Brand voice guidelines

Version 1.2, 2026-10-20. Owner: Afo.

Scope. Rules for anyone, person or agent, who writes Green Goods copy: product strings, the public website, docs, release notes, announcements, and funder or partner documents. Pull request and issue bodies are developer-facing; they follow the spelling and punctuation rules in section 6.7 but not this page's voice. Audiences by name, numbers, partner work, and anything else that should not sit in a public repository live in the internal strategy addendum (a shared Google Doc), not here.

How people use this file. Read sections 1 to 3 before writing. Run the checklist in section 9 before shipping. When this file and a source of truth disagree, the source wins and this file gets corrected; section 16 lists the sources and section 15 says what each one owns.

How agents use this file.

Routing: the repository agent guide (AGENTS.md, "Read when relevant") sends any task that writes copy a gardener, steward, funder, partner, or visitor reads here first, then to the owning surface DESIGN.md. The task-routing manifest names this page as a required authority for document review and design direction when the task carries brand-facing copy; visual-only design work and internal technical documents do not load it. Codex reaches the same files through the repository's agent guide, and the docs site publishes this page in its LLM index (llms.txt).
The same file, two paths: this page is the only copy. .claude/brand-voice-guidelines.md is a symbolic link to it, which is where the brand-voice plugin's enforce-voice command and other tools expect guidelines to live.
Enforcement: the mechanical rules are checked, not trusted. bun run check --only vocabulary rejects banned and retired terms, UK spellings, and the old Spanish and Portuguese role words in the three shared catalogs (packages/shared/src/i18n); the docs audit rejects UK spellings in docs pages and design guides; bun run check --only em-dashes rejects em dashes in those catalogs, docs pages, and design guides. The chat bot's copy in packages/agent is held to the same vocabulary file by the agent's copy tests (banned and retired terms, UK spellings, and the locked steward and garden words with their retired synonyms; the bot's promise and endowment terms are reviewed by eye), and the em dash check reads its chat copy modules too. Section 9 marks which checklist items a machine covers.
Review: a change to copy is reviewed against this page the way a change to layout is reviewed against the design skill. Dated copy decisions still get a row in the design decision log.

Writing conventions in this file. US spelling. No em dashes anywhere in Green Goods prose. In running text, "garden", the role words, and the domain nouns are lower case; Title Case belongs to titles, action labels, and named acts.

1. What Green Goods is

Brand line (website hero, keep as is): From good intentions to green outcomes.

Descriptor (the one sentence for product-overview metadata: the client site's meta description, the repository README's lead line, the docs site tagline, the deck one-liner; individual pages and package READMEs describe their own content):

Green Goods helps communities name their needs, make promises, document regenerative work, review it together, and turn the record into support.

Today a garden names its needs through an assessment, set by stewards and evaluators. Community members naming needs directly is a planned layer (the claims ledger lists it as planned), so copy about it says so.

Canonical problem statement:

The record of a community's regenerative work stays trapped, and three things keep it that way. Proving the work costs more than doing it. The tools for proving it were built for the office and the grant report, not for the field. And the record ends up on platforms the community does not own, so it cannot follow them to the next funder, partner, or tool. Inside the community, contribution is thanked but not valued. Outside, the people who would support it cannot read it.

Mission, in one sentence: Green Goods makes a community's regenerative work visible to the people who were not there, so support can follow delivery instead of paperwork.

The loop, in plain words: stewards and evaluators name what the garden needs and what counts as success in an assessment, members make promises, gardeners document the work, stewards review it and record their decision separately, and funders support more of it. Assessing what changed against that baseline is a planned step, so copy does not describe it as live.

What Green Goods is not (say so when a reader might assume otherwise):

Not a payroll. Approving work does not pay anyone. How a garden rewards its people is a separate step that each garden decides.
Not an audit. A steward's approval is a local decision by someone who knows the place, not an outside certification of ecological outcome.
Not a promise of a charitable receipt, nonprofit backing, or a tax deduction. A garden may have its own arrangement; readers check with the garden.

The current, generated list of what the product can and cannot do is the claims ledger (docs/claims.generated.md, generated from the ontology projections and kept out of the public site). Copy never claims more than that ledger does.

2. Who we write for

Five product personas carry fixed meanings in the ontology. One more audience appears in external copy.

