A shared contract for how the site looks, speaks, and behaves. It keeps new work useful and recognizable without turning consistency into sameness.
src/data/design-system.jsonsrc/data/design-system.schema.jsonThis is the shortest conforming route through the system. Use the detailed sections when a task calls for them; do not invent a parallel summary.
Name the textile task, intended audience, and realistic next action in the opening reading block.
Choose a canonical page pattern, keep related copy together, and give space only to information or interaction that earns it.
Use warm paper, ink, madder, restrained weld, bounded display type, readable sans text, and textile imagery that explains the work.
Keep source, date, scope, access, assumptions, limitations, and correction paths beside consequential claims.
Check the complete interaction state set, 320–1440px layouts, both themes, keyboard use, reduced motion, long content, and the required audits.
Read brand.strategy, voice, pagePatterns, implementation.componentInventory, accessibility
Deliver A task-first content order, a named page pattern, shared implementations where available, bounded claims, complete states, and responsive verification.
Read foundations, components, implementation.componentInventory, implementation.interactionStates
Deliver A documented information relationship, semantic tokens, every applicable state, 44px primary targets, and an implementation-map update.
Read brand.messaging, brand.trustModel, voice, voice.prohibitedPatterns
Deliver Job-first language, a realistic next action, the correct claim level, and no promotional authority or invented practitioner relationship.
Read brand.visualDirection, identity, imagery, iconography
Deliver One canonical visual mode, a useful focal subject, verified crop and construction detail, accessible treatment, and recorded provenance.
Read dataDisplay, components, accessibility, implementation.conformance
Deliver A named question, unit and denominator, source and date, evidence lane, directly readable values, and explicit interpretation boundaries.
An open textile workroom reference: practical enough for a first repair, rigorous enough for a production handoff.
Find the next useful source, tool, or working answer—and see what it can and cannot prove.
Make the next textile decision inspectable.
Practical textile knowledge becomes more useful when the route, source, assumptions, date, and limits remain visible beside the answer.
For makers, repairers, and apparel teams with a textile decision in front of them, Stitch Authority is an independent decision reference that connects working answers to maintained source routes and explicit limits.
A material, technique, measurement, pattern, or repair problem is stopping the work.
A product-development, fit, costing, sourcing, standards, or production handoff needs a defensible route forward.
Making, repair, materials, pattern work, standards, and production remain connected by the task instead of being split into unrelated legacy databases.
Visible proof: Task routes, site search, permanent source records, related guides, and working tools use the same decision vocabulary.
The source, checked date, scope, access, and limitation travel with consequential information.
Visible proof: Source-reviewed guides, portable records, public JSON exports, claim levels, and direct correction routes keep the basis inspectable.
The site helps turn reference material into a calculation, comparison, brief, checklist, or next check.
Visible proof: Calculators, printable working templates, comparison records, and task-specific handoffs preserve units, assumptions, and recovery.
Freshness is a dated process rather than an unsupported claim that the collection is current.
Visible proof: Review cadence, next-review dates, append-only history, a public maintenance queue, and per-record review packets expose upkeep.
Makers and repairers. Plain-language help choosing materials, learning techniques, solving measurements, finding patterns, and locating repair routes.
Apparel teams. Working references for product development, specifications, costing, standards, and production sourcing.
Editorial workroom, not heritage cosplay: warm paper, near-black ink, madder red, weld yellow, disciplined serif display type, readable sans text, and mono only where information benefits from calibration.
Means Name the task, object, unit, and next action in language a capable colleague would use.
Not Institutional posturing, vague mission language, or invented editorial labels.
Means Show assumptions, dates, jurisdiction, evidence, and the boundary of the result.
Not A severe tone, tiny technical copy, or false precision.
Means Let believable material, construction, tools, and work-in-progress details carry the textile character.
Not Decorative thread motifs, nostalgia filters, or craft-store ornament.
Means Expose adjacent terms, alternative routes, unresolved gaps, and ways to inspect further.
Not Endless recommendations, trivia, or a confident answer where the source remains uncertain.
Means Use typographic craft, calm density, and evidence to create confidence.
Not Oversized claims, badges, theatrical statistics, or a generic premium aesthetic.
