Colophon.

The hierarchy is deliberate: a warm-serif voice for the editorial register, a workhorse sans for prose and UI, a monospace for the numerical and structural register that surrounds market data. Each is loaded as a variable font so weight and italicization carry over without separate files per style.

The dark default reads as warm rather than neutral — every surface tone is shifted a few percent toward yellow. The accent is amber (#E8A62C), tuned to feel structural rather than decorative: used for rules, marks, citations, and live statuses, never as a flat background fill.

The research surface (/research/*) carries a separate sepia paper- reader theme — cream against a warmer-than-white background — so long reads feel like reading paper, not a UI. The light theme is a third skin, available via the header toggle.

Data-positive and data-negative carry their own tokens (#3DAA6F, #D0524F), separate from the general positive / negative semantic tokens — brighter, more chromatic, so ticker deltas read at a glance.

Motion tokens carry four durations (instant 80ms, fast 160ms, normal 240ms, slow 420ms) and a small set of eases. The standard out-ease (cubic-bezier(0.22, 1, 0.36, 1)) decelerates fast; the spring overshoots subtly. Page transitions crossfade 240ms; the kinetic hero word-reveal runs ~880ms on first paint.

The animation runtime is motion (the successor to framer-motion, swapped in P13). It's tree-shakeable per island; theming transitions apply selectively to the typographic spine (header, main, sections, articles, footers, h1-h4, p) rather than every node, which kept the ticker rail and chart canvas from getting paid the cost of a 240ms transition on every theme change.

The research and essay surfaces use a Tufte-inspired editorial register: sidenotes float into a 280px right gutter on wider viewports; pull quotes appear in two variants (centered with hairline amber rules, or floated to the right column); citations carry a dotted amber underline that opens a 280px popover on hover. Drop caps open methodology bodies; section numerals are roman, dynamically allocated so absent sections don't leave gaps in the numbering.

Each primitive has a defined role — sidenotes for running gloss that accompanies the prose, footnotes for one-off references, hover citations for trader/quant references where the reader might want a one-line gloss. Using the wrong one weakens the voice; the discipline is deliberate.

Every data widget on the site cites its upstream source explicitly. The intent is editorial — readers should be able to verify any number on the page by following the chain back to its issuer.

Authority cascade for the macro-event calendar (the FOMC / CPI / NFP / GDP / PCE markers on every chart): the Federal Reserve's published meeting schedule, the BLS CPI release calendar, the BEA GDP+PCE schedule, and the first-Friday rule for NFP. The dates are curated into a static config — accurate, stable, and independent of any live calendar feed's coverage. Values for the same releases (actual / consensus / prior) join from Finnhub's econ-calendar at runtime.

The site is a Next.js 16 static export served from Cloudflare Pages. Every page is pre-rendered at build time. Anywhere a request needs an API key — quotes, news, filings, the daily brief — a Cloudflare Pages Function intercepts at /api/* with the key held in environment variables. The client never sees a secret.

Pure handlers live in src/server/ with a thin wrapper in functions/api/; the test suite exercises the handlers as plain Request/Response functions without spinning up the Next runtime. Full-text search is powered by Pagefind, indexed at postbuild against any page carrying data-pagefind-body.