator computes the LCP from that trace's dependency graph. Each report therefore holds the real, unthrottled LCP (observedLargestContentfulPaint) and the simulated one.
| Mobile URL | Master LCP | After (A) | Reveal visible (B) | Real LCP A / B, ms | LCP element in A / B |
|---|---|---|---|---|---|
/ |
4293 | 4280 | 4281 | 109 / 111 | h1 text span, both |
/over-mij |
3535 | 3608 | 3306 | 977 / 51 | portrait photo, both |
/portfolio |
2926 | 3007 | 3607 | 55 / 197 | intro text in A, first grid photo in B |
/soorten-fotoshoots |
3750 | 3683 | 3383 | 990 / 56 | newborn photo |
| newborn | 3525 | 3610 | 3457 | 1047 / 62 | own photo |
| zwangerschap | 3676 | 3834 | 3682 | 1034 / 65 | own photo |
| gezin | 3678 | 3908 | 3758 | 1027 / 65 | own photo |
| loveshoot | 3768 | 4133 | 3907 | 1016 / 65 | own photo |
| portret | 3673 | 3759 | 3682 | 1007 / 63 | own photo |
/contact |
3345 | 3007 | 3006 | 50 / 64 | intro text |
6.2 Shoot pages are worse than master (up to +365 ms on loveshoot)
Two causes, one of which this unit removes.
- The photo waits for hydration. In A the photo is in the HTML but inside a hidden
Reveal; Chrome ignoresopacity: 0elements, so the photo only becomes the largest paint after hydration, the observer and the 0.8 s fade: real LCP 1007 to 1047 ms against 62 to 65 ms whenRevealis visible (B). The simulator, which waits for that late paint, includes everything requested before it. Master had the same wait (the SPA rendered the photo hidden too: real LCP 1008 to 1046 ms), so this cause is not new; A did not remove it. - The photo is now requested first and competes for the link. In A the photo is requested at about 11 ms with the 215 KB of scripts; in master it was requested after React rendered, behind the scripts. The simulator shares the 1.6 Mbps link between the concurrent requests, so the hydration that the paint waits for finishes later by about the photo's transfer time. Evidence: the regression against master grows with the photo's size: newborn 74 KB +85 ms, portret 97 KB +86, zwangerschap 104 KB +158, gezin 117 KB +230, loveshoot 153 KB +365. The scripts grew by 7 KB (208 to 215 KB), worth about 40 ms, not 365.
B shows what this unit achieves: it removes cause 1 (77 to 226 ms better than A, and 65 ms real LCP), but cause 2 stays: zwangerschap +6, portret +9, gezin +80, loveshoot +139 ms against master. newborn is 72 ms better than master.
6.3 /portfolio 3007 to 3607 ms with Reveal visible
In A the grid is hidden (BR-RVL-13 defect, section 1.4), so the largest paint is the intro paragraph at 55 ms, and the photos requested later (two requests at 194 ms, DeferredImage after the load event) are never candidates. In B the grid is visible, the first photo is the largest paint, and its real LCP is 197 ms. The simulator now includes everything requested up to 197 ms, among it those two photos (54 KB and 144 KB): +600 ms for +198 KB (3.0 ms per KB). The element changed from text to a photo; nothing got slower for the visitor, but the lab number did. This unit's design makes the grid visible (BR-RVL-03), so /portfolio is expected near B (score 0.90 against 0.94 in A) or, with the eager first row, somewhere else that is not measured yet. B had the photos deferred; the eager row requests tile 1 (54 KB at 412 px) with the scripts instead of at 194 ms, which could move the simulated LCP either way. The owner chose the eager first row (O2); BR-DMS-05 measures it against the deferred variant and reports to the owner if it is below 0.9 or below the deferred variant.
6.4 / stays at 4280 ms in both
The largest paint on / is the h1 text span at the first paint (real 109 and 111 ms), and all 14 Reveals of the page are below the fold at 412 x 823 (the first top is at 919 px), so Reveal plays no part. The simulated number is the cost of what was requested before 109 ms: the 19 scripts (216 KB), both logos (48 KB) and the hero photo (224 KB, requested at 14 ms with fetchPriority="high"). Compare /contact: same scripts, 54 KB of images, 3007 ms. The difference, 1273 ms, is what the hero's 224 KB costs at the simulated 184 KB/s (5.4 ms per KB): 1216 ms. The same slope appears in B on the shoot pages: from newborn (122 KB of images before the cut) to loveshoot (201 KB), +450 ms for +79 KB is 5.7 ms per KB. Master gave 4293 ms for the same reason (the same hero photo).
graph TD
Real["Real trace: LCP at 50 to 200 ms"] --> Cut["Simulator: take every request started before that moment"]
Cut --> Bytes["Scripts 200 to 220 KB, logos 48 KB, first photo 75 to 225 KB"]
Bytes --> Link["Share a 1.6 Mbps link, about 5.4 ms per KB"]
Link --> Sim["Simulated LCP 3.0 to 4.3 s"]
Hide["Reveal hides the photo"] --> Late["Real LCP moves to about 1.0 s"]
Late --> Cut
classDef real fill:#f6e05e,stroke:#c05621,stroke-width:1px,color:#000;
classDef model fill:#63b3ed,stroke:#2b6cb0,stroke-width:1px,color:#000;
classDef cause fill:#fbd38d,stroke:#92400e,stroke-width:1px,color:#000;
class Real,Late real;
class Cut,Bytes,Link,Sim model;
class Hide cause;
Text alternative: the simulated LCP follows the bytes requested before the real LCP moment, and a hidden Reveal pushes that moment from about 60 ms to about 1 s.