Audience	Who they are	Tone	Avoid	Technical vocabulary allowed
Gardener	Does and documents the work, often on a shared phone with weak signal, in the app or by chat	Warm, direct, action-oriented; short sentences	Technical terms, passive voice, option overload	None in ordinary copy; only the DL-094 safety cases (a privacy or permanence notice may name the network, a link names its explorer). About a 6th-grade reading level
Steward	Runs the garden: the season's plan, members, and review of work	Professional, efficient, task-focused, data-aware	Condescending simplicity, unstructured prose	Steward docs may describe on-chain review as a trust mechanism; product review screens follow DL-094; never contract internals
Evaluator	Sets baselines and evidence standards before work begins; assessing what changed against them is planned	Precise, evidence-based, structured; respects domain expertise	Oversimplification, vague impact language	Research and assessment terminology is fine
Funder (also called supporter)	Anyone who supports the work: an individual giving to a garden, a diaspora supporter, a grantmaker, an institution, a crypto ecosystem fund. They donate, endow, or fund through a grant or partnership, and the ontology's Funder persona means the same	Confident, transparent, evidence-first; lead with reviewed records	Hype, unsubstantiated claims, anything resembling an investment pitch	Traditional funders and institutions: describe Green Goods as a field data and MRV tool, no web3 words. In funder documents and decks for crypto ecosystem funders: attestation and chain words in an appendix, never the headline. On product funding surfaces DL-094 applies: a vault page may name the protocol that holds the deposit. Never "guaranteed returns"
Community member	A local person affected by the work, who confirms work done for them and, once Community Needs ships, names needs directly	Inclusive, plain, encouraging	Protocol jargon, assumed web3 literacy	None in ordinary copy; only the DL-094 safety cases. About a 6th-grade reading level
Partners and builders (external copy)	Ecosystem tools, integrators, researchers	Direct, specific, documented	Marketing superlatives	Protocol and chain names allowed on builder pages and in appendices

Framing rule. For traditional funders and non-technical audiences, Green Goods is a field data and MRV (measurement, reporting, and verification) tool for community-led regenerative work. Web3-native language appears only in crypto ecosystem contexts, and even there it stays in the appendix, never the lead.

Where protocol names may appear (DL-094). A protocol or chain name appears only where the reader needs it to act or to verify: a funding surface (a vault page names the protocol that holds the deposit), a builder page, the labeled "using your own wallet" box a community page may carry, a transaction step (a confirmation that spends gas or cannot be undone, a wallet action on a named network such as "Send G$ on Celo", and the note that follows it), a privacy or permanence notice that names the network making a photo public and undeletable, and a link that names the explorer it opens. Never as positioning, and never in onboarding, work capture, review, or community copy, where the product's own labels carry the meaning. Progress and status copy stays plain ("Recording your decision…", not "Syncing with blockchain…").

Priority rule when two audiences conflict. Gardener experience wins, unless that blocks steward onboarding; stewards are how a garden arrives.

3. Voice

Core voice. Green Goods speaks like a knowledgeable neighbor who runs the community garden. Warm but not gushing. Practical but not clinical. Knows soil science but explains it by pointing at the compost pile.

The one-sentence test. If a line could appear on a government form, rewrite it. If it could appear on a hand-painted garden sign, it is close.

We are / we are not

Pillar	We are	We are not
Grounded	Concrete, specific, rooted in real action: one tree, one bag, one panel, one class	Jargon-heavy, academic, blockchain-first
Inviting	Welcoming, assumes good intent, lowers barriers, gives permission where people hesitate	Sycophantic, over-enthusiastic, exclamation-heavy
Honest	Transparent about what works and what is still experimental; says the limit up front	Hedging, corporate disclaimers, vague promises
Active	Action-oriented, present tense, you-centered	Passive voice, abstract nouns, bureaucratic

Three dialects, one voice, and a fourth to come

Dialect	Surface	Speaks	Register
Admin dashboard	The steward workspace (`packages/admin`; the design skill calls the same surface the cockpit)	About the work, to the steward	Utility: status language, task framing, counts and states
Editorial website	The public browser site	To visitors and funders, about the gardens	Editorial: place-first, evocative headlines, numbers that breathe
PWA app	The installed garden journal (`packages/client`)	To the community, about the work	Warm and narrative; keeps drafts on the phone they were written on; never blames
Community app (future)	A surface for community members to name needs, witness work, and confirm promises	To the neighborhood, about the place	Not yet defined. It inherits the voice and the community-member tone row; its dialect gets written when the surface is designed

Section 5 has the patterns for each dialect and says which dialect other surfaces borrow.

Voice at the brand level. Green Goods speaks as Green Goods. No "built by" byline, no invitations to join any other organization, no network-wide link lists. Closing sections read "Stay in touch". The origin story may say that Green Goods grew out of the Greenpill network of builders and local communities, and nothing more.

