name.

You rarely have one name to make. You have a set — tiers of a plan, modules of a platform, agents in a pipeline — that should read as a family: obviously kin, never twins. The reliable way there is to pick one source domain whose internal structure mirrors how your parts relate, and let the names fall out of it. Set the relation; the generator does the falling-out.

The rest of this page is the thinking behind that box — why naming a set is a different problem, and why a single metaphor solves it.

01The shape of the problem: name a set, not a name

Most naming advice is about a single name — a company, a product, a feature. That advice is real but incomplete, because branded systems almost never name one thing. They name a constellation: tiers of a plan, modules of a platform, the agents in a pipeline, the rooms of an app. When those names are generated independently, the result is a zoo — one part named for a Greek god, one for a color, one a portmanteau, one an acronym. Individually fine; together, noise.

A family of names does work no individual name can. It tells the user these things belong together; it signals where one thing sits relative to the others; and — the test that matters most — it makes the next name obvious. If you can't name the sixth member without a committee, you never had a system.

The quality bar for a name family is not “are these five names good?” It's “does the sixth name write itself?”

02The distinctiveness spectrum

Before the family question, the per-name question. Brand-name theory (and, not coincidentally, trademark law) sorts names along a spectrum of distinctiveness. From weakest to strongest:

Two things to take from this. First, distinctiveness and protectability rise together — the law rewards names that don't describe, because a descriptive name is a word your competitors also need. Second, and more usefully here: most great name families live in the suggestive-to-arbitrary band. A metaphor-derived family (planets, rivers, ranks) is by construction arbitrary with respect to the product — “Saturn” has nothing to do with a storage tier — yet suggestive with respect to the family, because once you know the metaphor, each member's position is legible. That double life is exactly what you want. (The generator's coin it toggle nudges a family one notch further, toward the fanciful end.)

There's a second vocabulary hiding inside every domain: type words versus named instances. The water cycle's type words are Vapor, Rain, River; its named instances are the Thames, the Danube, the Mekong. Mountains give you Ridge, Crag, Summit — or Everest, Denali, Eiger. Winds give you gust and gale — or the Mistral and the Sirocco. Reaching for the already-named pays off because someone has already done the work of making those words feel like names. But mind two things. First, on the §02 spectrum, a proper instance sits further toward arbitrary — more ownable, but also more loaded: a named river may already be a company, a mountain a trademark. Second, named instances usually form a flat peer set; they shed the internal ordering the type words carried (Source→Mouth is structure; Thames→Nile is not). So reach for named instances when the relation is kindred or the tiers differ by vibe; keep type words when position has to read. The generator's name kind toggle switches between the two, or blends them.

Not all named instances are equally usable. Some are so heavily branded already that they're effectively spoken for — Apollo, Mars, Olympus, Tiffany read less like an open metaphor and more like someone else's logo. The generator marks these with a °, and a naming for production switch hides them outright so the palette only shows instances you have a fighting chance of owning. The freshest names from this layer are usually the deep cuts — not Orion but Bellatrix; not Everest but Eiger; not the Mistral but the Harmattan. (And if you fall for a loaded one anyway, coin it can carry it just far enough off the trademark.)

03Sound is meaning before semantics

Names are heard before they're parsed. The bouba/kiki effect — first observed by Köhler in 1929, replicated endlessly since — is the finding that ~95% of people map the nonsense word bouba to a round blob and kiki to a spiky star, across languages and ages. Sound carries shape, size, and texture before it carries meaning.

For naming this is leverage, not trivia. Front vowels (ee, i) read small, light, fast, precise — Nike, Twitter. Back vowels (oh, oo) read large, heavy, solid — Volvo, Google. Plosives (k, t, p, hard g) feel sharp, technical, abrupt; sonorants and fricatives (l, m, n, r, s, v) feel smooth, soft, continuous. A name's job should agree with its sound: a security product wants edges (kiki); a sleep product wants curves (bouba).

This matters for families too. Once you've chosen a source domain, it usually contains both sharp and soft members — the water cycle gives you the soft Mist and the crisp Crest. You can pick within the domain to match the product's character. The generator's phonetic lens tilts a family toward round or toward sharp on exactly this axis.

04The modal metaphor

Here is the core. To name a set coherently, don't reason about the objects — reason about the relations between them, then borrow a domain that already has those relations.

This rests on two well-worn ideas from cognitive science. Lakoff & Johnson's conceptual metaphor theory says we routinely understand one domain in terms of another (TIME IS MOTION, MORE IS UP); metaphor is a habit of thought, not a decoration on language. Dedre Gentner's structure-mapping theory sharpens the mechanism: a good analogy maps not surface features but the system of relations from a base domain onto a target. The planets aren't like your storage tiers in any property; they're like them in structure — an ordered series of distinct, nameable bodies along one axis.

The word modal does the selecting. Ask: in what mode are these things related?

