db-naming (M7)

Naming is the schema's documentation: consistent, predictable identifiers make joins obvious and reduce the quoting/casing bugs that creep in across ORMs and engines. It is low-weight (Naming category, design axis) — real but never the headline. This module is design-axis and never caps.

What it checks

• Casing / pluralization consistency: a mix of snake_case and camelCase, or singular and plural table names (user, orders, OrderItem) in one schema — pick one and hold it.
• Reserved words & case-folding traps: identifiers that are SQL reserved words (user, order, group, select) or that rely on case ("User" quoted) — Postgres folds unquoted to lowercase, MySQL is filesystem/case-config dependent; mixed quoting causes "relation does not exist" bugs.
• FK naming: FK columns not following a predictable <referenced>_id pattern, so joins aren't self-evident.
• Boolean naming: booleans not prefixed is_/has_/can_, or negative names (not_active) that invert logic.
• Opaque / ambiguous names: data, info, value, flag, temp, col1, abbreviations without a glossary, or timestamps not suffixed _at.

Axis & severity

• Axis: design; magnitude almost always low, banded honestly.
• Reserved-word unquoted identifier that breaks across engines: severity 2–3, warn.
• Schema-wide casing/pluralization inconsistency: severity 2, warn, fixable: proposed (rename is high-blast-radius — never auto).
• Opaque column name / missing _at suffix: severity 1, warn.
• M7 never caps any score; it is the smallest-weight design category.

Tier-0 static check

Parse DDL/snapshot via scripts/parse-schema.mjs: tally identifier casing styles and table pluralization; match identifiers against a reserved-word list per engine; check FK columns against the <table>_id pattern; flag boolean columns without is_/has_/can_ and opaque names. This is fully static and directional.

Tier-1 verification query

Confirm live identifier inventory and quoting:

-- $DATABASE_URL, read-only
SELECT table_name, column_name FROM information_schema.columns
WHERE table_schema NOT IN ('pg_catalog','information_schema')
ORDER BY table_name;   -- inspect casing/plurality/reserved words across the real catalog

Findings

Emit per schema/finding.schema.json. Examples:

• M7.schema.mixed_casing — schema mixes snake_case and camelCase identifiers (severity 2, warn, axis design, fixable: proposed).
• M7.user.reserved_word_table — table named user requires quoting and breaks unquoted refs (severity 2, warn, axis design).
• M7.accounts.active_boolean_unprefixed — boolean active not named is_active (severity 1, warn). Each finding: evidence.observed quotes the identifier(s) verbatim; verification.reproduce is the runnable query above (method: ddl_parse / schema_introspect); expected_impact is banded (typically low) + confidence directional with rationale.

Honesty

• Naming findings are advisory polish — never let them dominate the report or imply a functional defect.
• Renames are destructive and break app code/migrations; recommendations are proposed/advisory, never auto.
• Reserved-word and case-folding behaviour is engine-specific; scope each finding to the detected engine rather than asserting a universal rule.