4. Tone by context

Tone shifts by context while the voice stays constant.

Context	Tone	Model line
Onboarding	Encouraging, patient	"Start by describing what you see. We'll help with the rest."
Submitting work	Supportive, clear	"Add a photo and a few words about what you did today."
Hero moments in the app (first work submitted, a promise kept, a season closing)	Celebratory, genuine; one sentence, no confetti language. Admin completions (a garden created, a certificate minted) end in the admin register: a status line and Done, no celebration	"Your first contribution. This garden is growing because of you."
Errors	Calm, constructive; never blames the person; says what is safe	"That didn't go through. Your work is saved. Try again when you're ready."
Offline	Reassuring, matter-of-fact; promises only what the queue holds, and the queue waits for Upload all	"You're offline. Your work is saved on this phone. When you're back online, open Your Work and tap Upload all." Operations that need a connection (creating an assessment or a garden) say so instead: "Creating an assessment needs a connection. Nothing is lost." In the installed app a steward's decision on work, with a passkey or a wallet, queues like work does: "Your decision stays on this device until you upload it." In the admin cockpit a wallet decision needs a connection, and copy says so.
Admin and steward	Efficient, status-oriented	"3 submissions pending review. 1 flagged for follow-up."
Funding and impact	Respectful, concrete; numbers breathe	"This garden has documented [count] actions across [seasons] seasons." (both values from live data)
Failure states (a page or screen cannot load)	An ordinary page: a title, one sentence, one action; nothing lost, no codes	"This public record is temporarily unavailable. Please try again."
Docs (community and builder)	Second person, active, plain; why before what; one idea per sentence	"Your garden's stewards review your work. There is no fixed time."
Announcements and Green Goods channels	Warm, plain, pilot-focused; states limits up front; invites replies; thanks testers; at most one or two emoji; never "Gm" or "LFG"	"We're still pilot focused, learning alongside the communities using Green Goods."
Funder and partner documents	Evidence-first; claims carry a source tag ([D] documented, [I] inferred, [S] planned); chain detail in an appendix	"Approval is a steward's decision. It is not a payment and not an outside audit."
Chat reporting (Telegram bot)	Short, first person, one question at a time; says how many quick questions remain; shows a summary to check before publishing	"Two quick questions left, then I'll show you the summary."

5. The three dialects, and the surfaces that borrow them

Admin dashboard (steward workspace)

Utility copy only: status language and task framing, never marketing or narrative.
Headers are functional nouns: "Review Queue", "Garden Settings", "Member Activity".
Status text is counts plus state: "3 pending · 1 flagged · 12 approved".
Actions are verb-first and Title Case: "Approve", "Request Changes", "Flag for Review".
Empty states are terse: "No items in queue", not "Get started by adding…".
Outcome is the change the garden wants to see, one sentence about people or land. Metric is what is counted. Target is how much by the period's end. Define each once, above the form.
Don't say: "Welcome back!" / "Let's review some great work" / "You're crushing it".
Hero-moment, gallery, and celebration language stays out of the admin. The live list is prompt_vocabulary_admin_banned in scripts/data/banned-vocabulary.json.

Editorial website

Headlines are evocative and magazine-weight; subheads bridge feeling and information.
Garden descriptions lead with the place and what it grows, never with the technology.
Impact numbers breathe: "[count] documented actions" (a live figure) beats a sentence about how many actions community members documented.
Calls to action are warm imperatives: "Fund this garden", not "Contribute now".
Hero titles carry no closing period.
A failed page looks like the website: the banner hero, a title, one sentence, a reload link.

PWA app (garden journal)

Instructions are direct and second person: "Describe what you did", not "Please provide a description."
Labels are short nouns: "Photo", "Method", "Confidence".
Empty states encourage and suggest the next action: "No work yet. Start by documenting what you see."
Sync status is calm: "Saved locally", "Synced", "Syncing…". Recording work and deciding on work queue offline; creating an assessment or a garden needs a connection, and copy says which.
Errors never blame the person: "We couldn't submit that", not "Your submission failed."
A draft stays on the phone it was written on while the app and its data remain, and copy says so when it matters (keep the app installed, do not clear its data). Never promise more than that.
Say: "Let's see what's grown in your garden." / "Your work is in good hands." / "Go do more of what matters."
Don't say: "Processing submission..." / "Review queued." / "1 of 12 pending." / "Submit another?"
Cockpit, dashboard, and KPI language stays out of the app. The live list is prompt_vocabulary_client_banned in scripts/data/banned-vocabulary.json.

