The datum rule is a single selector in templates/report.css:87:

.center > main > section:first-of-type > .section-heading { margin-top: 0 }

Sessions and Quotes render <section> as the direct child of <main>, so it matches. Codebooks renders a fragment whose first child is <div class="codebook-header"> — a flex row holding the h1 and the button. Two mismatches at once: not a <section>, and the heading is a grandchild. The selector misses, the atom's margin: 2.5rem 0 1rem stands, and the title drops 40px.

This appeared on 6 Aug 2026, when Codebooks' h1 was given the .section-heading class it had always been missing (docs/design-lens-template.md § "Systematising the heading"). Before that it had no class, no rule, and — by luck — no margin either. The 22 Jul audit measured Codebooks at 84px, dead on the datum. Fixing the missing rule handed it the missing margin.

Because align-items: flex-start puts the button at the top of the row while the h1's own margin pushes it down inside its flex item, the button now sits higher than the title it belongs to. That is what the eye is objecting to, and it is why "move the button" and "fix the alignment" feel like the same request.

• Fixes the stated defect exactly. Title and button top-align; the row sits on the datum; Codebooks reads like every other lens.
• Zero markup change, zero locale change, zero desktop change. Wrap the fragment in <section> and the existing selector reaches it — no new rule at all.
• Leaves a filled-blue primary button as the loudest thing on a page whose actual job is organising tags. The heading row now shouts twice.
• Puts a primary button directly above the secondary "Codebook lab" button 16px below it — an experiment ranked louder than the shipping feature.

• The strongest alignment of the five — nothing shares the title's row, so the title is the first ink, exactly like Quotes and Analysis.
• Reading order becomes honest: what this page is → how to use it → the one thing to click. The button follows the sentence that motivates it.
• Structurally the simplest possible markup: drop the flex wrapper entirely, and the datum selector matches for free.
• Costs a band of vertical space at the top of the lens — roughly 50px before any tag appears, on the page most likely to be scrolled.
• A left-aligned filled button under body text reads as a form's submit. Softening it to secondary fixes that but then it competes with "Codebook lab" for the same weight.

• This is not a new idiom. ContentView.swift:2208–2229 already gives Quotes two contextual toolbar items (starred filter, tag sidebar) and Analysis one (heatmap). Codebooks is the lens with an empty contextual slot — and a loud in-pane button instead.
• Content pane opens perfectly clean: title, description, work. Identical to Quotes and Analysis.
• The action is already in the native Codes ▸ Browse Codebooks… menu (MenuCommands.swift:1157), so a toolbar item is the standard macOS promotion of an existing menu command — not an invention. books.vertical is already its symbol.
• The existing contextual items are toggles; this one opens a modal. Defensible (Export is a menu-opener in the same capsule, and Xcode/Keynote put library-openers in the toolbar) but it is a widening of the slot's meaning.
• Web serve mode has no native toolbar, so the SPA still needs an in-page entry. That is what ct() exists for — show the in-page button on web, hide it on desktop — but it is a two-surface answer, not one.
• Toolbar real estate on Codebooks would then be Export · Library · Search, and Search there is currently a disabled placeholder. Adding a live item beside a dead one sharpens the dead one.

• Same one-line geometry fix as A, and it settles the second problem A leaves behind: the page now has a consistent title-left / action-right rhythm at three descending scales — h1 + Library, then "foo tags" + Codebook lab, then each framework + its switch.
• Matches what the button actually is: a browse affordance, not the page's primary act. The page's real primary acts (rename, drag, merge) have no buttons at all.
• Loses the only bit of colour drawing a first-time user toward the one place codebooks come from. Discoverability of the Library is already thin — the sidebar's "Codebook Library →" button was removed on 14 Aug precisely because it duplicated this one (CodebookSidebar.tsx:159).
• Three same-weight bordered buttons stacked down the left-to-right axis can read as undifferentiated.

