P1: file parsers, sample sheet, discovery, CLI inspect

Readers for IDAT v3, BPM (locus entries v4-v8), CSV manifest, EGT (data blocks 5-9) and GTC v3-v5 (plus a layout-preserving GTC writer); sample-sheet validation (IN-001..IN-014), IDAT discovery and pairing (IN-020..IN-022), message catalogue and exit codes (CLI-014/016), JSON Lines logging (CLI-007/008), atomic writes (CLI-011), run manifest, and `simurg inspect`. Validation (docs/validation/P1.md): GTC, BPM and EGT readers agree exactly with Illumina BeadArrayFiles on GSA v4 / GDA v1 files; GTCs round-trip byte-identically; BPM and CSV manifests are consistent.
2026-09-27 16:56:39 +01:00
1 parent 0a421dffc9
commit 5a20e7131c
@@ -0,0 +1,68 @@
+from __future__ import annotations
+
+import gzip
+
+import numpy as np
+import pytest
+from hypothesis import given
+from hypothesis import strategies as st
+
+from simurg.core.errors import FormatError
+from simurg.formats.binutil import BinaryReader, BinaryWriter, load_bytes
+
+
+@given(st.text(max_size=400))
+def test_string_roundtrip(s: str) -> None:
+    w = BinaryWriter()
+    w.string(s)
+    r = BinaryReader(w.getvalue(), "E_GTC_CORRUPT")
+    assert r.string() == s
+    assert r.remaining() == 0
+
+
+@pytest.mark.parametrize("n", [0, 1, 127, 128, 16383, 16384, 300_000])
+def test_varint_boundaries(n: int) -> None:
+    w = BinaryWriter()
+    w.string(b"x" * n)
+    r = BinaryReader(w.getvalue(), "E_GTC_CORRUPT")
+    assert r.varint_length() == n
+
+
+def test_invalid_utf8_is_replaced_not_raised() -> None:  # IN-033, EDGE-002
+    w = BinaryWriter()
+    w.string(b"ok\xff\xfe")
+    assert BinaryReader(w.getvalue(), "E_GTC_CORRUPT").string() == "ok��"
+
+
+def test_string_past_eof_is_format_error() -> None:
+    buf = bytes([50]) + b"short"
+    with pytest.raises(FormatError) as e:
+        BinaryReader(buf, "E_GTC_CORRUPT", "x.gtc").string()
+    assert e.value.code == "E_GTC_CORRUPT"
+
+
+def test_bad_varint_is_format_error() -> None:
+    with pytest.raises(FormatError):
+        BinaryReader(b"\xff\xff\xff\xff\xff\xff", "E_BPM_FORMAT").varint_length()
+
+
+def test_array_bounds_and_seek() -> None:
+    r = BinaryReader(np.arange(4, dtype="<i4").tobytes(), "E_IDAT_FORMAT")
+    assert r.array("<i4", 4).tolist() == [0, 1, 2, 3]
+    r.seek(0)
+    with pytest.raises(FormatError):
+        r.array("<i4", 5)
+    with pytest.raises(FormatError):
+        r.seek(100)
+
+
+def test_load_bytes_gzip(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "a.bin.gz"
+    p.write_bytes(gzip.compress(b"hello"))
+    assert bytes(load_bytes(p)) == b"hello"
+    q = tmp_path / "b.bin"
+    q.write_bytes(b"world")
+    assert bytes(load_bytes(q)) == b"world"
+    e = tmp_path / "empty.bin"
+    e.write_bytes(b"")
+    assert bytes(load_bytes(e)) == b""
@@ -0,0 +1,139 @@
+"""BPM and EGT readers on synthetic files (spec IN-040, IN-044; plan §6.2 normalization-bin rule)."""