Community app (future)

Not yet designed. When it is, its dialect is written here before its first strings ship, using the community-member row in section 2.

Surfaces that borrow a dialect

Surface	Borrows	Adjustments
Docs, community track	Editorial website, made plain	Second person, active, about an 8th-grade reading level; why before what; condition before instruction; one idea per sentence; sentence-case headings; contractions fine; no superlatives. The product's own labels are the vocabulary; protocol and chain names appear only where DL-094 allows them (the wallet box, a funding or transaction step, a privacy or permanence notice, a link that names its explorer). Permission-giving where people hesitate; strict where the integrity of the record is at stake. Money and privacy statements describe what the code and contracts do and never repeat a guarantee from interface copy. Task pages end with next steps, and links say where they go
Docs, builders track	Docs community register, for developers	Same rules; generated reference pages stay terse; protocol names allowed
Release notes	Editorial website, trimmed	One plain opening sentence on what the release does for people; sections by audience ("For gardeners", "For stewards"); a short "Protocol and platform" section; links. Deferred or disabled capabilities are stated in the same note
Announcements and community posts	PWA app warmth, website clarity	One-line opener, what changed in two or three short paragraphs, an invitation to try it and reply, thanks, a link. Spanish and Portuguese versions are reviewed by a speaker before posting
Chat reporting bot	PWA app, in conversation	Short first-person turns, one question at a time, a count of what remains, a summary to check before publishing. Gardeners and stewards message the bot directly to report; other roles, including owner-only accounts, are not supported yet
Funder and partner documents	Editorial website, evidence-first	Claims carry source tags; chain detail in an appendix; short sections under headers, two or three sentences each

6. Terminology

6.1 Core terms, in plain words

The Glossary is generated from the ontology and is the authority; this table is the short form a writer needs.

Term	Meaning
garden	A community and its shared place. It gives people, roles, records, and funds one accountable home. It does not have to involve plants: a solar team, a waste crew, a classroom all qualify.
action	A reusable work instruction. It makes expected work and evidence clear before anyone submits.
work	A field record of completed activity, with notes and photos.
work approval	A steward's recorded decision on submitted work, kept separate from the record of what happened. Stewards approve work; evaluators assess, they do not approve.
assessment	A starting point and a plan: what a place needs and what good evidence should show. Assessment comes before work, not after it.
impact certificate	One claim that bundles a body of approved work. Turning approved work into a certificate is in build (the claims ledger says so), so copy describes what a certificate represents, not a funding path that is live.
vault	A garden's long-term fund. A deposit can stay in place while what it earns supports ongoing work; a strategy or contract loss can reduce what comes back or delay a withdrawal, and copy says so wherever it invites a deposit.
cookie jar	A shared fund for near-term needs, with a limit on each claim.
promise	What a member offers or asks for in a pool: a workshop, a planting, a repair. "Promise" is the word in every user-facing surface, in all three languages. "Commitment Pooling" remains the feature name, and the ontology's formal entity names (Commitment Pool, Commitment Cycle, Commitment Contributor) keep the word on builder pages and in the glossary's builder reference; everywhere else copy says promise, and "commitment" otherwise stays in code and contracts.
pool	The container that holds a garden's promises. "Pool" names only the container.
season, campaign	A garden's cycle of promises and work. A garden's season is its pool's Season-type cycle. Their details sheets show only what the record holds.
gardener, steward, evaluator, funder, community member	The five personas. Definitions in section 2. The ontology's garden roles also include owner, which has no persona of its own; name it when enumerating roles or permissions.

6.2 Use this, not that

Use	Don't use	Why
garden	project, organization, DAO	Gardens are the metaphor
gardener	user, contributor, member (for the role)	People who do the work have a name
steward	admin, manager, operator	They steward the garden; "operator" is retired
evaluator	reviewer, auditor, assessor	They evaluate impact, not compliance; a steward, not an evaluator, reviews and approves work
funder, supporter	investor, backer	Funding a garden, not buying a return. A person who gives through the Donate path is still a funder
community member	visitor, viewer, spectator	Part of the community, not an audience
work	task, activity	Regenerative work is the core concept. "Submission" names the act of sending work and the Work Submissions list, never the work itself
action	action type	The thing a gardener can do. "Action template" names the browsable template records on the public site
promise	commitment (in user-facing copy)	The plain word for what a member offers
fund (verb)	invest	Funding a garden. "Fund" is the call to action; "contribute" is the umbrella word for giving before a supporter picks the Donate or Endow path
donate	(allowed only as the name of the direct-giving path)	The path a supporter takes to give directly; "fund" everywhere else
impact	result (as a vague synonym)	Bridges community and chain. "Output" and "Outcome" are stages of the impact model (activity, output, outcome, impact) and keep their meaning; "Metric" stays the name of the assessment field (Outcome, Metric, Target)
document (verb)	log, capture	Gardeners document their work. "Record" is fine in plain explanation of what the product keeps
approved submissions	approved work (as a public count label)	Public count label
garden members	hands at work, participants	Public count of gardeners and stewards, each address counted once
waste management	cleanup	Domain name
solar (hub development)	hub	Domain name
agroforestry	agro (in prose)	"Agro", "Edu", "Solar", and "Waste" are the compact domain labels on chips and short lists; prose and headings use the full names
domain	vertical	Taxonomy term

