Baseline: 51.0% total statements (2026-10-03, after PRs #41-#43). Scope: pgsql, sqlexec, template, plus non-reader/writer packages. Other readers/writers deferred.

Order

#	Plan	Package(s)	Now
1	pgsql.md	readers/pgsql, writers/pgsql, pkg/pgsql	16.0 / 74.0 / 87.8
2	sqlexec.md	writers/sqlexec	19.4
3	template.md	writers/template	8.5
4	models.md	pkg/models	20.4
5	cmd.md	cmd/relspec, pkg/jobs	49.3 / 72.0
6	ui.md	pkg/ui	3.8
7	diff-merge.md	pkg/diff, pkg/merge	65.5 / 75.1
8	sqltypes.md	pkg/sqltypes	67.0

Conventions

Same package as code under test; table-driven; must pass -race.
Existing data first: tests/assets/*, examples/*.dbml, tests/postgres/init.sql, tests/postgres/issue21. Generate new data only where listed under "Data needed".
New fixtures go in tests/assets/<format>/ or package testdata/.
Live-DB tests: skip unless the env var is set (pattern in pkg/readers/pgsql/reader_test.go). Use tests/dbtest/dbtest.sh (podman/docker; see tests/dbtest/README.md):
- dbtest.sh up|down <postgres|mssql|mysql|all>
- eval "$(dbtest.sh env postgres)" sets RELSPEC_TEST_PG_CONN (mssql: RELSPEC_TEST_MSSQL_CONN, mysql: RELSPEC_TEST_MYSQL_CONN)
- dbtest.sh test <db> [pkgs] runs up, go test, down
Fixtures for live DBs: postgres tests/postgres/init.sql, mssql test_data/mssql/test_schema.sql, mysql tests/dbtest/init/mysql.sql.
Prerequisite: working container networking (currently blocked until reboot into matching kernel; tun module).
Prefer pure-function tests over DB tests wherever the logic can be isolated.
Output assertions must not depend on map order (see memory: map iteration determinism).

Targets

Package	Target
pgsql (all three)	>= 85
sqlexec	>= 80
template	>= 85
models	>= 80
cmd/relspec	>= 65
jobs	>= 85
ui	>= 40 (data ops/pure helpers; screens via smoke tests)
diff, merge	>= 85
sqltypes	>= 85

Verify

go test -race -coverprofile=c.out ./pkg/<pkg>/ then go tool cover -func=c.out
make test before commit