Begin with the decision or action a person is trying to complete, then reveal specialist terms and source models as useful filters.
Trust comes from named sources, checked dates, visible limits, and correction history rather than status language or inventory totals.
A newcomer should reach a plausible next step quickly while a practitioner can inspect units, assumptions, provenance, and handoff detail.
Freshness, review cadence, prior state, and correction routes remain visible wherever they affect a decision.
Tokens, information relationships, and interaction behavior stay consistent while page composition changes to fit the work.
The voice is confident because the work is inspectable, not because the site declares itself authoritative.
Name what the user can do before explaining the collection or taxonomy.
Find a clothing repair service in Norway.
State the source, observation date, and limit beside consequential claims.
The checked catalogue exposed 679 listings; that count is not a provider census or quality rating.
Prefer concrete nouns and verbs. Remove institutional posturing, motivational filler, and theatrical labels.
Compare access, geography, evidence, and limits.
A record may document a useful route without recommending, ranking, recruiting, or approving anyone in it.
Use this directory to make a shortlist, then verify the provider directly.
Finish instructions and empty states with one realistic recovery or next step.
Try the country name, item, or service—or browse Repair and conserve.
Orient someone to the next useful task without making an authority or completeness claim.
Tone: Short, direct, and task-first.
Find a source, tool, or working answer.
Answer first, then explain the work, evidence, limits, and next check.
Tone: Instructional, plainspoken, and specific about materials, units, and sequence.
Test the seam on a scrap before cutting the garment.
Describe what the route is useful for and where direct verification begins.
Tone: Factual, bounded, and free of ranking or approval language.
Use this directory to make a shortlist, then verify the provider directly.
State the result with units, assumptions, rounding, and a reversible correction path.
Tone: Exact without implying that a calculation replaces a sample, test, specification, or qualified review.
Result: 1.6 metres. Add your chosen shrinkage and cutting allowance before purchase.
Explain what happened, preserve valid work, and name the next action.
Tone: Calm, specific, and non-blaming.
No source matched that country and task. Keep the task filter or clear the country.
Separate observed source material, interpretation, unresolved risk, and delivery status.
Tone: Cautious, source-led, and explicit about what Stitch Authority has not established.
This route helps locate the current standard; confirm applicability and the controlled text before acting.
| Claim level | When it can be published | Example |
|---|---|---|
| Observed | A named source was checked on a stated date and the interface describes only what was visible. | The directory exposed a Patternmaking & Related Services filter when checked on 26 July 2026. |
| Bounded inference | The conclusion follows from cited observations and is clearly labeled as an interpretation or next-check recommendation. | Use the filter to form a shortlist, then confirm service scope and availability with the professional. |
| Not established | Never present the claim as fact without new evidence. State the limitation instead. | Listing presence does not establish identity, credentials, availability, file compatibility, or quality. |
Use the needle-and-thread mark with the Stitch Authority wordmark in primary navigation and formal brand contexts.
Warm neutrals carry the page. Madder identifies links and primary actions. Weld is a sparse secondary accent, not a second primary color.
--bgPrimary canvas
#FBF8F3 · #17150F--surfaceGrouped sections and controls
#F4EEE4 · #211E17--surface-2Secondary surface and image placeholder
#E9E0D2 · #2C2820--borderSeparation, never the only control boundary
#D9CDB8 · #3A3529--fgPrimary text
#23201B · #EDE6D8--mutedSecondary text
#57503F · #C4BBA8--muted-2Metadata and tertiary text
#6E6753 · #9A917D--brandLinks, primary actions, focus, and brand mark
#6E2B2E · #D79A94--weldSparse craft-track accent and warm emphasis
#C0801E · #E4B45C--positiveSuccess, always paired with words or an icon
--warningCaution, always paired with words or an icon
--dangerErrors and destructive actions, always paired with words or an icon
These roles keep dark editorial and operational surfaces stable in either theme.
--inverse-bgPersistent dark canvas inside either theme--inverse-surfaceRaised or interactive surface on the inverse canvas--inverse-fgPrimary text on the inverse canvas--inverse-mutedSecondary text on the inverse canvas--inverse-tertiaryMetadata on the inverse canvas--inverse-borderRules and control boundaries on the inverse canvas--inverse-accentSparse warm accent on the inverse canvasDisplay type gives the reference a voice. Sans text carries the work. Mono marks calibrated information. None of them is permission to make copy tiny.