+
+from __future__ import annotations
+
+import numpy as np
+import pytest
+
+from simurg.core.errors import FormatError
+from simurg.formats.bpm import infer_assay_type, parse_control_config, read_bpm
+from simurg.formats.egt import read_egt
+from simurg.synth.files import SynthLocus, encode_bpm, encode_egt
+
+
+def _loci() -> list[SynthLocus]:
+    return [
+        SynthLocus("rs1", "1", 100, "[A/G]", 11, 0, 0, norm_id=1),
+        SynthLocus("rs2", "2", 200, "[A/T]", 12, 13, 1, norm_id=1, ilmn_strand="BOT", ref_strand="-"),
+        SynthLocus("rs3", "X", 300, "[C/G]", 14, 15, 2, norm_id=99),
+        SynthLocus("rs4", "Y", 400, "[T/C]", 16, 0, 0, norm_id=2, intensity_only=1),
+        SynthLocus("rs5", "MT", 500, "[A/C]", 17, 0, 0, norm_id=99),
+    ]
+
+
+def test_bpm_roundtrip(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "m.bpm"
+    p.write_bytes(encode_bpm(_loci(), control_config="0000000001,Staining,Red,DNP (High)\n"))
+    m = read_bpm(p)
+    assert m.names == ["rs1", "rs2", "rs3", "rs4", "rs5"]
+    assert m.chroms == ["1", "2", "X", "Y", "MT"]
+    assert m.map_infos.tolist() == [100, 200, 300, 400, 500]
+    assert m.address_a.tolist() == [11, 12, 14, 16, 17]
+    assert m.address_b.tolist() == [0, 13, 15, 0, 0]
+    assert m.assay_types.tolist() == [0, 1, 2, 0, 0]
+    assert m.intensity_only.tolist() == [False, False, False, True, False]
+    assert m.ref_strands == ["+", "-", "+", "+", "+"]
+    assert m.ilmn_strands[1] == "BOT"
+    assert m.controls == [{"address": "0000000001", "category": "Staining", "color": "Red", "name": "DNP (High)"}]
+    assert m.header_lines == ["Illumina, Inc."]
+
+
+def test_norm_bins_mod256_and_rank(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    # rs3: 99 + 100*2 = 299 -> 43 (byte wrap) ; rs2: 1 + 100 = 101
+    p = tmp_path / "m.bpm"
+    p.write_bytes(encode_bpm(_loci()))
+    m = read_bpm(p)
+    assert m.norm_bin_ids().tolist() == [1, 101, 43, 2, 99]
+    # sorted distinct: [1, 2, 43, 99, 101] -> ranks
+    assert m.normalization_lookups().tolist() == [0, 4, 2, 1, 3]
+
+
+def test_entries_mapped_by_name(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "m.bpm"
+    p.write_bytes(encode_bpm(_loci(), entry_order=[4, 3, 2, 1, 0]))
+    m = read_bpm(p)
+    assert m.map_infos.tolist() == [100, 200, 300, 400, 500]
+
+
+@pytest.mark.parametrize("entry_version", [4, 6, 7, 8])
+def test_entry_versions(tmp_path, entry_version) -> None:  # type: ignore[no-untyped-def]
+    loci = [
+        SynthLocus("rs1", source_seq="ACGT[A/G]TTGA", probe_a="ACGTA"),
+        SynthLocus("rs2", address_b=5, assay_type=1, source_seq="AAAT[A/T]AAAC", probe_a="ACGTA", probe_b="ACGTT"),
+    ]
+    p = tmp_path / "m.bpm"
+    p.write_bytes(encode_bpm(loci, entry_version=entry_version))
+    m = read_bpm(p)
+    assert m.assay_types.tolist() == [0, 1]
+    assert (m.base_fractions is not None) == (entry_version >= 7)