• Reuses an idiom this exact page already teaches twice: .new-group-placeholder ("+ New group") and the library modal's own .tile-create. A dashed card meaning "one more of these" needs no explanation here.
• Puts the affordance where the need arises — at the end of the codebooks you have, in the moment you notice you want another. Header buttons ask before the user has the question.
• Header is fully clean; the datum problem disappears with the wrapper.
• Below the fold on any project with more than one codebook installed. A researcher who never scrolls never learns the Library exists — the opposite of A's problem.
• A dashed card is a create gesture everywhere else in the app; here it opens a browse modal. Near-miss on an established meaning, which is worse than an unfamiliar one.
• New CSS/markup, not a one-liner. The largest build of the five.

The atom sets max-width: none so the rule spans the pane — but the h1’s containing block is <div style="flex:1;min-width:0">, the left column of the flex row. “Full width” resolves to the width of that column, which is the pane minus the button. Every other lens draws its rule pane-wide because its heading’s containing block is the pane.

So the wrapper causes both defects at once: the datum miss (heading is a grandchild, selector doesn’t reach) and the short rule (heading is column-width, not pane-width). Any fix that keeps a flex column around the heading keeps the second one.

• R1 — move the border to .codebook-header and null it on the h1. Works, and it is precisely the drift the atom was created to stop: Codebooks’ rule would be drawn by a different box than every other lens’s, re-opening the “a class you have to remember” failure the atom’s own header comment describes.
• R2 — .section-heading becomes a flex row, action inside the h1. Border spans the pane by construction, one CSS change, and .framework-section-header on this very page is the precedent one scale down. Cost: a <button> inside an <h1> is valid HTML but VoiceOver reads “Codebooks Codebook Library, heading level one”. The h2-scale precedent avoids this only because its title is a plain div.
• R3 — the atom becomes a wrapper (recommended). <div class="section-heading"><h1>…</h1>{action}</div>. The border sits on one box that spans the pane by construction; the button is a sibling of the heading, so the heading announces as just “Codebooks”; the datum selector is untouched (it matches the class, not the tag); and the static-render templates that write <h2 class="section-heading"> keep working, because the CSS never requires an inner element. All 19 call sites are unchanged — SectionHeading.tsx absorbs the shape, which is the entire reason that component exists.
• R4 — sidestep. Put the web button under the description (option B) so nothing shares the row on either surface. Zero atom change, but it abandons the right-aligned action D was chosen for, and leaves the atom unable to carry a zone-level action the next time one is wanted.

Worth noting the stakes: under C the desktop heading row is bare — the button is in the toolbar — so the truncation only ever shows on the web SPA. That lowers the urgency, not the correctness. Doing R3 now makes a zone-level action free on Quotes, Sessions and Analysis too, which is the same argument that made SectionHeading a component rather than a class.

.section-heading is now the row, not the text: a wrapper carrying the keyline, with the <h1> and an optional action inside. The shape does not vary with the action — every zone title in the app is the same box, so a change to the title treatment lands on all five lenses identically, and a lens that grows an action later gets the geometry the others already have. .codebook-header is deleted; it was the only per-lens header wrapper and it caused both defects.

No new exceptions. The action slot is the same construction as .framework-section-actions one scale down (flex, centred, --bn-space-sm, never shrinks), so the three title-left / action-right rows a lens stacks are one pattern at three sizes. The button is the stock .bn-btn atom, demoted from .bn-btn-primary. Desktop reuses the existing Codes ▸ Browse Codebooks… action string, symbol and help text; the only new locale key is the toolbar’s one-word label.

Every option fixes the datum, because every option removes the flex wrapper or flushes past it. The alignment is therefore not the deciding question — it is free in all five. The deciding question is how loudly a browse affordance should speak on a page whose real work is silent (drag, rename, merge).

C + D compose. Toolbar item on desktop, quiet secondary in-pane on web, one flush heading on both. That is the shape the codebase is already bent toward: an empty contextual toolbar slot on exactly this lens, a native menu item already wired, and a ct() helper built for this fork.