test: finish adapter validation (importer coverage + export conformance)

docs/architecture/integration-adapters.md

@@ -60,9 +60,16 @@ was built first as the reference and the rest mirror it.
 - **Shared helpers removed duplication.** `metaseed._http.request_json`
   (retry/backoff) and `metaseed.isatab.to_isatab` are written once and reused by
   all the relevant adapters.
-- **Schema-valid output where it counts.** The ENA exporter's study/sample/run
-  XML **validate against ENA's official SRA XSDs** — i.e. the output is not just
-  well-formed, it would be accepted.
+- **Every exporter has a conformance test.** The ENA export **validates against
+  ENA's official SRA XSDs** (external schema — the gold standard); the PRIDE,
+  MetaboLights, and BrAPI exports have structural/shape conformance tests
+  (required px `MTD`/`FMH`/`FME` lines, the ISA-Tab sections + study/assay/MAF
+  files, and the BrAPI v2 object shape via jsonschema). All are `@network`
+  round-trips against a real record.
+- **Importers tested across diverse accessions.** Each importer was run against
+  several real, varied accessions (ENA run/sample/study incl. a 495-file study;
+  four PXD projects incl. a 2384-file one; four MTBLS studies) without surfacing
+  new mapping gaps.
 
 ## What didn't work (and the lessons)
 
@@ -90,9 +97,9 @@ was built first as the reference and the rest mirror it.
 
 ## Honest status
 
-The eight adapters are a real first step, **not production-validated.** Each was
-smoke-tested against a single live record (which already caught real bugs), and
-the clients now page and retry. Still open: broader accession coverage, export
-validation for BrAPI/PRIDE/MetaboLights (ENA is XSD-valid), and verification that
-exports are accepted by the live submission systems. Tracking: umbrella #75,
-importers epic #76.
+The eight adapters now page and retry, each importer has been exercised across
+several diverse live accessions, and every exporter has a conformance test
+(ENA against the official XSD; the rest structural/shape). The one remaining
+unknown is **acceptance by the live submission systems** — no export has yet
+been round-tripped through a real ENA/PRIDE/MetaboLights/BrAPI submission
+endpoint. Tracking: umbrella #75, importers epic #76.

pyproject.toml

@@ -41,6 +41,7 @@ dev = [
     "httpx>=0.27",
     "jupyter>=1.0",
     "xmlschema>=3.0",
+    "jsonschema>=4.0",
 ]
 docs = [
     "mkdocs>=1.6,<2.0",

tests/test_brapi/test_conformance.py

@@ -0,0 +1,42 @@
+"""Conformance of the BrAPI export to the BrAPI v2 object structure.
+
+Uses jsonschema to assert the v2 shape: every object carries its DbId, and an
+observation unit nests its level inside observationUnitPosition (never a
+top-level ``observationLevel`` — the bug fixed in this adapter). Opt-in (network).
+"""
+
+from __future__ import annotations
+
+import pytest
+from jsonschema import validate
+
+from metaseed.brapi import import_brapi, to_brapi
+
+BASE_URL = "https://test-server.brapi.org/brapi/v2"
+
+_OU_SCHEMA = {
+    "type": "object",
+    "required": ["observationUnitDbId"],
+    "not": {"required": ["observationLevel"]},  # must be nested, not top-level
+    "properties": {
+        "observationUnitPosition": {
+            "type": "object",
+            "properties": {"observationLevel": {"type": "object"}},
+        }
+    },
+}
+_REQUIRED_ID = {
+    "studies": "studyDbId",
+    "germplasm": "germplasmDbId",
+    "observationUnits": "observationUnitDbId",
+}
+
+
+@pytest.mark.network
+def test_brapi_export_matches_v2_shape():
+    bodies = to_brapi(import_brapi(BASE_URL))
+    for collection, id_field in _REQUIRED_ID.items():
+        for obj in bodies.get(collection, []):
+            assert id_field in obj, f"{collection} object missing {id_field}"
+    for unit in bodies.get("observationUnits", []):
+        validate(instance=unit, schema=_OU_SCHEMA)