6.3 Retired words

"Operator" and "operators" in user-facing role copy, in all languages. The deployed role identifier keeps the old name as code, and the glossary keeps "Garden Operator" as an explanatory alias so builders can map older records; neither is a license to write it in copy. "Commitment" in user-facing copy, except inside the feature name and the ontology's formal entity names on builder surfaces. "Approved work" as a count label. "Hands at work". "Mutual credit" as a domain name in copy (see 6.5). "Dotación", "dotação", and the invented verbs "endonar", "endotar", "endowar", "endossar" for endowment. The vocabulary check rejects the retired role words in every catalog.

6.4 Banned vocabulary

Rejected by the vocabulary check in every i18n string: streak, countdown, leaderboard, FOMO, urgent, limited time, re-engagement, retention hook. These signal growth-hacking patterns that conflict with the regenerative lens.
Nowhere: "guaranteed returns", "passive income", or any phrase that promises the supporter a return or frames a gift as an investment. Describing how an endowment earns (a vault's investment earnings fund the work) is allowed where the deposit's risks and the loss limitation sit beside it, as the shipped endowment copy does.
Surface-specific design and prompt vocabulary (what stays out of the admin, what stays out of the app) lives in scripts/data/banned-vocabulary.json and is summarized in section 5. The app and cockpit unification plan may revise those lists; the file is the authority, not this summary.

6.5 Domains

Copy names four domains: agroforestry, waste management, solar (hub development), and education, the same four the ontology, the contracts, and the app carry. Mutual credit is not a domain. The owner retired it; commitment pooling and promises serve the need it named, and no catalog carries a label for it. A partner integration's service-credit instrument gets its own name when that integration ships; it is not a domain.

6.6 Capitalization

Running text: garden, gardener, steward, evaluator, funder, and community member (DL-090), and the domain nouns work, action, assessment, promise, pool, season, campaign, vault, cookie jar, and impact certificate (DL-093) are lower case, in English, Spanish, and Portuguese ("this garden", "este jardín", "o responsável", "your work", "a new promise").
Title Case: card, section, dialog, flow, and step titles; action labels (buttons, menu items, aria-only action labels); named acts ("Create Garden", "Submit Work", "Request Changes"); and a garden's own name ("Community Garden"). Spanish and Portuguese keep native sentence-style casing even in these.
Sentence case: status and outcome copy ("Connection lost", "Update needs a restart"), field labels, helper text, placeholders, kickers, body copy, docs headings.
Hero titles on the public site carry no closing period.
Plain titles, methods in helper text: a step or field title is a plain question or noun ("What You'll Measure"); a framework or method name appears only in parentheses in the helper text. A flow never preselects a choice the person has to make.

6.7 Spelling and punctuation

US spelling in all English copy: neighbor, color, center, organize, canceled, gray, license, behavior. The vocabulary check rejects UK forms in catalogs; the docs audit rejects them in docs pages and design guides. Quoted third-party names keep their own spelling.
No em dashes anywhere: prose, product strings, docs, release titles, announcements, in any language. Restructure with periods, colons, commas, or parentheses, never a hyphen or a spaced en dash. bun run check --only em-dashes enforces it in the catalogs, the bot copy, docs pages, and design guides; release packs under docs/releases/ are reviewed by eye. Canonical ontology definitions keep their existing punctuation, and a lone em dash standing in for a missing or unreadable figure in a table or metric is a data placeholder, not prose, and stays.
Contractions are welcome in app and docs copy. Exclamation marks are rare: one per screen at most, never in errors or status.
Ellipsis for in-progress states ("Syncing…") and, on an action label, to say the action opens another step before anything happens ("End Season…", "Archive…"); never in prose.

6.8 Numbers

Live product counts (the public site's own figures) may appear anywhere.
Hand-typed figures: at most one owner-cleared headline figure, reviewed each quarter. No other garden counts, country or continent counts, or traction figures in any copy. Until a figure is cleared, say "pilot communities".
Empty or not-yet-started records stay quiet. Never present an invented statistic.