Aa Bb 123
--font-displayPage titles and editorial headings only
Aa Bb 123
--font-sansBody copy, navigation, forms, buttons, and UI
Aa Bb 123
--font-monoEyebrows, metadata, measurements, and calibrated tables
Aa Bb 123
--font-serifExisting quotation treatments only; do not introduce into new UI
--step--1Metadata12.5–13.4px--step-0Body and controls16–17.3px--step-1Lead and card title19.2–22.4px--step-2Section heading24–31.2px--step-3Page or major section title28.8–43.2px--hero-titleMaximum product headline52px desktop · 46px tablet · 36px mobileA 12-column mental model with 24px gutters; collapse by content pressure, not device names.
--sp-28pxTight internal relationships--sp-312pxLabel-to-content and compact control gaps--sp-416pxDefault component padding and row rhythm--sp-624pxColumn gutter and mobile page edge--sp-832pxDesktop page edge and component group--sp-1248pxSection sub-group--sp-1664pxPage intro and major division--sp-2496pxMaximum standard section breathing room--control-min44px / 2.75rem. Primary controls and touch routes use the 44px house minimum. The 24px WCAG floor is not the product target.
--shadow-floatUse surface contrast or a one-pixel rule before adding shadow. Navigation panels and temporary trays that must read above document flow.
--z-sticky40Persistent site header--z-menu60Navigation menus above the header and document--z-overlay100Full-screen navigation and blocking overlaysHelp someone recognize a material, understand a textile action, or inspect a working context. An image must carry information before it carries mood.
Reference still life · 16:9Use: Swatches, surfaces, structure, drape, seams, tools, and construction details shown clearly enough to compare.
Compose: Fill the frame with the relevant material or detail, preserve useful scale, and keep color and texture believable.
Use: Hands, tools, and intermediate states when the action itself helps explain a technique or decision.
Compose: Show one legible action with a clear focal point. Hands belong only when they demonstrate work, not as generic maker atmosphere.
Use: A restrained arrangement of garments, swatches, blank working sheets, measuring tools, and notions for guide covers.
Compose: Use a quiet overhead or shallow oblique view, tactile neutrals, one madder or weld accent, and enough negative space for responsive cropping.
Sparse, functional, and subordinate to language. Use simple currentColor line icons or the existing typographic arrow and chevron vocabulary.
| Use | Ratio | Crop rule |
|---|---|---|
| Homepage or feature image | 16:9 | Keep the primary subject away from the text-safe side and verify the crop at 390px and 1440px. |
| Guide cover | 4:3 | The image must survive a 16:10 card crop without losing the instructional subject. |
| Material or stitch reference | 1:1 | Show repeat, scale, and surface without unrelated props. |
Choose components by information relationship and next action, never to fill a grid.
