Load-bearing lines
Declarations that look optional and are not, and the class names already claimed by style.css. Both fail silently, and both have caught us more than once.
We found these by reading the older development diary at content/akb/knowledge-base.html against the stylesheets that are running now. Several fixes described there are still in the CSS, still doing their job, and carry no comment saying so. Anyone tidying those files, including us, would read them as leftovers and delete them.
The problem with all six is the same: none of them breaks anything visible on a desktop browser. They fail on phones, or when a page is opened from disk, or in a way that looks like a design choice rather than a fault. So this is the record of what they are for.
The drawer's height, twice
assets/css/drawer.css
height: 100%;
height: 100dvh;
The repetition is deliberate. 100vh and 100% do not account for the address bar on mobile Safari, which slides away as you scroll, so a full-height drawer is cut off at the bottom or overshoots. 100dvh is the dynamic viewport height and tracks the bar as it moves. The first line is the fallback for browsers that do not know dvh; the second overrides it where it is supported.
Delete either line and it still looks correct on a laptop. The same pair appears in reader.css for the knowledge base rail.
Momentum scrolling inside the drawer
assets/css/drawer.css
-webkit-overflow-scrolling: touch;
Without it, a scrollable panel on older iOS drags stiffly instead of gliding and stopping the way every other list on the phone does. It reads as the site being broken rather than as a missing property.
Trapping the scroll
assets/css/drawer.css
overscroll-behavior-y: contain;
When you reach the bottom of the drawer and keep dragging, this stops the scroll from passing through to the page behind it. Without it, the page underneath moves while the menu is open, and closing the menu leaves the reader somewhere they did not choose to be.
Blending the logo mount
assets/css/drawer.css, on .drawer-logo-mount
mix-blend-mode: screen;
This one has changed meaning and is worth a decision. It was added when the logo was AI-generated artwork rendered on a pure black background. Screen blending drops pure black to transparent, so the art could sit on any dark background without a visible box around it. That was genuinely load-bearing.
The logo is now flat SVG bars with no background at all, so the rule no longer removes anything. It still lightens the bars against the drawer glass, which is a look rather than a fix. Keep it or drop it on appearance alone, but it is no longer holding anything up.
Why the drawer disappears when you open a page from disk
drawer.js builds the menu by fetching /drawer.html and injecting it. Browsers block that fetch for pages opened directly from the filesystem, so double-clicking any page in Finder gives you the content with no navigation.
Nothing is broken. It needs a server. This is why the logo archive in _shelf/logos/ has all of its CSS inlined: it is meant to be opened from disk years from now, so it cannot depend on anything being fetched.
Why every shared path starts with a slash
Stylesheets, scripts and the drawer are all linked as /assets/css/... rather than assets/css/.... A relative path resolves from the folder the page is in, so a page at content/consulting/proposals/ looking for assets/css/style.css searches inside that folder and finds nothing. The leading slash resolves from the site root, so one link works at every depth.
The same trap caught the knowledge base builder on 26 August 2026, where the depth calculation assumed pages sat at the site root and produced links that resolved to folders that did not exist. Different mechanism, same mistake: assuming you know how deep you are.
The names that are already taken
A different trap in the same family, and one that has now caught us three times.
style.css is loaded by most pages on this site, and it claims a set of bare element selectors and some very ordinary class names. Anything you write afterward that happens to use one of those names inherits behavior you did not ask for, silently, with no error.
Bare elements
These apply to every matching tag on any page that loads style.css:
| Selector | What it quietly does |
|---|---|
p | Sets max-width: 700px and margin: auto, which centers every paragraph |
h1 | Uppercase, 3 to 5rem, and painted through a transparent gradient fill |
h2 | 1.6 to 2.2rem, bold |
* | Zeroes all margins and padding, including list indents |
The p rule is the most surprising. margin: auto on a width-capped block centers it, so body copy on a new page arrives centered for no visible reason. The 30-day plan has five separate max-width: none declarations undoing exactly this.
Class names claimed by style.css
container, section, hero, hero-content, glow-bg, reveal, footer, footer-grid, footer-info, footer-bottom, form-group, glass-container, cta-group, btn-primary, btn-outline, terminal-output, sys-msg, sys-alert, cursor-blink, and the v6- prefix.
section is the one that bit most recently. On 28 August 2026 a new page used class="section" for its own content blocks and every one of them came out centered, because style.css sets text-align: center on that name for landing pages.
What to do instead
Prefix page-specific classes. The knowledge base reader page uses kb-section. The 30-day plan uses doc- throughout. The conversion band uses own-band-. A two or three letter prefix costs nothing and makes the collision impossible.
Undo the element rules inside your own container. If a page carries body copy, expect to write max-width: none and a real margin on p inside whatever wraps it, because the global rule will otherwise center it.
Check before naming. grep -n "\.your-name" assets/css/*.css takes a second and settles it.
Why this keeps happening
Three separate instances so far: the drawer's submenu losing to an id-level rule, every plain body link falling back to the browser default because nothing claimed a, and now .section. None of them errored. In every case the CSS read correctly and only the cascade disagreed, which is why they were all found by measuring computed styles in a browser rather than by reading the file.
The rule
Two habits, one for each half of this entry.
Before removing a declaration that looks redundant, check here first, and check on a phone rather than in a narrow desktop window. Four of the six declarations above fail only on a real device.
Before naming a new class, run grep -n "\.your-name" assets/css/*.css. If style.css already claims it, prefix yours instead.
Both halves fail the same way: no error, no warning, and CSS that reads perfectly while the cascade quietly disagrees.