v1.0 · artifact as produced · raw file
# 05 — Implementation Mode: `static-artifact` · Direction: A — Specimen Sheet · Date: 2026-08-24 ## Files | File | Role | | --- | --- | | `05-build-source.html` | Authored source, with `__IMG_*__` placeholders | | `05-build.html` | Built output — placeholders replaced by data URIs (201 KB) | | `thumbs/*.png` | Gallery images, generated from this run's own artifacts | Build step: one script substitutes each placeholder with a base64 data URI. It is a scratch tool, not a deliverable, and is deleted at the end of the run. The published file has no build dependency — it is one HTML file with everything inside it. ## Two-pass record (taste calibration) **Pass 1 — the plan.** - *Colour*: warm paper ground, cool cyan-leaning ink and hairlines, one drafting red. Derived in `color-protocol`, values in `tokens.json`. - *Type*: two families, three roles. Tight semibold sans display, sans body, and monospace doing real work on labels, captions, filenames and stamps. - *Layout*: bordered sheet, 88px vertical index rail, figure-numbered sections separated by captioned rules. - *Signature*: the callout system — figure captions knocked out of a hairline, and `GATE` stamps on exactly the two steps where a person decides. **Pass 2 — critique of the plan before coding.** Three things were changed: 1. *Would I produce this for any similar brief?* The sheet frame and figure captions — no. The four-across card grid for the steps — **yes**, that is the default answer for "show seven things". Revised: the grid is bordered as a single figure with internal hairlines, so it reads as one drawing rather than eight cards. Cards are the thing everyone ships; a bordered figure is what a specification does. 2. *Is the hero image-led by habit?* The wireframe left the hero-visual question open. Answer: **no image**. The project has no logo, and a stock-feeling visual would contradict the page. The figure grid immediately below acts as the hero's visual weight. 3. *Is the mark colour earning itself?* Initially the drafting red was on all eight figure numbers. That spreads boldness evenly, which is exactly what the restraint rule forbids. Revised: red on the step labels, and the two gate cells get a soft red field plus a stamp — so the one thing the eye catches is the structural claim the page is making. ## Content decisions - **No invented social proof.** No counts, no logos, no testimonials. There are none. - **Install commands are visibly stamped `PLACEHOLDER`.** This is a modelled UI state, not an oversight — the page says plainly that shipping a failing command is worse than none. - **Acknowledgements name all four upstream projects with licences**, and state explicitly that no text was reproduced. This is the only place proper names appear. ## Bilingual implementation One DOM, not two. Every translatable node carries `data-i18n`; English is captured from the markup at load, Vietnamese lives in one object. The toggle swaps `innerHTML`, sets `documentElement.lang`, and moves `aria-pressed`. Vietnamese is written, not machine-translated, and carries the same argument at the same length. Both faces are system stacks with full diacritic coverage, so the second language is not degraded and the file needs no font download. ## Known defects at the end of Step 5 Found by looking at the render. Left for the Step 7 loop deliberately — that is the step that exists to catch them, and fixing them silently here would make the review theatre. 1. `h2` inherits `line-height: 1.6` from body. At display sizes it reads as two loosely stacked lines instead of one heading. 2. Section rhythm compounds: `section` bottom padding plus `.figrule` top margin produce voids between sections that read as absence rather than composition — the precise failure the direction-gate reference warns about. 3. The two-column trail section leaves a short right column, so the gap below it is uneven. ## Accessibility built in, not retrofitted Skip link · semantic landmarks (`header`/`nav`/`main`/`section`/`footer`) · every image has a real alt description · the language control is two real buttons in a labelled group with `aria-pressed` · focus-visible outline on every stop · no hover-only affordances · press feedback gated behind fine pointer.