| Component | Purpose | Non-negotiable rule |
|---|---|---|
| Page intro | Name the page, state what it helps accomplish, and expose the primary next action. | Use a compact left-aligned stack. Operational references, ledgers, coverage maps, and taxonomy pages use --step-3; feature, guide, and source-record titles may use the bounded --hero-title token. Supporting text stays near the title and within 66ch. |
| Button | Trigger or navigate to a clear outcome. | Use one filled primary action per view when there is a true priority. Maintain a 44px minimum hit area. Use verb-first labels. |
| Text link | Navigate within prose, metadata, and secondary actions. | Use the madder token and a visible underline where context alone does not make the link obvious. |
| Resource row | Expose name, job, geography, source type, access, evidence date, and limitations without hiding the comparison basis. | Prefer rows or compact cards. Show one primary job and every evidence-backed additional job; do not duplicate a maintained source to force it into another task filter. Do not use provider logos, stars, rank, or approval badges unless the underlying evidence explicitly supports them. |
| Source model row | Explain which owner or maintainer can answer a question before someone enters a large resource collection. | Expose when to use the model, what it can establish, what it cannot establish, the next check, and representative live routes. Treat the model as an evidence boundary, never a quality tier, score, or ranking. |
| Operational fact strip | Expose the date, cadence, state transition, or rule needed to act on an operational reference. | Use compact definition rows with descriptive values. Show workflow facts such as the next due date or append-only rule, never catalogue totals, unsupported quality measures, or promotional proof. |
| Maintenance work queue | Turn published review dates into ordered work that a maintainer or agent can reproduce without reconstructing the process from prose. | State the as-of date and due threshold, lead with overdue or due-soon work, show an honest nothing-due state, and attach the current record, source, dates, review focus, portable packet, allowed outcomes, and append-only finish rule. Queue status is never a provider score. |
| Evidence note | Place source scope and decision boundaries near a consequential claim. | Use plain headings such as Evidence, Limitations, or Checked on. Do not invent archival or editorial labels. |
| Record usefulness check | Ask whether a permanent Stitch Authority record supported the visitor's next decision without rating the external source. | Use one neutral three-outcome question, bounded reason choices, and honest sending, received, and retry states. Send no name, free text, account, persistent identifier, or business information. Treat responses as anonymous interaction counts that never establish provider quality, unique users, or qualified participant validation. |
| Research participation funnel | Move an eligible volunteer from invitation to a consented realistic task without hiding the study boundary or front-loading the complete operating protocol. | Use a compact task-first introduction, direct track-to-runner links, participant-centered track labels, a concise data summary beside positive consent, and progressive disclosure for the complete method and withdrawal terms. Keep self-guided discovery, moderated evidence, internal QA, listed providers, and represented practitioners separate. |
| Research task invitation | Offer an optional self-guided task check at the point someone is already using the product, with a measurable handoff into consent. | Keep the prompt compact and adjacent to the active task. State time, activity, and omitted data in plain language; use participant-centered track choices; preserve only a bounded aggregate entry point; and distinguish an invitation click, a consented start, self-guided discovery, and qualified moderated validation. Never use clicks as participant counts or evidence of completion. |
| Filter chip | Apply one visible catalogue constraint. | Show selected state with shape, text, and color. Preserve state in the URL. Keep each target at least 44px high. |
| Data table | Compare repeated fields or exact values. | Use headers, captions, tabular figures where helpful, and a contained horizontal-scroll region on small screens. |
| Empty or error state | Explain why no useful result is visible and provide a recovery path. | Preserve user input, name the constraint in plain language, and offer one primary correction plus an optional browse route. |
| Site navigation | Expose the main jobs and preserve orientation across a large reference collection. | Use descriptive destination labels, keep the logo linked to home, and move secondary destinations into grouped menus or the footer before compressing primary labels. |
| Search result | Show why a page, tool, or source record matches a query and make result types distinguishable. | Lead with the title and useful excerpt, show the content type in text, preserve the query, and never manufacture relevance with a popularity badge. |
| Working template | Turn a guide into a record someone can complete, print, save, or hand off. | Use 16px fields, explicit labels, 44px actions, local save and clear feedback, print-safe layout, and a clear statement of what the template cannot establish. Never send entered values, field labels, or exported records in behavior events; the measurement payload may contain only the static template ID, action name, and public or internal-QA evidence-lane marker. |
| Disclosure | Keep optional methods, records, and supporting detail available without forcing them through the primary task path. | Use native details and summary semantics, a descriptive label, an optional one-line deck, a visible chevron, a 44px minimum target, visible focus, and an expanded print fallback. Do not hide information required to understand the immediate decision, consent, price, safety, or consequence. |
| Form field | Collect one understandable unit of information with a visible label and recovery path. | Keep labels persistent, instructions close, errors specific, and valid input intact. Do not use placeholder text as the only label. |
| Correction intake | Challenge one published record without obscuring its current state or implying that a prepared packet reached a maintainer. | Start from a searchable current-record selection, show the permanent record and current source, bind identity and decision metadata again on the server, and separate supporting or replacement evidence from the challenged state. Reject private contact details from public packet text. Offer copy and download as portable actions; expose a queue action only after the destination is verified signed-out as public and issue-enabled. |
| Figure | Connect an image, diagram, or specimen to the exact point it helps explain. | Use useful alternative text, a caption when provenance or interpretation matters, and the canonical image ratios. Never use an image as unsupported proof. |
One primary action, one useful alternative, and no ornamental button row.