+    assert m.has_ref_strand == (entry_version >= 8)
+
+
+def test_bpm_errors(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    good = encode_bpm(_loci())
+    cases = {
+        "magic": b"XPM" + good[3:],
+        "truncated": good[: len(good) // 2],
+        "bad_assay": encode_bpm([SynthLocus("rs1", address_b=0, assay_type=1)]),
+        "bad_norm": encode_bpm([SynthLocus("rs1", norm_id=150)]),
+    }
+    for label, data in cases.items():
+        p = tmp_path / f"{label}.bpm"
+        p.write_bytes(data)
+        with pytest.raises(FormatError) as e:
+            read_bpm(p)
+        assert e.value.code == "E_BPM_FORMAT", label
+
+
+def test_long_manifest_name(tmp_path) -> None:  # type: ignore[no-untyped-def]  # EDGE-006
+    name = "X" * 200 + "_ünicode"
+    p = tmp_path / "long name (1).bpm"
+    p.write_bytes(encode_bpm(_loci(), manifest_name=name))
+    assert read_bpm(p).manifest_name == name
+
+
+def test_infer_assay_type() -> None:
+    assert infer_assay_type("ACGT", "", "AC[A/G]TT") == 0
+    assert infer_assay_type("ACGA", "ACGT", "AAA[A/T]TTT") == 1
+    assert infer_assay_type("ACGA", "ACGT", "CCC[C/G]GGG") == 2
+    assert infer_assay_type("", "", "") == 255
+
+
+def test_control_config_parse() -> None:
+    txt = "0027630314:0027630314,Staining,Red,DNP (High)\n\nbad line\n1,Hyb,Green,Hyb (High), extra\n"
+    assert parse_control_config(txt)[1]["name"] == "Hyb (High),extra"
+    assert len(parse_control_config(txt)) == 2
+
+
+@pytest.mark.parametrize("dbv", [9, 8, 7, 6, 5, 4])
+def test_egt_versions(tmp_path, dbv) -> None:  # type: ignore[no-untyped-def]
+    names = ["rs1", "rs2"]
+    tm = [[0.05, 0.5, 0.95], [0.1, 0.45, 0.9]]
+    rm = [[1.0, 1.1, 0.9], [0.8, 0.9, 1.0]]
+    p = tmp_path / "c.egt"
+    p.write_bytes(encode_egt(names, theta_mean=tm, r_mean=rm, intensity_threshold=[0.2, 0.0], data_block_version=dbv))
+    e = read_egt(p)
+    assert e.names == names
+    np.testing.assert_allclose(e.theta_mean, tm, rtol=1e-6)
+    np.testing.assert_allclose(e.r_mean, rm, rtol=1e-6)
+    assert e.n.tolist() == [[100, 100, 100]] * 2
+    if dbv >= 7:
+        assert e.intensity_threshold.tolist()[0] == pytest.approx(0.2)
+    else:
+        assert np.isnan(e.intensity_threshold).all()
+    np.testing.assert_allclose(e.mean_r(), np.mean(rm, axis=1), rtol=1e-6)
+
+
+def test_egt_count_mismatch(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "c.egt"
+    p.write_bytes(encode_egt(["rs1"], theta_mean=[[0.1, 0.5, 0.9]], r_mean=[[1, 1, 1]], corrupt_counts=True))
+    with pytest.raises(FormatError) as e:
+        read_egt(p)
+    assert e.value.code == "E_EGT_LOAD"
+
+
+def test_egt_bad_version(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "c.egt"
+    p.write_bytes(b"\x04\x00\x00\x00" + b"\x00" * 50)
+    with pytest.raises(FormatError):
+        read_egt(p)
@@ -0,0 +1,95 @@
+"""CLI behaviour (CLI-007, CLI-014, CLI-020) and the message catalogue (CLI-016)."""
