Vocabulary

The words this site uses

Motif leans on a few words — fidelity, run log, audit, tier — that mean something specific here and something vague elsewhere. Each entry below is defined the way the site actually behaves, with a live example to check it against. If a definition and a page disagree, the page is right and this list is the bug.

Fidelity score457 runs scored: How closely a model's output matched the brief it was given.; Scored 0–100 per run by comparing the rendered result with the brief's checklist — layout blocks present, type scale honoured, motion described in the block, and no invented sections. It measures agreement with the brief, not beauty: a plain page that follows the brief scores higher than a beautiful one that ignores it.; AI tool landing with terminal motif — 93% average →
Run log116 prompts: Every model, date, score and failure kept on the prompt page.; A prompt ships with the runs that produced it, including the ones that failed to build. A prompt page with no failed runs says so rather than implying universal success — the log is a record, not a highlight reel.; Run logs on the prompts hub →
Concept render: The poster drawn from a prompt, not a screenshot of a model's output.; Prompt posters are generated from the prompt's own metadata — blocks, palette, industry — and are labelled as concept renders because the site ships no model screenshots. Calling a drawing a screenshot would be the easiest lie to tell here, so the label is part of the component.; Poster gallery →
Audit (a11y / quality)289 assets scored: Two 0–100 scores on every asset, with the checks named.; The a11y score covers keyboard reach, focus visibility, label text and reduced-motion handling; the quality score covers originality, token coverage and the size budget for its kind. Both are automated checks against the source, not a human certificate, and the audit page prints what they do not cover.; Worked example: Inline Status Banner (97/94) →
Tier (performance): Low, mid or high — what a background costs before you use it.; Every background is placed in a tier from its paint cost: low is a static gradient or a single CSS animation, mid adds layered blur or many small elements, high leans on WebGL or hundreds of animated nodes. The tier is a promise about the cheap path, not a claim about your device.; Backgrounds with tiers →
Size budget289 assets checked: A per-kind ceiling in KB, checked at build time.; Elements get 8 KB, animated pieces 10 KB, sections 8 KB and templates 28 KB. The number is the measured source size of the component, and an asset that goes over shows the overage rather than hiding it.; Budgets and their current standings →
Run-tested: A prompt that has been executed, not merely written.; “Run-tested” means at least three recorded runs exist and the prompt body matches the one that was executed. It is not a quality claim about the prompt's subject.; See which prompts qualify →
Browser-local: Saved lists, streaks and stickers that never leave the machine.; Motif has no accounts, so anything personal is stored under a motif:* key in your browser and is readable only there. Every panel names its key; clearing site data is the delete button.; The habits drawer →
Skeleton: A placeholder that reserves the space content will occupy.; Skeletons are only useful if their box matches the final content — a skeleton that shifts when it resolves is worse than a spinner. The pattern is used in the demo panels and the preloader scene, both of which reserve the exact final height.; Skeleton card component →
Reduced motion: The visitor's own setting, honoured by every moving scene.; Where a scene would loop or travel, prefers-reduced-motion stops it and leaves the end state or a static equivalent. It is treated as a preference, not a fallback: the reduced version is designed, not merely disabled.; How the scenes handle it →

Why a glossary earns its page

Two reasons, one of them selfish. The selfish one: people search for definitions, and a term defined precisely is a page that answers a real question instead of chasing a keyword. The honest one: this site asks you to trust numbers — scores, budgets, run counts — and numbers are only trustworthy when the words around them are pinned down. “Fidelity” without a definition is decoration.

The markup on this page is a DefinedTermSet, which is a real schema.org type for exactly this. There is no FAQPage here, because the same content repeated under two types is spam with extra steps — the rules for which markup this site ships are on /quality/schema.