7. Writing style rules

Concrete over abstract. If the reader cannot picture it, rewrite it.
Active voice, present tense, second person wherever a person is acting.
One idea per sentence. Short does not mean clipped: a sentence beats a label with a colon.
Why before what; condition before instruction.
Blockchain is implementation, not copy. On-chain words appear only where section 2 allows them.
External copy is short and scannable: a one-line intro, headers, two or three sentences per section, hyphen bullets in plain-text fields. People do not read dense things.
Errors state what is safe ("Your work is saved") before what to do next. No codes, no stack traces, no blame.
In the app, every empty state suggests the next action in one sentence. On the website, an empty state suggests an action only where the visitor has one; a public record with nothing yet stays a quiet "nothing here yet" line, never an invented statistic or an unrelated call to action. Admin empty states stay a terse acknowledgment ("No items in queue"), as section 5 says.
Authored community and builder task pages end with next steps, and links say where they go. Reference and generated pages stay terse.
Permission-giving where people hesitate; strict where the integrity of the record is at stake.
Product strings are localized: a new string in the shared catalogs or the bot copy ships with Spanish and Portuguese, accents intact, locked terms used. The docs site is English only today. Community announcements get Spanish and Portuguese versions when those communities are the audience, reviewed by a speaker.
Shorter than your first draft.

8. Claims discipline

Never state or imply:

that approving work pays anyone, or that approval proves an ecological outcome;
that settlement, contributor payouts, or new reward promises are live when a release defers them;
that federation between pools exists or works;
a charitable receipt, nonprofit backing, or tax deductibility;
instant, self-service, or universal availability; Green Goods is pilot-focused until it is not;
automatic impact verification or AI verification of work;
"guaranteed returns", yield promises, or any framing of a gift as an investment for the supporter; saying that a vault's earnings fund the work is fine when the risks and the loss limitation sit beside it.

Always:

describe what the contracts and code do, and label fixtures, illustrations, and synthetic screenshots as such;
state a limit in the same breath as the capability ("Approving work does not pay anyone; each garden decides rewards separately");
keep claims traceable: in funder and partner documents, tag each claim [D] documented, [I] inferred, or [S] planned.

9. Pre-ship checklist

Items marked (machine) are covered by a check in the validation pipeline; the rest are judgment calls.

10. Model lines

Drawn from shipped copy and approved release material. Lines marked "adapted" were adjusted to the casing or punctuation rules above.

Line	Source	What it shows
"Already part of a garden? Update the app, give it a try, and tell us what feels easier, or where you get stuck."	2.0 announcement (adapted)	Inviting, plain ask
"Empty or not-yet-started records stay quiet rather than presenting invented statistics."	2.0 release notes	Honest
"We're still pilot focused, learning alongside the communities using Green Goods."	2.0 announcement	Limits up front
"Start by describing what you see. We'll help with the rest."	DESIGN.md tone spectrum	Patient onboarding
"Add a photo and a few words about what you did today."	DESIGN.md tone spectrum	Supportive instruction
"You're offline. Your work is saved on this phone. When you're back online, open Your Work and tap Upload all."	DESIGN.md tone spectrum	Reassuring, bounded to the queue and its Upload all step
"3 submissions pending review. 1 flagged for follow-up."	Admin copy voice	Status, no narrative
"We couldn't submit that."	Client prompt contract	Error without blame
"Connect the wallet you endowed with, and we'll show you the gardens your support is growing."	Public fund page, endowments view	Warm, concrete, no custody claim
"The people who do the work should be the ones who are seen for it."	Why We Build	Belief, plain
"Good work often leaves no trace."	How It Works	Grounded opening
"This public record is temporarily unavailable. Please try again."	Browser design guide	Calm failure state
"No pudimos mostrar esta página en este momento. Recarga para intentarlo de nuevo."	es.json, page error	Spanish failure voice
"Não foi possível mostrar esta página agora. Recarregue para tentar novamente."	pt.json, page error	Portuguese failure voice

Instead of this, write this

Instead of	Write
"Processing submission..."	"Your work is in good hands."
"Your submission failed."	"We couldn't submit that. Your work is saved."
"Welcome back! Let's review some great work."	"3 submissions pending review."
"Contribute now"	"Fund this garden"
"Please provide a description."	"Describe what you did."
"Make a commitment to the pool."	"Make a promise."
"Verified on-chain impact for DAOs"	"Work reviewed by the people who know the place."
"Join the guild"	"Stay in touch"
"Operators approve work."	"Stewards review work and record their decision."
"13 live gardens on five continents"	"Built with pilot communities."
"Hands at work: 24"	"Garden members: 24"