+
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+import numpy as np
+from typer.testing import CliRunner
+
+from simurg.cli.main import app
+from simurg.core.errors import CATALOGUE, render_markdown
+from simurg.synth.files import SynthIdat, SynthLocus, encode_bpm, encode_egt, encode_idat
+
+ROOT = Path(__file__).resolve().parents[2]
+runner = CliRunner()
+
+
+def _files(tmp_path: Path) -> dict[str, Path]:
+    idat = tmp_path / "200000000001_R01C01_Grn.idat"
+    idat.write_bytes(
+        encode_idat(SynthIdat(address_ids=np.arange(1, 6, dtype=np.int32), mean=np.arange(5, dtype=np.uint16)))
+    )
+    bpm = tmp_path / "m.bpm"
+    bpm.write_bytes(encode_bpm([SynthLocus("rs1"), SynthLocus("rs2", address_a=2)]))
+    egt = tmp_path / "c.egt"
+    egt.write_bytes(encode_egt(["rs1"], theta_mean=[[0.1, 0.5, 0.9]], r_mean=[[1, 1, 1]]))
+    return {"idat": idat, "bpm": bpm, "egt": egt}
+
+
+def test_version() -> None:
+    r = runner.invoke(app, ["--version"])
+    assert r.exit_code == 0 and r.stdout.startswith("SimurgArray ")
+
+
+def test_inspect_json(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    f = _files(tmp_path)
+    r = runner.invoke(app, ["inspect", "--json", str(f["idat"]), str(f["bpm"]), str(f["egt"])])
+    assert r.exit_code == 0, r.output
+    out = json.loads(r.stdout)
+    assert [o["kind"] for o in out] == ["idat", "bpm", "egt"]
+    assert out[1]["num_loci"] == 2 and out[2]["num_records"] == 1
+
+
+def test_inspect_missing_file_exit2() -> None:
+    r = runner.invoke(app, ["inspect", "/nonexistent/x.idat"])
+    assert r.exit_code == 2
+    assert "E_FILE_NOT_FOUND" in r.output
+
+
+def test_inspect_partial_failure_exit3(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    f = _files(tmp_path)
+    bad = tmp_path / "bad.gtc"
+    bad.write_bytes(b"gtc\x05" + b"\x00" * 3)
+    r = runner.invoke(app, ["inspect", "--log-json", str(f["idat"]), str(bad)])
+    assert r.exit_code == 3
+    line = next(ln for ln in r.output.splitlines() if ln.startswith('{"schema"'))
+    rec = json.loads(line)
+    assert rec["schema"] == "simurg.log/1" and rec["code"] == "E_GTC_CORRUPT" and rec["level"] == "error"
+
+
+def test_unknown_option_is_usage_error(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    r = runner.invoke(app, ["inspect", "--no-such-option", "x"])
+    assert r.exit_code == 2
+
+
+def test_catalogue_integrity() -> None:
+    for code, d in CATALOGUE.items():
+        assert code == d.code
+        assert code[:2] in ("E_", "W_", "I_")
+        assert d.level in ("error", "warning", "info")
+
+
+def test_messages_doc_is_current() -> None:
+    doc = ROOT / "docs" / "messages.md"
+    assert doc.read_text(encoding="utf-8") == render_markdown(), "regenerate: python -m simurg.core.errors"
+
+
+def test_locale_independent_output(tmp_path) -> None:  # type: ignore[no-untyped-def]  # EDGE-001
+    f = _files(tmp_path)
+    outs = set()
+    for loc in ("C.UTF-8", "tr_TR.UTF-8", "de_DE.UTF-8"):
+        env = {"LC_ALL": loc, "PATH": "/usr/bin:/bin"}
+        r = subprocess.run(
+            [sys.executable, "-m", "simurg.cli.main", "inspect", "--json", str(f["bpm"])],
+            capture_output=True,
+            text=True,
+            env=env,
+            cwd=ROOT,
+        )
+        assert r.returncode == 0, r.stderr
+        outs.add(r.stdout)
+    assert len(outs) == 1
@@ -0,0 +1,120 @@
+"""IDAT discovery (IN-020 … IN-022, EDGE-005) and atomic writes / overwrite policy (CLI-010, CLI-011)."""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+import numpy as np
