# Source coverage

Build a full-source ledger before drafting. The ledger accounts for every transcript segment exactly once as reviewed or as an explicitly unexplored region with a concrete reason. Do not stop at the opening, the cleanest quote, or the first plausible article.

For each source, inventory these evidence classes from exact segment IDs:

- money;
- percentages;
- numbers;
- names;
- case studies and results;
- mechanisms and processes;
- claims;
- quotations.

The submitted available-segment inventory must match the host’s inventory. Used IDs must belong to that class and to the reviewed region. Meet the configured coverage floor for every available class; never satisfy it with IDs from another call.

For every article-worthy opportunity, record one internal `articleDecision` with a defensible claim, supporting segment IDs, audience, pillar, offer route, acquisition intent, destination, archetype, one primary angle, and at least two alternative angles. A drafted decision maps one-to-one to a full article. A discarded decision has a concrete reason. These decisions are provenance, not Atomic ideas and not a human approval step.

Quality decides the article count. Do not pad a batch. A fully covered source may end as `no_viable_articles`; state why each considered opportunity was discarded. Never turn a possibility into a result, a third-person account into first-person experience, or a speaker’s opinion into a verified fact.