• Related in space? — reach for places, cartography, architecture.
• Related in time, as a cycle? — reach for the water cycle, the seasons, the hours, the moon's phases.
• Related in time, as a sequence? — reach for a river's course, a plant's growth, a forging process.
• Related by rank? — reach for nobility, military grades, musical tempo.
• Related as parts of one whole? — reach for anatomy, a ship, a loom, a book.
• Related as opposites or poles? — reach for day/night, the winds, light and shadow.
• Just a flat peer set? — reach for trees, birds, gems, constellations.
• Spread across a magnitude axis? — reach for sizes, measures, the cosmic scale.
• Each one nested inside the next? — reach for map zoom, administrative tiers, shells and cores.
• A hub and its satellites? — reach for the wheel, the orrery, hub-and-spoke.
• A forking tree, or a causal chain, or a lineage? — reach for a tree's form, ignition, a dynasty.

The mode is the question; the metaphor is the answer; the family is automatic. Two seed intuitions make it concrete: objects related in space → placenames, objects related in time → the water cycle. Generalize the move and you have a method — and a generator.

05Why families beat one-offs — the evidence

The technology industry has been running this experiment for decades, and the durable schemes are all modal metaphors:

• Apple ran big cats (Cheetah → Jaguar → Leopard → Lion), a kindred set, then switched to California places (Mavericks, Yosemite, Big Sur) — a spatial set, because they'd run out of impressive cats. Note the tell: the scheme can be exhausted, which means it was a real system.
• Android shipped desserts in alphabetical order (Cupcake, Donut, … Pie) — a sequential spine grafted onto a kindred theme, so the order was load-bearing.
• Ubuntu uses Adjective Animal, both starting with the same letter, incrementing alphabetically (Feisty Fawn, Gutsy Gibbon) — a generative template, not a fixed list.
• Debian names releases after Toy Story characters — a closed cast, deliberately whimsical, deliberately bounded.

The pattern across all four: a single source domain, a clear rule for membership, and an obvious next member. They differ in whether the family is ordered (Android, Ubuntu) or merely kindred (Debian, early Apple) — which is precisely the distinction the modal question resolves for you up front.

06The method, in seven moves

1. Inventory the elements. List the things needing names. Five tiers? Three agents? Seven modules?
2. Name the relation, not the elements. How do they relate to each other? Pick the mode — spatial, cyclical, sequential, hierarchical, part–whole, oppositional, or simply kindred (a flat set with no order).
3. Choose an isomorphic source domain. One whose own members carry that same structure. The mode shortlists it; taste (and the register control) picks the winner. Prefer a domain with more members than you need — headroom is extensibility.
4. Harvest the members in register. Pull the words. If the relation is ordered, keep them in their native order so position encodes rank/stage.
5. Run the phonetic pass. Tilt the selection toward the product's character — round for soft/calm, sharp for technical/fast (§03). Drop members that mispronounce or collide.
6. Gut-check availability. A real word from a real domain is, by the spectrum (§02), arbitrary for your product — good for protection — but check it isn't already owned in your category, and that the .com/handle/namespace isn't a fight you'll lose. (When everything's taken, coin it.)
7. Apply the extensibility test. Can a stranger, told only the metaphor, name the next member correctly? If yes, ship it. If no, the domain is too sparse or too arbitrary — go back to move 3.

The generator at the top runs moves 2–5 on a library of ~60 source domains, cross-indexed by fourteen relation modes and nine registers, many of them thickened with named instances (the Thames, Everest, the Mistral) alongside their type words. Moves 1, 6, and 7 are yours.

07Pitfalls

• Choosing a metaphor that's prettier than it is structural. “Constellations” sounds lovely, but if your objects are an ordered pipeline and constellations have no order, the metaphor fights the product. Structure first, charm second.
• Picking an exhaustible domain for an open-ended set. Seven planets is fine for seven things and a trap for a platform that will grow to twenty. Match the domain's cardinality to your roadmap.
• Over-cleverness that needs a footnote. If users must be told the metaphor for the names to make sense, you've bought obscurity. The best families reward the insight but don't require it.
• Forgetting the family is a promise. Once you're “the rivers company,” every future name must be a river. Choose a domain you'll still want to live in at member number fifteen.

08Sources

• Gentner, D. — Metaphor as Structure Mapping: The Relational Shift; Structure-Mapping: A Theoretical Framework for Analogy.
• Lakoff, G. — The Contemporary Theory of Metaphor (conceptual metaphor theory).
• Pathak, A. et al. — Assessing sound symbolism (bouba/kiki paradigm); Sounds good: Phonetic sound patterns in top brand names.
• Brand-name taxonomy — Every type of brand name in one chart; The Five Types of Brand Names (the distinctiveness spectrum).
• Naming systems in practice — Ubuntu version history; The developer obsession with code names.