+import pytest
+
+from simurg.core.atomic import atomic_open, atomic_write_bytes, check_outputs_absent
+from simurg.core.errors import SimurgError
+from simurg.formats import discovery as disc
+from simurg.formats.samplesheet import parse_samplesheet
+from simurg.synth.files import SynthIdat, encode_idat
+
+
+def write_pair(
+    folder: Path,
+    barcode: str,
+    pos: str,
+    scan: str = "12/27/2023 5:52:21 PM",
+    channels: tuple[str, ...] = ("Grn", "Red"),
+) -> None:
+    folder.mkdir(parents=True, exist_ok=True)
+    s = SynthIdat(
+        address_ids=np.arange(1, 11, dtype=np.int32),
+        mean=np.arange(10, dtype=np.uint16),
+        barcode=barcode,
+        position=pos,
+        run_info=[(scan, "Scan", "", "iScan", "1")],
+    )
+    for ch in channels:
+        (folder / f"{barcode}_{pos}_{ch}.idat").write_bytes(encode_idat(s))
+
+
+def test_pairs_and_unpaired(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    write_pair(tmp_path / "a", "200000000001", "R01C01")
+    write_pair(tmp_path / "b" / "deep", "200000000001", "R02C01", channels=("Grn",))
+    d = disc.discover_idats(tmp_path)
+    assert sorted(d.pairs) == ["200000000001_R01C01", "200000000001_R02C01"]
+    assert d.pairs["200000000001_R01C01"].complete
+    assert not d.pairs["200000000001_R02C01"].complete
+    assert [c for c, _ in d.warnings] == ["W_IDAT_UNPAIRED"]
+
+
+def test_duplicate_pairs_error_lists_all_paths(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    write_pair(tmp_path / "run1", "200000000001", "R01C01")
+    write_pair(tmp_path / "run2", "200000000001", "R01C01")
+    with pytest.raises(SimurgError) as e:
+        disc.discover_idats(tmp_path)
+    assert e.value.code == "E_DUP_IDAT"
+    assert "run1" in e.value.message and "run2" in e.value.message
+
+
+def test_duplicate_first_and_newest_deterministic(tmp_path, monkeypatch) -> None:  # type: ignore[no-untyped-def]
+    write_pair(tmp_path / "z_old", "200000000001", "R01C01", scan="1/1/2020 1:00:00 PM")
+    write_pair(tmp_path / "a_new", "200000000001", "R01C01", scan="1/1/2024 1:00:00 PM")
+    first = disc.discover_idats(tmp_path, on_duplicate="first").pairs["200000000001_R01C01"]
+    newest = disc.discover_idats(tmp_path, on_duplicate="newest").pairs["200000000001_R01C01"]
+    assert first.folder.name == "a_new"  # bytewise smallest path
+    assert newest.folder.name == "a_new"
+    # shuffled directory listing must not change anything
+    real_scandir = os.scandir
+
+    def shuffled(path):  # type: ignore[no-untyped-def]
+        return list(reversed(list(real_scandir(path))))
+
+    monkeypatch.setattr(os, "scandir", shuffled)
+    assert disc.discover_idats(tmp_path, on_duplicate="first").pairs["200000000001_R01C01"].folder.name == "a_new"
+
+
+def test_symlink_loop(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    write_pair(tmp_path / "a", "200000000001", "R01C01")
+    (tmp_path / "a" / "loop").symlink_to(tmp_path, target_is_directory=True)
+    assert list(disc.discover_idats(tmp_path).pairs) == ["200000000001_R01C01"]
+
+
+def test_resolve_with_sheet(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    write_pair(tmp_path / "idats", "200000000001", "R01C01")
+    write_pair(tmp_path / "idats", "200000000001", "R03C01")
+    sh = tmp_path / "s.csv"
+    sh.write_text(
+        "[Header]\nRootFolder,idats\n[Data]\nSentrixBarcode_A,SentrixPosition_A,Sample_Name\n"
+        "200000000001,R01C01,alpha\n200000000001,R02C01,beta\n"
+    )
+    sheet = parse_samplesheet(sh)
