Glossary
Every term you'll meet in the app, the API, and this knowledge base, in one place.
Accuracy hold: what stops a written post from publishing until a person looks. The kernel checks claims as it writes and holds a draft that lost a table, image, code block or quote from its input; on a company's own blog, a post where most figures have no source is held too. See Held draft.
Arena: an adversarial cluster of research agents. One run turns a topic into a sourced, fact-checked outline with its sources, its direction and its stat ledger. What the product used to call the research agent. KB-206.
Auto-Pen: the one-click writer: Research → Write → the guaranteed passes on a post, with the post locked until it finishes. KB-301.
Backfilled mention: your brand mention added to a post after the fact, once that post is already being cited by AI engines but doesn't name you yet. A deliberate follow-up step, not a mistake being corrected. See KB-213.
Balance: how many direct runs of a primitive you have left. Separate from the AI usage allotment, and per primitive. KB-407.
Brand mention: how much a post names your company: off, subtle, or woven. On phantoms, a mention rate sets what share of posts mention you at all.
Brief: what a kernel writes from: a headline, notes, an outline, or an existing article. KB-703.
Canvas: a visual style: cover-image styles and diagram styles, made on the Canvas page and referenced by collections.
Capability: a permission carried by an API key. KB-801.
Collection: a publication. A standing configuration that produces and publishes posts on a cadence, with its own topic map, voice and approval posture. A phantom is a collection whose site we host; your own blog is a collection publishing to your CMS. Both run on the same engine.
Constraints: what a run was configured with, recorded on the run itself. The answer to "what did this actually run with", in one read.
Control Room: a collection's home page in the app: headlines, topics, setup.
Direct / internal: whether a run was asked for by you (direct, and billed to your balance) or by another part of Letterstory acting on your standing instruction (internal, recorded but billed to that product).
Direction check-in: an optional pause after Research where you pick the article's angle from a short list.
Endpoint: a named, callable configuration for a primitive: a kernel endpoint, a shredder endpoint, an arena, a seer, a collection. It has a public id, a one-file configuration, and a run history.
Enhance pass: one of the guaranteed passes that run after the writer: brand mention, visuals, internal links, external sources, metadata. Each is a switch on the collection's Auto-Pen card, and each can be run on one post (run_enhance_pass). A pass adds; it never rewrites your prose. KB-303.
Held draft: a written post that won't publish until a person releases it, because an accuracy check stopped it. KB-306.
Identity: how AI should describe you, and what you want to be the answer for. Part of Strategy.
Kernel (writing kernel): the voice a draft is written in: a pipeline with house rules and a reviewer panel. Eight built-in voices exist, and an organization can fork and tune its own under Primitives ▾ → Kernels. KB-701.
Landing page: a URL plus context that a collection can weave into posts automatically.
Lede: a cold-email opening line chosen for one prospect, from a curated library, with the company behind them enriched first. A newer one, with the same endpoint-and-run shape as the rest.
Masthead: a phantom site's name. Never contains the client's name.
MCP: Model Context Protocol; how an AI agent discovers and calls Letterstory's tools. KB-402.
Paper trail: the public sources published with an article, listed in the Publish panel and curatable before you publish. KB-211.
Phantom (phantom site): a stand-alone publication Letterstory hosts, independent of your brand. KB-601.
Pillar / spoke: the two levels of a topic map: a pillar is ground a searcher owns ("Destination guides"); a spoke is one future post under it.
Planner: autopilot for a collection: picks spokes, writes them, publishes on a cadence. Its runs are ordinary collection runs, stamped as the cadence's rather than yours.
Positioning: your company description, priority topics, stances, and avoid-topics. The heart of Strategy.
Primitive: one of the engine's building blocks, each callable on its own: kernel, shredder, enhance, seer, arena, collection, lede (opener). KB-407.
Rebuild: regenerate a phantom site from its published posts so a change shows. Refresh content in the app.
Research: the post editor's name for a hosted arena run, one that writes its outline into that post.
Run: one execution of a primitive. Every run has an id, a status you poll, the constraints it ran with, and a verification read-back.
Seer: a watcher on an outside signal (GitHub, Notion, a spec URL, a news or regulation query) that produces headline suggestions or drafts.
Shred: rewrite each sentence of a draft across several model providers so no single model's signature remains. Also the unit a shredder endpoint spends: one shred per article. KB-217.
Shredder endpoint: a callable URL for shredding, carrying standing defaults. KB-217.
Skeleton: the shape of an article sent to a kernel as data: which sections are lists, steps, prose, or fixed text. Templates produce one.
Stance: a position you hold and argue for. Shapes every piece.
Strategy: the app page labelled Context in the header (and the settings behind it) for positioning, identity, competitors, cadence, and context.
Template: an ordered list of blocks saying what shape a post takes. Binds to a collection, a post, or a seer.
Tell: a mechanical sign of machine writing (colon headings, em-dash overuse, a missing brand mention) that the platform detects and, in a tell pass, proposes fixes for.
Theme: a phantom site's visual look. KB-804.
Topic map: a collection's pillars and spokes; what it plans to cover.
Verification: the checks a finished run reads back against its own constraints, each pass, fail or skipped. Reported, never hidden, and never blocking.