Use it to find a Norwegian clothing or textile repair lead. Listing presence is not provider approval.
Make comparison, coverage, maintenance state, and measured behavior easier to inspect without presenting totals as proof of usefulness or completeness.
These patterns keep page structure tied to the job. They are required information sequences, not rigid wireframes.
Help a visitor identify the relevant task and reach a useful destination quickly.
Avoid: Promotional inventory counts, unsupported authority claims, abstract mission copy, or oversized cards with one link.
Let an eligible person understand the commitment, choose the closest resource-use track, consent to limited task data, and begin without mistaking the activity for training, provider recruitment, or completed validation.
Avoid: Long protocol text before the first meaningful choice, duplicated page and runner introductions, provider-recruitment language, hidden consent scope, preselected consent, decorative evidence totals, or internal QA presented as participant proof.
Let someone express a need, narrow visible constraints, compare results, and recover from no result.
Avoid: Hidden filter state, rank without a stated basis, provider approval signals, or a dead empty page.
Help someone choose the right kind of source or route before asking them to navigate the complete collection.
Avoid: Giant network diagrams, promotional inventory totals, quality scores, decorative taxonomy cards, or definitions disconnected from working routes.
Let someone inspect current state, understand the maintenance rule, find the next due action, and challenge an earlier decision.
Avoid: Promotional record totals, baseline counts, giant dark mastheads, split section headers, oversized definition cards, or state claims without dates and source routes.
Answer the immediate question, show the decision boundary, and support the next real action.
Avoid: Decorative throat-clearing, false certainty, a generic hero, or source lists separated from the claims they support.
Produce one understandable result while keeping units, assumptions, and reversibility visible.
Avoid: Unlabeled fields, invented typical project values, hidden fit, ease, or reserve adjustments, area scaling presented as an accurate garment or shaped-project quantity, results that appear authoritative without assumptions, destructive reset behavior, or motion that delays the answer.
Document what an external resource is useful for, what was checked, and what remains unverified.
Avoid: Stars, awards, approval badges, provider logos as proof, census language, or claims that Stitch Authority manages listed people.
Let agents and tools reuse a maintained evidence collection without scraping the visual interface or dropping its decision boundaries.
Avoid: Provider rankings, completeness claims, undocumented fields, unstable generated timestamps, silent vocabulary changes, or exports that strip evidence and limitations from names and links.
Loading, empty, error, success, focus, and reduced-motion behavior must be designed before a flow is considered complete.
The control is visibly actionable and its label names the result.
Pointer feedback changes color, surface, or underline without moving layout or carrying essential information.
A high-contrast outline remains visible and unobscured. Focus never depends on hover styling.
Press feedback is immediate and subtle. Use color or transform without changing layout.
The state is exposed semantically and visually. Disabled actions do not masquerade as ordinary muted text.
Keep the action label or outcome context, expose busy state, prevent duplicate submission, and avoid layout shift.
Name the active constraint, preserve input, and provide one correction plus an optional browse route.
State the cause and recovery beside the affected control while preserving valid input.
Confirm what changed, what did not change, and the next realistic action.
Name the active constraint, preserve the query, and offer one correction plus a browse route.
Put the cause and correction beside the field. Never erase valid input.
Confirm what happened, what did not happen, and the realistic next step.
These are product-specific failures, not a mood board. A new implementation must not reintroduce them under another name.
Avoid: Any page title above the 52px desktop, 46px tablet, and 36px mobile caps.
Why: Scale should establish hierarchy without turning one sentence into the whole viewport.
Avoid: A monumental title isolated on the left with small supporting copy stranded across an empty row.
Why: The relationship is weak, the reading path is artificial, and the small copy becomes subordinate to empty space.
Avoid: Very tall or wide boxes containing only a label, one sentence, and an arrow.
Why: Component area must be earned by the information or interaction it contains.
Avoid: Copy forced into a sliver so ordinary words stack one per line.
Why: A label is not a decorative spine; it must remain readable at a glance.