+    folder, notes = disc.resolve_input_folder(None, sheet)
+    assert folder == tmp_path / "idats" and notes == []
+    samples, notes = disc.resolve_samples(disc.discover_idats(folder), sheet)
+    assert [(s.sample_name, s.status) for s in samples] == [("alpha", "READY"), ("beta", "MISSING_INPUT")]
+    assert notes[0][0] == "I_UNLISTED_IDATS"
+    _, notes = disc.resolve_input_folder(tmp_path / "other", sheet)
+    assert notes[0][0] == "W_SHEET_ROOTFOLDER_OVERRIDDEN"
+
+
+def test_input_rules() -> None:
+    with pytest.raises(SimurgError) as e:
+        disc.resolve_input_folder(None, None)
+    assert e.value.code == "E_NO_INPUT"
+
+
+def test_atomic_write_and_failure(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "out" / "x.bin"
+    atomic_write_bytes(p, b"abc")
+    assert p.read_bytes() == b"abc"
+    with pytest.raises(RuntimeError), atomic_open(tmp_path / "y.bin") as fh:
+        fh.write(b"partial")  # type: ignore[arg-type]
+        raise RuntimeError("boom")
+    assert not (tmp_path / "y.bin").exists()
+    assert not (tmp_path / "y.bin.part").exists()
+
+
+def test_overwrite_policy(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "x"
+    p.write_text("1")
+    with pytest.raises(SimurgError) as e:
+        check_outputs_absent([p, tmp_path / "missing"], force=False)
+    assert e.value.code == "E_OUTPUT_EXISTS"
+    check_outputs_absent([p], force=True)
@@ -0,0 +1,152 @@
+"""GTC codec: synthetic round-trips, property tests, corruption handling (spec IN-032, IN-033, EDGE-002)."""
+
+from __future__ import annotations
+
+import numpy as np
+import pytest
+from hypothesis import given, settings
+from hypothesis import strategies as st
+
+from simurg.core.errors import FormatError
+from simurg.formats.gtc import (
+    CANONICAL_ORDER_V3,
+    CANONICAL_ORDER_V5,
+    XFORM_DTYPE,
+    Gtc,
+    ScannerData,
+    encode_gtc,
+    read_gtc,
+    write_gtc,
+)
+
+
+def make_gtc(n: int = 20, version: int = 5, seed: int = 0) -> Gtc:
+    rng = np.random.default_rng(seed)
+    xf = np.zeros(3, XFORM_DTYPE)
+    xf["version"] = 1
+    xf["offset_x"] = rng.normal(size=3)
+    xf["theta"] = rng.normal(size=3)
+    g = Gtc(
+        version=version,
+        num_snps=n,
+        sample_name="NA12878 ünicode",
+        cluster_file="c.egt",
+        snp_manifest="m.bpm",
+        imaging_date="12/27/2023 5:53:34 PM",
+        autocall_date="01/01/26 12:00 PM",
+        autocall_version="simurg-0.1",
+        normalization_transforms=xf,
+        controls_x=rng.integers(0, 60000, 8).astype(np.uint16),
+        controls_y=rng.integers(0, 60000, 8).astype(np.uint16),
+        raw_x=rng.integers(0, 60000, n).astype(np.uint16),
+        raw_y=rng.integers(0, 60000, n).astype(np.uint16),
+        genotypes=rng.integers(0, 4, n).astype(np.uint8),
+        base_calls=np.frombuffer(b"AG" * n, np.uint8).reshape(n, 2).copy(),
+        genotype_scores=rng.random(n).astype(np.float32),
+        scanner_data=ScannerData("N103", b"\x01\x00\x00\x00", b"\x02\x00\x00\x00", "4.2.1", "user"),
+        call_rate=0.99,
+        gender=b"F",
+        logr_dev=0.12,
+        gc10=0.4,
+        b_allele_freqs=rng.random(n).astype(np.float32),
+        logr_ratios=rng.normal(size=n).astype(np.float32),
+        percentiles_x=np.array([1, 2, 3], np.uint16),
+        percentiles_y=np.array([4, 5, 6], np.uint16),
+        slide_identifier="200000000001",
+    )
+    g.sample_data["p50gc"] = 0.7
+    g.sample_data["num_calls"] = n - 2
+    g.sample_data["num_no_calls"] = 2
+    return g
+
+
