Known test gaps

Tests that exist but are skipped, and why. Each is skipped rather than
asserted-to-pass so it cannot mask a real defect.

Unfinished agent-written tests (2026-09-06)

Eleven tests were written during a coverage push whose authoring agents were cut
off mid-file by a rate limit. Their expectations were never reconciled against
actual behaviour, so it is unknown whether each failure is a wrong expectation or
a real bug. They are marked it.skip with an UNFINISHED: comment at the call
site.

File	Skipped	What to check
`tests/dialects/duckdb-builders.test.ts`	3	`[model, field]` order pairs; `dropPartition()` child-table + metadata cleanup; whether RLS/security policies throw
`tests/dialects/postgres-sql-builders.test.ts`	1	`createSequence` schema-qualification and option ordering
`tests/cli-io-prompt-keychain.test.ts`	7	keychain backend selection (`security` vs `secret-tool`) — likely environment-dependent `spawnSync` stubbing rather than a product bug

To resolve one: run it, decide whether the expectation or the code is wrong,
then either fix the expectation or file the bug and keep it skipped with a
// BUG: note. Do not simply rewrite the assertion to match current output —
that is how a silent defect gets frozen in.

One test from the same batch was resolved: trino-builders.test.ts expected
CREATE TABLE … "day" DATE from a column declared { type: 'DATEONLY' }. The
product is correct — a raw string type is passed through verbatim by design,
and DATEONLY is a DataTypes key, not a SQL type. The test now uses
DataTypes.DATEONLY() and asserts DATE, which is what actually exercises the
type mapper.

Integration suites (pre-existing)

13 suites / 510 tests are gated on INTEGRATION=1 and need live services
(PostgreSQL, MySQL, MariaDB, SQL Server, CockroachDB, ClickHouse, Oracle,
Elasticsearch, Qdrant, etcd, InfluxDB, Neo4j, Weaviate, ArangoDB). Start them
with npm run test:db:up, then npm run test:integration. They skip by design
when docker is not running.

Where the tests are

Counted with ls tests/*.test.ts | wc -l for the root and
find tests/<dir> -name '*.test.ts' | wc -l for each directory; 648 test files
in total (find tests -name '*.test.ts' | wc -l). "other" is compliance (2),
associations, cloudflare, core and extensions (1 each).

The chart shows that the root directory (cross-cutting suites: models,
transactions, type mapping, CLI, …) and the per-dialect suites hold most of
the files; a dialect or store without a file under tests/dialects or
tests/nosql is one to check before relying on it.

What changed since 4.8.2

The three commits that make up the next release (

git log --oneline
a3ce5fba..477f13fb

: type mapping and the transaction contract, 42 new
dialects, prorm models pull), grouped by top-level src/ directory with
git diff --numstat a3ce5fba..477f13fb summed per directory. A chart would be
dominated by the dialect row, so this is a table.

Area	Files	Lines added	Lines removed
`src/dialects`	90	64,178	8,793
`src/codegen` (new)	20	6,472	0
`src/query-builders`	18	5,982	2,045
`src/*.ts` (root modules: `prorm.ts`, `transaction-support.ts`, `query-pipe.ts`, …)	22	5,419	1,480
`src/models`	7	1,406	559
`src/decorators`	14	1,025	240
`src/compliance`	5	995	153
`src/cli`	10	860	104
`src/nosql`	3	709	10
`src/query-optimizers`	5	536	98
`src/types`	1	410	92
`src/migrations`	3	114	16
everything else under `src/`	12	146	91
`tests/`	239	59,213	1,641
`docs/`	49	7,587	16

Going further — the rest of the reference material

Architecture — where each area sits

Unfinished agent-written tests (2026-09-06)#

Integration suites (pre-existing)#

Where the tests are#

What changed since 4.8.2#

Related reading#

Unfinished agent-written tests (2026-09-06)

Integration suites (pre-existing)

Where the tests are

What changed since 4.8.2

Related reading