Dense and value-dense with no concept-overexplaining (no 'what is a seed' preamble); every section — coverage matrix, format table, executable script, validation, anti-patterns — earns its tokens, matching the lean score-3 anchor.

Provides copy-paste-ready executable code (FactoryBot seed.rb, rake/grep shell commands) with concrete file paths; the deferred Python equivalent is explicitly justified flexibility rather than pseudocode.

Five explicitly numbered, sequenced steps with a validation checkpoint ('assert the row count') and a fix-and-retry loop ('If the count is off, fix... and re-run'), matching the score-3 anchor for clear sequence with feedback loops.

Core workflow lives inline while verbose CI/ephemeral-env orchestration is split into a real, clearly-signaled one-level-deep reference (references/seed-scripts.md, verified to exist) with no nested indirection.

Lists multiple concrete actions — 'picks a representative... cross-section', 'generates the rows via... factory library', 'persists the dataset as a checked-in fixture', and 'wires it into the test bootstrap' — matching the score-3 anchor for multiple specific concrete actions.

Explicitly answers both what (builds/persists/wires a reproducible seed dataset) and when via the 'Use when starting E2E coverage... or when an existing seed has drifted' clause.

Uses natural domain phrasing a testing engineer would actually say ('no seed strategy', 'seed has drifted', 'E2E seed dataset', 'test bootstrap') with good coverage of variations rather than abstract jargon.

Carves a clear niche (reproducible E2E seed datasets via specific factory libraries) with distinct triggers unlikely to fire for general test or data-generation skills.