+@pytest.mark.parametrize("version", [3, 5])
+def test_roundtrip_canonical(tmp_path, version) -> None:  # type: ignore[no-untyped-def]
+    g = make_gtc(version=version)
+    p = tmp_path / "s.gtc"
+    write_gtc(g, p)
+    h = read_gtc(p)
+    assert h.toc_order == (CANONICAL_ORDER_V5 if version == 5 else CANONICAL_ORDER_V3)
+    assert h.sample_name == g.sample_name
+    assert np.array_equal(h.raw_x, g.raw_x) and np.array_equal(h.genotypes, g.genotypes)
+    assert np.array_equal(h.genotype_scores, g.genotype_scores)
+    assert h.scanner_data == g.scanner_data
+    assert h.gc50 == pytest.approx(0.7) and h.num_calls == 18
+    assert h.gender_char == "F"
+    assert h.base_call_strings()[0] == "AG"
+    assert encode_gtc(h) == p.read_bytes()  # byte-identical round-trip
+    if version == 5:
+        assert np.array_equal(h.logr_ratios, g.logr_ratios)
+
+
+def test_noncanonical_layout_preserved(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    g = make_gtc()
+    g.toc_order = list(reversed(CANONICAL_ORDER_V5))
+    g.data_order = [t for t in CANONICAL_ORDER_V5 if t > 3][::-1]
+    data = encode_gtc(g)
+    p = tmp_path / "s.gtc"
+    p.write_bytes(data)
+    h = read_gtc(p)
+    assert h.toc_order == g.toc_order and h.data_order == g.data_order
+    assert encode_gtc(h) == data
+
+
+@settings(max_examples=25, deadline=None)
+@given(n=st.integers(0, 200), seed=st.integers(0, 10**6), name=st.text(max_size=300))
+def test_property_roundtrip(tmp_path_factory, n, seed, name) -> None:  # type: ignore[no-untyped-def]
+    g = make_gtc(n=n, seed=seed)
+    g.sample_name = name
+    data = encode_gtc(g)
+    p = tmp_path_factory.mktemp("g") / "x.gtc"
+    p.write_bytes(data)
+    h = read_gtc(p)
+    assert h.sample_name == name and h.num_snps == n
+    assert encode_gtc(h) == data
+
+
+def test_truncated_and_bad_magic(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    data = encode_gtc(make_gtc())
+    for label, blob in {
+        "trunc": data[: len(data) // 3],
+        "magic": b"xtc" + data[3:],
+        "tiny": data[:5],
+        "version": data[:3] + b"\x09" + data[4:],
+    }.items():
+        p = tmp_path / f"{label}.gtc"
+        p.write_bytes(blob)
+        with pytest.raises(FormatError) as e:
+            read_gtc(p)
+        assert e.value.code == "E_GTC_CORRUPT", label
+
+
+def test_bad_string_length_is_clean_error(tmp_path) -> None:  # type: ignore[no-untyped-def]  # EDGE-002
+    g = make_gtc()
+    data = bytearray(encode_gtc(g))
+    header = 8 + 6 * len(CANONICAL_ORDER_V5)
+    data[header] = 0xFF  # sample-name length prefix now runs past EOF
+    data[header + 1] = 0xFF
+    data[header + 2] = 0xFF
+    data[header + 3] = 0x7F
+    p = tmp_path / "s.gtc"
+    p.write_bytes(bytes(data))
+    with pytest.raises(FormatError) as e:
+        read_gtc(p)
+    assert e.value.code == "E_GTC_CORRUPT"
+
+
+def test_incomplete_write_detection(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    g = make_gtc()
+    g.toc_order = [t for t in CANONICAL_ORDER_V5 if t != 1016]
+    g.data_order = [t for t in g.toc_order if t > 3]
+    p = tmp_path / "s.gtc"
+    p.write_bytes(encode_gtc(g))
+    with pytest.raises(FormatError):
+        read_gtc(p)
+    assert read_gtc(p, check_write_complete=False).num_snps == 20
+
+
+def test_length_mismatch_rejected_on_write() -> None:
+    g = make_gtc()
+    g.raw_x = g.raw_x[:-1]
+    with pytest.raises(ValueError):
+        encode_gtc(g)
@@ -0,0 +1,128 @@
+"""IDAT reader: synthetic files + malformed-input matrix (spec IN-023, EDGE-004, EDGE-033)."""