Avoid: Promotional resource totals, evidence totals, or price claims used as homepage trust theater.
Why: Inventory describes the database at one time. It does not prove usefulness, quality, completeness, or endorsement.
Avoid: Invented desk names, numbered notes, archival metaphors, and slogans that do not help complete a task.
Why: Labels must orient, not perform expertise.
Avoid: Language implying Stitch Authority recruits, manages, represents, or has approved practitioners or makers.
Why: The product is a resource hub. It may document external populations but does not own them.
Avoid: A centered slogan followed by three equal feature cards, decorative gradients, and unsupported social proof.
Why: Structure should follow textile decisions and working information, not a default generation pattern.
Use the repository-native instruction file when one is available, then read the local canonical source. Outside the repository, fetch the public JSON endpoint. Every adapter must point here instead of restating visual values. Do not infer the system from a screenshot or a single page.
DESIGN.mdsrc/data/design-system.jsonA rel=alternate application/json link points to the canonical machine reference.
The key-pages list names the human reference, canonical JSON contract, and public JSON Schema.
Vendor-specific entry points remain thin adapters that direct local contributors to the same canonical JSON before user-facing work.
Agents can consume the maintained evidence records and exact vocabulary without scraping the rendered directory or inferring fields, then return a correction or source review through record-bound action links.
Agents can find due work, the next scheduled batch, source-specific review focus, allowed outcomes, and portable record packets without calculating cadence or reconstructing the workflow.
src/data/design-system.jsonsrc/app/globals.cssChoose the page pattern first, then use the global title tokens and keep orientation copy in the same reading block.
src/components/ui/Button.tsxsrc/app/globals.cssUse for primary and secondary actions instead of rebuilding action states in a page module.
src/app/globals.csssrc/app/globals.cssUse semantic link color, underline where context is not sufficient, and preserve visible focus.
src/app/directory/page.tsxsrc/components/directory/DirectoryExplorer.module.cssReuse the directory decision metadata and compact comparison order.
src/app/directory/landscape/page.tsxsrc/app/directory/landscape/landscape.module.cssPreserve can-establish, cannot-establish, next-check, and live-route fields.
src/app/directory/maintenance/page.tsxsrc/app/directory/maintenance/maintenance.module.cssUse for dates, cadence, next due work, and append-only rules rather than catalogue totals.
src/app/directory/maintenance/page.tsxsrc/app/directory/maintenance/maintenance.module.cssReuse the as-of status, due threshold, compact work rows, machine queue, source handoff, portable packet, honest empty state, and append-only outcome contract.
src/app/directory/[id]/page.tsxsrc/app/directory/[id]/resource-record.module.cssKeep evidence, limitations, checked date, and verification instructions adjacent.
src/components/directory/ResourceUsefulnessFeedback.tsxsrc/components/directory/ResourceUsefulnessFeedback.module.cssUse only after the record's evidence and limitations. Ask about the record, use closed response codes, confirm server receipt before showing success, and keep the signal outside moderated validation.
src/components/research/FieldTestRunner.tsxsrc/components/research/FieldTestRunner.module.cssStart from the participant page's direct track links, keep the data summary and positive consent before the begin action, put the full protocol in progressive disclosure, and preserve the self-guided, moderated, internal-QA, and withdrawal evidence boundaries.
src/components/research/ResearchTaskPrompt.tsxsrc/components/research/ResearchTaskPrompt.module.cssPlace beside an active product task, offer bounded track choices, disclose time and omitted data, record only intentional handoff clicks, and preserve entry attribution only after positive consent in the validation lane.
src/components/ui/FilterChips.tsxsrc/components/ui/FilterChips.module.cssReuse for a small visible filter group; use directory controls for complex faceted filtering.
src/data/design-system.jsonsrc/app/globals.cssBuild from semantic table elements and the data-display rules; keep exact values and relationships intact on small screens.
src/app/search/page.tsxsrc/app/search/search.module.cssName the active constraint, preserve the query, and offer a correction plus a browse route.
src/components/Header.tsxsrc/components/NavMenu.module.cssPreserve desktop and mobile navigation, keyboard behavior, descriptive labels, and semantic layer tokens.