6.5 What was not determined
- The reports contain no trace, so the simulator's inner workings (which requests it includes, how it shares the link) are inferred from the correlation above, not read from a trace. The fit is close (hero: 1216 predicted against 1273 observed; slope 5.4 against 5.7 ms per KB), but it is a fit. BR-DMS-02 keeps a trace to confirm it.
- Why Chrome never reports the hero photo as the largest paint on
/(one entry only, the text). A viewport-filling image being treated as background is likely; not confirmed (V-D10). - Whether the residual increase on four shoot pages (cause 2) is the whole of it: it fits the photo sizes, no other candidate was found.
What this unit's design does about it: removes cause 1 (fade-in gate), fixes the empty mobile portfolio, removes 25 KB of never-shown logo on ten pages (BR-PRE-05, an estimated 140 ms), and leaves the bytes of photos, scripts and the 500 x 500 logos to the responsive-photos feature, unit 3 and the owner's per-page decision (BR-DMS-04).
7. Testable Properties (PBT-01)
Properties extend the numbering of unit 1 (P1 to P17). fast-check runs in Vitest with the existing seed handling. src/lib/** and src/seo.ts have 100 % coverage thresholds, so each property file also carries example tests.
| Id | Function | Category | Property |
|---|---|---|---|
| P18 | startsHidden(top, viewportHeight) |
Invariant | For any finite numbers: an element whose top is below the viewport height is hidden; one whose top is smaller is never hidden (BR-RVL-03). Equal gives hidden. |
| P19 | startsHidden |
Metamorphic (monotonic) | If an element with top t is hidden then every element with top t' >= t (same viewport) is hidden; shifting top and viewport height by the same amount does not change the result. |
| P20 | nextRevealPhase(phase, event) |
Invariant (state machine) | For any phase and any sequence of events the phase never returns from revealed to hidden or settled, and never goes from settled to revealed (BR-RVL-09). |
| P21 | nextRevealPhase |
Idempotence | Applying the same event twice gives the same result as once (StrictMode runs effects twice). |
| P24 | firstRow(index, columns) |
Invariant | For any column map with increasing breakpoints and column counts and any tile index: the result is 'always' exactly when index is smaller than the base column count, null exactly when it is not smaller than the largest column count, and otherwise { media } of the first breakpoint whose column count exceeds index. |
| P25 | firstRow and firstRowPreloads |
Metamorphic (monotonic) | If a tile is eager at some width it is eager at every wider width (the set of eager tiles only grows with the column count); and the number of tiles eager at a width equals the column count at that width. firstRowPreloads(items) has one link per tile with a media, none for 'always' or null tiles, href equal to the tile thumb's src, and never more links than max columns - base columns (no duplicates). |
| P22 | preloadImage(image, options) |
Invariant | For any image and options, href equals image.src, rel is preload, as is image, fetchPriority is high, media is present exactly when given and equals it, and no other key exists. |
The check that the CSS constant has no </, url( or @import and balanced braces (BR-NAV-02) is an example test on one constant, not a property.
Components with no properties:
| Component | Result |
|---|---|
Reveal, useReveal |
No PBT properties identified beyond P18 to P21: the hook reads the DOM and renders; the decisions are in revealPhase. Example tests with a stubbed rectangle and the fake observer. |
DeferredImage |
No PBT properties identified: fixed markup from one image; example tests with renderToString and hydrate(). |
NoScriptStyles, NoScriptNav |
No PBT properties identified: constant markup; example tests and the browser tests of BR-DTS-03. |
NavigationBar, MobileNav, MobileNavGroup, NavSubmenu, Hero*, FilterBar, Footer, FooterColumn, Testimonials |
No PBT properties identified: attribute additions on components with small fixed inputs. |
Route head() of /contact and /portfolio |
No PBT properties identified: they call preloadImage and firstRowPreloads (P22, P24, P25) with fixed content. |
PortfolioGrid |
No PBT properties identified: markup from a list; the choice per tile is firstRow (P24, P25); example tests with renderToString and hydrate(). |
8. Requirement coverage
| Requirement | Where |
|---|---|
| FR-PR-09 | BR-RVL-01 to 06, 09 to 14; section 1 |
| FR-PR-10 | BR-RVL-07 |
| FR-PR-11 | BR-IMG-01 to 10; BR-RVL-01; section 2 |
| FR-PR-12 | BR-NAV-01 to 12; section 3 |
| FR-PR-13 | BR-NJS-01 to 11; BR-IMG-07, 08; section 4 |
| FR-PR-21 | BR-PRE-01 to 12; section 5 |
| FR-PR-22 | BR-NJS-02 (placeholders hidden, address stays absent) |
| NFR-PR-P-01, P-03 | BR-DMS-01 to 05; section 6 |
| NFR-PR-Q-01, Q-03 | BR-DTS-01 to 04 |
| NFR-PR-SEC-01 | BR-NAV-02 |
9. Security Baseline and PBT compliance (unit 2)
| Rule | Status | Note |
|---|---|---|
| SECURITY-04 | Compliant | CSP unchanged: the new CSS needs style-src 'unsafe-inline', already present and documented; no new script, no inline handler |
| SECURITY-13 | Compliant | No external resource added; preloadImage links are same-origin hashed assets |
| SECURITY-09, 10, 11, 15 | Compliant / unchanged | No server output, dependency, endpoint or error path changes |
| PBT-01 | Compliant | Section 7 |