# Atomic Article contract

An Atomic Article is one complete 500–800 word argument grounded in one primary call source. It is ready for human judgment as an article, not as an outline, list of ideas, or promise to write later.

Choose one archetype that fits the source:

- `belief_shift`: replace a costly assumption with a source-backed way to see the problem;
- `how_to`: teach a mechanism or sequence the source actually supports;
- `mistakes_and_lessons`: show a concrete mistake, consequence, and learned correction.

Each article must include:

- a precise title and unique stable slug;
- the exact source ID, version, hash, and internal decision ID;
- a coherent opening, developed argument, and earned close without generic Introduction or Conclusion headings;
- explicit claim records with exact supporting segment IDs;
- a complete claim ledger where every published paragraph appears verbatim in at least one claim after whitespace normalization;
- audience, pillar, offer route, acquisition intent, and destination identical to the admitted decision;
- the exact active Voice revision ID and SHA-256 plus the Voice rules applied;
- first-person language only when an explicitly marked first-person claim cites first-person source evidence.

Keep one primary claim per article. Use source language where it is distinctive, but do not manufacture quotations. Attribute other speakers when ownership matters. Narrow or cut any claim the transcript cannot defend.

The batch candidate must bind to the current batch/version, workflow version, skill digest, context receipt and manifest hash, and every per-source release receipt. Producer-supplied check booleans are hints only; the host ignores them as proof.