src/app/search/page.tsxsrc/app/search/search.module.cssPreserve the query, content-type label, useful excerpt, and exact destination.
src/components/content/WorkingTemplate.tsxsrc/components/content/WorkingTemplate.module.cssReuse for printable or locally saved guide records and retain the privacy-safe event boundary.
src/components/content/Disclosure.tsxsrc/components/content/Disclosure.module.cssReuse for optional long-form methods, worksheets, and supporting detail after the immediate answer remains visible.
src/components/tools/ToolShell.tsxsrc/components/tools/ToolForm.module.cssUse persistent labels, 16px inputs, nearby units and help, specific errors, and preserved valid input.
src/components/directory/ContributeForm.tsxsrc/components/directory/ContributeForm.module.cssReuse the published-record binding, packet preparation, privacy rejection, and delivery-state boundaries.
src/components/content/Figure.tsxsrc/components/content/Figure.module.cssReuse for sourced or instructional imagery that needs alt text, a caption, or provenance.
Read the canonical source before changing user-facing layout, typography, color, components, interaction, or product copy.
src/data/design-system.jsonsrc/data/design-system.schema.jsonsrc/app/globals.cssMigration: Remove universal yardage ranges and prefilled project dimensions or reserves. For sample scaling, require a user-supplied equivalent area, show base and reserve separately, state comparability limits, and hand whole-package planning to the dedicated workflow.
Migration: Consolidate overlapping calculators around the actual user decision. Remove preset project sizes and automatic fit assumptions unless a current controlled source and exact applicability are present; show raw arithmetic and reversible alternatives instead.
Migration: When a calculator receives a query whose requested quantities are not convertible, state that boundary beside the controls, identify the ambiguous term, and route to the maintained explanation instead of manufacturing a formula.
Migration: Use the shared ResearchTaskPrompt where an active research round needs volunteers. Do not count its link events as people, consent, task completion, or validation.
Migration: Use the shared Disclosure component when optional long-form content overwhelms the primary task. Do not use it to conceal information a visitor needs before acting.
Migration: Repository agents should begin with their native instruction file, then follow AGENTS.md to src/data/design-system.json. Do not copy tokens, component rules, or prohibited patterns into a vendor-specific adapter.
Migration: Use --deployment=https://deployment-url for the protected candidate, then run the audit without arguments after promotion. Do not disable deployment protection to make a public fetch pass.
Migration: After deployment, run the production-parity audit against the immutable deployment URL and then the canonical domain. Treat any mismatch as release drift, not as absent behavior.
Migration: Resource-export consumers should adopt schema 2.0.0 and preserve correctionUrl and reviewPacketUrl with each record. Do not treat an opened action as delivered or completed work.
Migration: No visual or manifest-field migration. Consumers should validate the unchanged canonical shape against the stricter public schema; malformed derivative documents and undocumented fields now fail.
Migration: Research invitations and participant entry pages should use the shared field-test flow. Put the track choice and limited-data consent before the full protocol, but retain complete data, retention, recovery, withdrawal, and facilitation information on the same route.
Migration: Automated consumers should validate against /design-system.schema.json, then use agentConsumption.defaultDecisions and taskRecipes before reading the relevant detailed sections. Do not create a separate agent summary or brand guide.
Migration: For maintained directory work, consume /directory/maintenance.json and its per-record packet routes instead of inferring urgency from check dates or reconstructing review outcomes from prose. Keep queue status separate from provider quality.
Migration: For maintained evidence collections that need reuse outside the interface, expose a deterministic versioned export and schema. Do not publish names and links without their evidence, limitations, dates, and claim boundaries.
Migration: Place the shared usefulness check after source evidence and limitations. Use only the closed outcome and reason codes; do not add free text, identity, stars, rankings, or endorsement language.
Migration: No visual migration. Use valid roles when a specimen needs an accessible group or image name; hide decorative swatches when adjacent text already names their values.
Migration: No visual migration. Agents may discover the canonical JSON from any rendered page instead of relying on a known URL.
Migration: Use componentInventory instead of guessing whether a named component is shared. Replace off-scale reference-page labels with the canonical metadata or body token.
Migration: No breaking change.
Migration: No breaking change.