tests/test_metabolights/test_conformance.py

@@ -0,0 +1,46 @@
+"""Conformance of the MetaboLights export to the ISA-Tab structure.
+
+Structural validation (ISA-Tab has no installable schema here): the investigation
+file must carry the required ISA sections, and the bundle must include the study,
+assay, and MAF files with their mandatory headers. Opt-in (network).
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from metaseed.metabolights import import_accession, to_metabolights
+
+_REQUIRED_SECTIONS = (
+    "ONTOLOGY SOURCE REFERENCE",
+    "INVESTIGATION",
+    "STUDY",
+    "STUDY FACTORS",
+    "STUDY ASSAYS",
+    "STUDY PROTOCOLS",
+    "STUDY CONTACTS",
+)
+
+
+def _isatab_violations(docs: dict[str, str]) -> list[str]:
+    errors = []
+    inv = docs.get("i_Investigation.txt", "")
+    errors += [f"missing section {s}" for s in _REQUIRED_SECTIONS if s not in inv]
+    if "Investigation Identifier" not in inv:
+        errors.append("no Investigation Identifier row")
+    if not [n for n in docs if n.startswith("s_")]:
+        errors.append("no study (s_) file")
+    if not [n for n in docs if n.startswith("a_")]:
+        errors.append("no assay (a_) file")
+    if not [n for n in docs if n.startswith("m_")]:
+        errors.append("no MAF (m_) file")
+    for name in (n for n in docs if n.startswith("s_")):
+        if "Sample Name" not in docs[name].splitlines()[0]:
+            errors.append(f"{name} missing 'Sample Name' header")
+    return errors
+
+
+@pytest.mark.network
+def test_metabolights_export_is_isatab_conformant():
+    docs = to_metabolights(import_accession("MTBLS1"))
+    assert _isatab_violations(docs) == []

tests/test_pride/test_conformance.py

@@ -0,0 +1,55 @@
+"""Conformance of the PRIDE ``submission.px`` export to the px-submission format.
+
+Structural validation: PRIDE has no external schema (unlike ENA's SRA XSD), so
+this asserts the px-submission-tool structure — the required ``MTD`` metadata
+lines and a well-formed ``FMH``/``FME`` file-mapping section. Opt-in (network):
+exports a real project and checks the result would be structurally accepted.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from metaseed.pride import import_accession, to_pride_submission
+
+# MTD keys the px-submission tool requires for a project.
+_REQUIRED_MTD = (
+    "project_title",
+    "project_description",
+    "submitter_name",
+    "submitter_email",
+    "submission_type",
+)
+
+
+def _px_violations(text: str) -> list[str]:
+    lines = [ln for ln in text.splitlines() if ln]
+    mtd = {
+        parts[1]
+        for ln in lines
+        if (parts := ln.split("\t"))[0] == "MTD" and len(parts) >= 3
+    }
+    errors = [f"missing MTD {k}" for k in _REQUIRED_MTD if k not in mtd]
+    if "species" not in mtd:
+        errors.append("no species line")
+    if "instrument" not in mtd:
+        errors.append("no instrument line")
+
+    if not any(ln.startswith("FMH\t") for ln in lines):
+        errors.append("missing FMH header")
+    fme = [ln for ln in lines if ln.startswith("FME\t")]
+    if not fme:
+        errors.append("no FME file entries")
+    for ln in fme:
+        cols = ln.split("\t")
+        if len(cols) != 4:
+            errors.append(f"FME has {len(cols)} columns, expected 4: {ln!r}")
+        elif not cols[1].isdigit():
+            errors.append(f"FME file_id not numeric: {ln!r}")
+    return errors
+
+
+@pytest.mark.network
+def test_pride_export_is_px_conformant():
+    px = to_pride_submission(import_accession("PXD000001"))["submission.px"]
+    assert _px_violations(px) == []