Symmetry tests never run either language's schema validator over the other's documents #155

Joint issue — the Python half is VH-Lab/DID-python#28. The gap is symmetric and the fix has to land on both sides to be meaningful.

The problem

The cross-language symmetry suite verifies that each language can read and round-trip the other's database. It never verifies that either language considers the other's documents schema-valid.

Validation lives in database.add_docs in both languages ('Validate', true here, validate=True in Python — the default on both). Each language calls add_docs only on documents it created itself, during makeArtifacts. The readArtifacts step on both sides opens the other language's SQLite file, re-summarizes it, and compares summaries — it never calls add_docs, so validate_docs / validate_doc_vs_schema never runs.

MATLAB (tests_symmetry/+did/+symmetry/+readArtifacts/+database/buildDatabase.m):

db = did.implementations.sqlitedb(dbPath);
liveSummary = did.util.databaseSummary(db);
report = did.util.compareDatabaseSummary(savedSummary, liveSummary);

Python (tests/symmetry/read_artifacts/database/test_build_database.py):

db = SQLiteDB(db_path)
live_summary = database_summary(db)
report = compare_database_summary(saved_summary, live_summary)
...
doc = db.get_docs(expected_id, OnMissing="ignore")   # get_docs does not validate

So a document that one language writes can be invalid under the other's schema validator while the suite stays green.

What the suite does and does not catch

Covered by compareDatabaseSummary: branch names, branch hierarchy, per-branch document count, document IDs, class_name, the .value field of demoA/demoB/demoC, and depends_on name/value pairs.

Not covered — anything validate_doc_vs_schema would check:

• field types and ranges outside those three .value fields
• did_uid / timestamp / char-length / matrix-shape conformance
• required (mustbenotempty) depends_on entries being present and non-empty
• inherited superclass property lists being present and well-formed
• document_class.validation pointing at a resolvable schema
• files / file_info conformance via checkfiles

Two concrete instances

1. Python document IDs were UUID4 (now fixed, but the suite never noticed). base.schema.json types base.id as did_uid — 16 hex digits, an underscore, 16 more, exactly what did.ido.unique_id produces and did.ido.isvalid accepts. Python's IDO.unique_id() returned a UUID4 instead, so every Python-generated document was schema-invalid and this repo's add_docs would have rejected all of them. The symmetry suite was green throughout, because MATLAB only ever read them, never validated them. Fixed in VH-Lab/DID-python#27; the detection gap that let it persist is not.

2. base.datestamp format — believed still live. Python writes it with a space separator:

Python:  '2026-08-28 11:14:42.595475'      str() of a naive datetime
MATLAB:  '2026-08-28T10:11:16.934Z'        char(datetime(...,'TimeZone','UTCLeapSeconds'))

base.schema.json types datestamp as timestamp, and did.database.validate_field_type_and_value checks it with:

value = regexprep(value,'Z$','');
java.time.LocalDateTime.parse(java.lang.String(value));

LocalDateTime.parse uses ISO_LOCAL_DATE_TIME, which requires the T separator. If that reading is right, this validator would raise DID:Database:ValidationFieldTimeStamp on every Python-produced document — and the symmetry suite would still be green. Python's own validator accepts both forms (it has strptime fallbacks), so the Python side is quiet too.

This one is reasoning from source, not an observed failure. It was written without a MATLAB runtime, so it needs confirming here before anyone acts on it — the quickest check is to run validate_doc_vs_schema over a document from a Python-produced artifact and see whether base.datestamp trips. Note it is pre-existing and not introduced by the recent Python work.

Proposed fix

Add a validation pass to readArtifacts on both sides, so each language runs its own validator over documents the other language wrote:

• MATLAB — in tests_symmetry/+did/+symmetry/+readArtifacts/+database/buildDatabase.m, after opening the database, read the documents and call db.validate_docs(...), failing the test on a validation error.
• Python — the mirror in test_build_database.py, calling db.validate_docs(docs) (public since Port document-vs-schema validation from MATLAB to Python DID-python#27).

That turns the suite into a real cross-language contract check rather than a round-trip check, and would have caught instance 1 the day it was introduced.

Per AGENTS.md, anything implemented here needs a MATLAB environment to run and validate.