11. Localization

The product ships in English, Spanish, and Portuguese: a new string in the shared catalogs or the bot copy is not done until its Spanish and Portuguese strings exist, with accents intact (an accents guard is planned; until it lands, review by eye). The docs site is English only today.
Locked terms, applied on every surface including hero copy, count notes, and the chat bot. The vocabulary check (shared catalogs) and the agent's copy tests (bot chat copy) enforce the steward and garden words and reject their retired synonyms; the promise and endowment terms are checked in the shared catalogs by the locale-coverage test and reviewed by eye in the bot chat copy. Steward is "responsable" (es) and "responsável" (pt); garden is "jardín" (es) and "jardim" (pt); endowment is "provisión" / "proveer" (es) and "provisão" / "provisionar" (pt); promises are "promesas" (es) and "promessas" (pt). The check also rejects the old words: coordinador, custodio, administrador, and huerto in Spanish; coordenador, zelador, and horta in Portuguese; the English word "garden" left untranslated.
Spanish and Portuguese keep native sentence-style casing everywhere, including titles and action labels.
Gender follows the noun: "promesa" and "promessa" are feminine, so articles, adjectives, and pronouns that point back at a promise agree with it, including count strings beside the noun. The verb for taking a promise up is "aceptar" / "aceitar".
The same voice applies: warm, plain, no blame, no em dashes, no protocol names where English has none.
Tone beyond translation (warmth, formality, regional vocabulary) gets a one-page addendum written with a native-speaking steward from the product team. Until it exists, translators follow this page and the locked terms.
Community announcements in Spanish or Portuguese are reviewed by a speaker before posting.

12. Visual identity, in brief

Voice copy only needs to know the following. Everything visual belongs to the design system. The app and cockpit unification plan (.plans/backlog/app-cockpit-unification/) may change product materials, surfaces, and names; nothing in this page depends on those outcomes.

The visual kit is the Brand kit page, with data in docs/src/data/brand-kit-v3.json. Logo and symbol color is the canonical dark green #367D44, with white and black monochrome versions. A UI accent is not a different logo color.
Kit typography pairs Fraunces (headlines, editorial moments) with Inter (body, labels). Which typeface each product surface uses is the design system's call, not this page's.
The product's design language is Warm Earth; its tokens, materials, and motion live in DESIGN.md and .claude/skills/design/language.md. Copy never describes a material ("glass", "liquid") or a layout pattern; those words belong to design prompts and are governed there.
Open visual items (clear space, minimum sizes, horizontal and wordmark-only logos, secondary colors, usage terms) are tracked on the brand kit page.

13. Confidence by section

Section	Confidence	Basis
1 Positioning	High	Owner-approved wording; shipped Welcome, FAQ, the claims ledger
2 Audiences	High	Ontology and product context; the framing rule and protocol-name placement are owner decisions
3 Voice and dialects	High for voice and the three dialects; the fourth is a placeholder by design	DESIGN.md; admin and client contracts; owner direction
4 Tone by context	High for the eight product rows; Medium for docs, announcements, funder documents, and chat rows	Product rows are canonical; the others rest on one approved release pack and two contracts
5 Dialect patterns	High for the admin, website, and app; Medium for the bot	Admin and client prompt contracts, the docs guide, the decision log, and the bot's shipped reporting copy (`packages/agent/src/services/reporting/copy/`)
6 Terminology	High	Ontology, glossary, banned-vocabulary file, decision log, owner decisions
7 Style rules	High	Contracts and owner feedback
8 Claims discipline	High	The claims ledger, release notes, FAQ
10 Model lines	High	Shipped or approved copy
11 Localization	Medium	Locked terms are codified and checked; tone adaptation is assigned, not written
12 Visual identity	High at this level of detail	Brand kit page and kit data

14. Open items

Legal attribution. Brand-facing mentions of other organizations are gone from docs chrome and contributor copy. The docs footer names Green Goods. The copyright holder line on the License page stays until the operating entity is settled; then the holder is decided once.
Community app dialect. Written when the surface is designed, before its first strings ship.
Spanish and Portuguese tone addendum. Owned by a native-speaking steward on the product team; one page; due with the next localization pass.

15. Relationship to other guidance

This page is the entry point for voice. It restates only what a writer needs to apply a rule, and points to the owner for everything else.

