Status: Split out of the former writer disk-format validation table on 2026-05-23.
This note holds durable validation rules for writer-emitted Access disk-format bytes. The cross-feature coverage map lives in writer-disk-format-validation-matrix.md.
Local builds and targeted test runs are allowed for both ordinary regressions and DAO checks. Select the methods or classes relevant to the change; DAO tests need an Access-equipped host and may use locally built or matching CI-built binaries. Full test suites and all benchmarks run only on CI. Required hosted acceptance gates still apply; see AGENTS.md.
Selected .NET validation also requires a hosted DAO120 compatibility inventory
for each runtime/library combination. Disposable Windows hosts install the pinned
Microsoft Access Runtime and validate the actual loaded ACEDAO DLL version.
Office/MSACCESS is pinned to 16.0.20527.20112; its separately pinned ACEDAO
component reports 16.0.20527.20016. The component pin reflects three independent
installs of that exact release; product and component build numbers are distinct.
The lane consumes its matching analyzed runner artifact without rebuilding;
every required method must pass and no skip is accepted. Ordinary partitions
retain guarded DAO skips where Access is absent. DAO3.6/Jet3 native evidence and
feature variants outside the required inventory remain separate obligations.
Use the strongest feasible automated signal for the mutation under test:
DAO.DBEngine.120 against writer output and materializes recordsets or DAO index seeks.Manual Microsoft Access UI checks are supplemental evidence. They are useful for workflows that DAO does not expose, such as saving an attachment back to disk through the UI, but they are not stronger than DAO automation when the same scenario can be automated.
DAO-authored byte comparisons and FormatProbe output are diagnostic oracles for undocumented format details. When a probe identifies a root-cause invariant, promote that invariant into a focused regression test instead of leaving the probe report as the only evidence.
Access-authored fixtures checked into this repository, DAO-authored scratch databases, and Access/DAO-authored payload bytes are the trusted sources of truth for engine behavior and undocumented disk-format details. Use them as the oracle side of a comparison or as the host database for writer mutations.
Writer-created fresh databases are never fixture oracles. They are subjects under test: useful for validating AccessWriter.CreateDatabaseAsync, bootstrap catalog scaffolding, generated linked metadata, or other writer-owned output, but not a trusted baseline for unrelated DAO/Access compatibility questions.
If a test needs a smaller or cleaner trusted base than Northwind, create that base through DAO/Access or add a checked-in Access-authored fixture. Do not replace an Access-authored compact/open-recordset fixture with a fresh database produced by this library just to make the test faster.
Prefer Access-authored fixtures such as NorthwindTraders.accdb for DAO compact tests that validate writer mutations. That keeps the trusted base in Microsoft Access bytes and isolates the writer-owned changes.
DAO-created scratch databases are acceptable trusted hosts when the scenario only needs a small engine-authored container, for example direct DAO OpenRecordset or index-seek probes over writer-added user tables.
Writer-created fresh databases are still valuable for bootstrap coverage, but they make catalog scaffolding part of the subject under test. Use them only when the feature being validated is the bootstrap or generated writer output itself.
CompactDatabase exit code 0 is not enough. Always reopen the compacted output and assert the feature that motivated the test: rows, indexes, relationships, complex payloads, encryption/password handling, or page-reuse invariants.
Encrypted source compaction must use DAO’s five-argument form, with the source password supplied separately from the destination locale/password. The historical details live in round-trip-openrecordset-hypothesis.md.
If a design note says Microsoft Access Compact and Repair validation is pending, replace the blanket warning with a link to the relevant matrix row in writer-disk-format-validation-matrix.md or to the feature doc that owns the residual byte-level gap.
Access-equipped hosts currently cover DAO OpenRecordset row counts, DAO primary-key seeks, AutoNumber continuation, MEMO/OLE/LVAL fidelity, FK enforcement, FK CompactDatabase, encrypted CompactDatabase, fresh ACCDB bootstrap compact, complex-column compact, storage-maintenance compact, advanced ACE/Jet4 index compact, linked Access-file Type 6 CompactDatabase plus compacted OpenRecordset, linked text/CSV Type 6 CompactDatabase plus compacted OpenRecordset, cached-schema ODBC Type 4 CompactDatabase, and conditional Jet3 index compact when the installed DAO engine can open Access 97 .mdb files. Coverage rows that mention writer-created fresh databases are bootstrap or writer-output coverage rows, not fixture-oracle rows.
Reader/regression coverage also guards writer-created linked Access/ODBC/text catalog rows: catalog-only negative object-id allocation, low-24 collision avoidance, Type 4/6 MSysObjects catalog-index splicing, fixture-aligned flags, linked-object ACE rows, and cleanup when a linked catalog splice fails. Access-file and text Type 6 links now have DAO coverage with null Lv/LvProp/LvModule/LvExtra cache columns, DAO-shaped Database LVAL storage, CompactDatabase, and compacted OpenRecordset. ODBC Type 4 links have CompactDatabase coverage when created with a caller-supplied Access/DAO cached-schema LvProp; generated ODBC links now use parseable table-level LvProp, and the source-column overload adds generated schema-cache targets.
Use writer-disk-format-validation-matrix.md as the active coverage inventory for validation promotions and residual byte-comparison gaps.