coin

coin — name a thing on purpose

Replicates Lexicon Branding's process (David Placek: PowerBook, BlackBerry, Sonos, Azure, Vercel, Windsurf). You are coining, not brainstorming: generate a wide funnel of directions with judgment suspended, then score against a fixed rubric. A name is an experience the market will have, not a word the team likes. Comfort is a warning sign; polarization is a sign of strength.

Arguments

• Free text or a path to a brief file → new run (brief file skips the interview).
• Path to an existing naming-runs/<run>/ workspace → jump to Round 2.
• --depth quick|default|deep (default default) → funnel widths below.
• --runs-dir <path> → parent dir for the run workspace (default ./naming-runs/).

depth	per-lane directions	triage keep	score keep	shortlist
quick	100	30	10	4–5
default	250	60	20	5–8
deep	250 ×2 lanes-per-role	100	30	8–10

Phase 1 — Identify

1. Interview (skip if a brief file was given). One question at a time:
   • The diamond: how do you define winning? What do you have to win (assets, traction)? What do you need to win? What do you have to say to win?
   • Behavior and experience: how does the product behave now, and how should it behave in the future? How should the market behave toward it? Which 3–6 experience qualities should the name evoke (e.g. flow, precision, warmth)?
   • Constraints: markets (default: global + Japan), name type (company / product / feature), must it work as a CLI command or package name?
2. Landscape: spawn one research subagent to map competitor names and the category's stock language (the words everyone uses — to avoid; descriptive names imitate, and imitation forfeits distinctiveness).
3. Framework: synthesize framework.md — target experience qualities, behaviors, what the name must be able to say (now and later), category language to avoid, and sound priorities chosen from the letter table in references/linguistics.md.
4. Disguise briefs: draft the two blind briefs (see lane roles below) — a fictional company in an adjacent category, and a physical object with the same experience qualities. Neither may mention the real product, category, or company.
5. Sign-off gate: present the framework and both disguise briefs to the user and stop. Phase 1 is complete only when the user has approved them — no invent tokens are spent before that. On approval, create the run workspace <runs-dir>/YYYY-MM-DD-<slug>/ and write framework.md, landscape.md, briefs.md.

Phase 2 — Invent + screen (one Workflow run)

Load references/workflow.md and run its script verbatim via the Workflow tool, passing the framework, briefs, funnel widths, and the absolute path of references/linguistics.md through args. The script is frozen so every run takes the same shape: four blind lanes → triage → linguistic scoring → screening. Do not paraphrase the lane prompts — the disguise lanes work because the script, not you, holds the briefs.

Lane roles (defined in the script):

1. truth — full real brief.
2. disguise — the fictional-adjacent-company brief.
3. object — the physical-object brief.
4. mining — no names: raw words, metaphors, sports, natural phenomena, morphemes around the experience qualities; a synthesis step compounds them into candidates ("Windsurf was sitting on a word list").

Screening semantics (the script's screeners follow these; hold them when interpreting results):

• A taken .com is a price flag, never a kill — domains are area codes now; record and move on.
• Trademark results are advisory — not legal clearance; the presentation must carry that label verbatim.
• Cultural panel: English, Japanese, Spanish, Portuguese, German, Mandarin. Eliminate the terrible, flag the worrisome.

Phase 2 is complete when the workflow has returned and you have written funnel.md (every cut logged with what was dropped and why — no silent truncation), directions/ (raw lane output), scores.md, and screening.md into the workspace. Read the workflow journal if a result looks empty; never assume.

Phase 3 — Implement

Produce presentation.html in the workspace: a single self-contained page (no external requests) presenting the shortlist. For every finalist:

• The story — rationale written as Lexicon writes it for a client: what the name lets the company say, and how the market will behave toward it. Start a story, don't make a statement.
• Sound analysis — how its letters and rhythm serve the framework's target qualities, grounded in references/linguistics.md.
• Screening table — domains, registries, collisions, trademark advisory, cultural flags.
• Prototypes — the name rendered in developer contexts: a terminal (pip install <name> / npx <name>), a GitHub repo header, a landing-page hero, a launch tweet. Seeing the name executed is what flips decisions.

End the page with the user's drills, addressed to them:

• Speculate, don't evaluate: ask "what could this name do for us?", never "do I like it?".
• Polarization test: if the team splits hard over one name, that's energy — Grove chose Pentium over ProChip because of the argument.
• Competitor drill: tell a friend "we just got a new competitor called X — what do you think?". You're not collecting opinions; you're watching what the name does to them. The winning reaction: "I don't know what they do, but they're not like the other guys."
• If the team is comfortable with a name, chances are it isn't the name yet.

Phase 3 is complete when every finalist has all four blocks and the page opens clean from the file system. Close by offering Round 2 and, in one line, suggest recording the eventual decision wherever the project keeps decisions.

Round 2 — resumable, offered never automatic

Given an existing workspace:

1. Read framework.md, funnel.md, scores.md, screening.md, plus any prior round-2/ state.
2. Collect the user's reactions to the shortlist and derive framework deltas ("more tactile, less abstract; compounds outperformed coined words") into round-2/deltas.md.
3. Re-run the Workflow script with args.round2: two lanes seeded by the deltas instead of four, plus rescue — near-misses from the persisted funnel that the deltas now favor re-enter scoring without regeneration.
4. Co-creation: any names the user proposes enter the same scoring and screening stages as generated ones. Same rubric, no home-team bias.
5. Produce round-2/presentation.html showing round-2 finalists side-by-side with round-1 survivors — humans decide by comparing.