Guidance	Owns	This page's relationship
`DESIGN.md` § Voice & Copy	The voice core: statement, pillars, tone spectrum, terminology table, writing checklist, and the dated casing, spelling, and punctuation rules	Restated here in sections 3, 4, 6, and 9 so a writer has one place to read; DESIGN.md wins on any divergence
Design skill (`.claude/skills/design/`: `language.md`, `prompt-contract.md`, `client-prompt-contract.md`, `decision-log.md`, `review-checklist.md`)	Surface dialects for AI design tools, materials, motion, prompt vocabulary, dated decisions	Section 5 summarizes the copy half of each dialect; visual and prompt rules stay there; copy decisions still get a decision-log row
App and cockpit unification plan (`.plans/backlog/app-cockpit-unification/`)	Future material, layout, and naming changes across the admin and the app	This page names dialects by what they say, not how they look, so it survives the unification; if surfaces are renamed, section 3's table follows
Ontology, generated glossary, `banned-vocabulary.json`	Terms, personas, domains, enforced vocabulary	Section 6 is a reading of them; the files are machine-checked and win on any divergence
Docs guide (`docs/AGENTS.md`)	Docs page conventions	Both docs registers, community and builder, are written out in section 5, which owns them; the plan hubs that supplied those rules are not authorities once they close
`.claude/context/product.md`	Persona tone rows and the audience priority rule	A row describes how a persona arrives and the tone that fits, never what the product promises; the claims ledger and section 8 outrank it
Brand kit page and kit data	Visual identity	Section 12 is a pointer
Release packs (`docs/releases/`)	Approved copy per release	Section 10 draws model lines from them; each release pack applies this page
Internal strategy addendum (shared Google Doc)	Audiences by name, numbers, channel plans, partner-specific registers, byline and entity decisions	This page carries the rules those decisions produce, never the names or figures
General prose skills available to agents (for example a humanize-writing skill)	Clarity, concreteness, natural prose	Apply on top of this page; this page supplies the brand-specific rules they lack

16. Sources of truth

Source	Owns
`DESIGN.md` § Voice & Copy	Core voice, pillars, tone spectrum, terminology table, writing checklist, casing, spelling, and punctuation rules
`.claude/skills/design/prompt-contract.md` and `client-prompt-contract.md`	Admin and app dialects, say and don't-say pairs, prompt vocabulary
`packages/shared/src/ontology/green-goods-ontology.json`	Entities, personas, domains, allowed surfaces
`docs/docs/reference/glossary.generated.mdx`	Plain-language definitions (generated)
`scripts/data/banned-vocabulary.json`	Banned terms, retired terms, US spelling list, locked locale terms, prompt vocabulary
`packages/shared/src/i18n/en.json`, `es.json`, `pt.json`	Shipped strings and locked translations
`packages/agent/src/services/reporting/copy/en.ts`, `es.ts`, `pt.ts`	The chat reporting bot's shipped copy
`.claude/skills/design/decision-log.md`	Dated copy decisions (DL-012, DL-013, DL-047, DL-071, DL-077, DL-079, DL-083 to DL-085, DL-087, DL-090 to DL-094)
`.claude/context/product.md`	Persona tone rows and the audience priority rule, outranked by the claims ledger on what the product promises
`docs/docs/community/welcome.mdx`, `why-we-build.mdx`, `how-it-works.mdx`, `docs/docs/reference/faq.mdx`	Public explanatory voice and the "is not" claims
`docs/claims.generated.md`	What the product can honestly claim (generated, kept out of the public site)
`docs/releases/` (the 2.0.0 announcement and release notes today)	Curated release notes and approved announcement copy per release
`docs/docs/reference/brand.mdx`, `docs/src/data/brand-kit-v3.json`	Visual identity

1. What Green Goods is​

2. Who we write for​

3. Voice​

We are / we are not​

Three dialects, one voice, and a fourth to come​

4. Tone by context​

5. The three dialects, and the surfaces that borrow them​

Admin dashboard (steward workspace)​

Editorial website​

PWA app (garden journal)​

Community app (future)​

Surfaces that borrow a dialect​

6. Terminology​

6.1 Core terms, in plain words​

6.2 Use this, not that​

6.3 Retired words​

6.4 Banned vocabulary​

6.5 Domains​

6.6 Capitalization​

6.7 Spelling and punctuation​

6.8 Numbers​

7. Writing style rules​

8. Claims discipline​

9. Pre-ship checklist​

10. Model lines​

Instead of this, write this​

11. Localization​

12. Visual identity, in brief​

13. Confidence by section​

14. Open items​

15. Relationship to other guidance​

16. Sources of truth​