+
+from __future__ import annotations
+
+import gzip
+import struct
+
+import numpy as np
+import pytest
+
+from simurg.core.errors import FormatError
+from simurg.formats.idat import check_pair, parse_idat_filename, parse_run_time, read_idat
+from simurg.synth.files import SynthIdat, encode_idat
+
+
+def _synth(n: int = 50, **kw) -> SynthIdat:  # type: ignore[no-untyped-def]
+    rng = np.random.default_rng(1)
+    ids = np.sort(rng.choice(10_000_000, n, replace=False)).astype(np.int32)
+    return SynthIdat(
+        address_ids=ids,
+        mean=rng.integers(0, 60000, n).astype(np.uint16),
+        n_beads=rng.integers(0, 30, n).astype(np.uint8),
+        **kw,
+    )
+
+
+def test_roundtrip(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = _synth(sample_id="NA12878", plate="P1", well="A01")
+    p = tmp_path / "200000000001_R01C01_Grn.idat"
+    p.write_bytes(encode_idat(s))
+    i = read_idat(p)
+    assert i.n_addresses == 50
+    assert np.array_equal(i.address_ids, s.address_ids)
+    assert np.array_equal(i.mean, s.mean)
+    assert np.array_equal(i.n_beads, s.n_beads)
+    assert (i.barcode, i.position, i.sample_id, i.plate, i.well) == ("200000000001", "R01C01", "NA12878", "P1", "A01")
+    assert i.scan_time_iso == "2023-12-27T17:52:21"
+    assert i.scanner_software == "4.2.1.729"
+    assert i.mid_block_itemsize == 4
+
+
+def test_midblock_uint8_inferred(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "x.idat"
+    p.write_bytes(encode_idat(_synth(n=40, mid_block_itemsize=1)))
+    assert read_idat(p).mid_block_itemsize == 1
+
+
+def test_gzip(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "200000000001_R01C01_Red.idat.gz"
+    p.write_bytes(gzip.compress(encode_idat(_synth())))
+    assert read_idat(p).n_addresses == 50
+
+
+def test_metadata_only(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "x.idat"
+    p.write_bytes(encode_idat(_synth()))
+    i = read_idat(p, load_arrays=False)
+    assert i.n_addresses == 50 and len(i.mean) == 0 and i.barcode == "200000000001"
+
+
+@pytest.mark.parametrize(
+    "mutate,code",
+    [
+        (lambda b: b"XDAT" + b[4:], "E_IDAT_FORMAT"),  # wrong magic
+        (lambda b: b[:4] + struct.pack("<q", 4) + b[12:], "E_IDAT_FORMAT"),  # unsupported version
+        (lambda b: b[:4] + struct.pack("<q", 1 | (7 << 32)) + b[12:], "E_IDAT_ENCRYPTED"),  # encrypted v1
+        (lambda b: b[: len(b) // 2], "E_IDAT_FORMAT"),  # truncated
+        (lambda b: b[:18] + struct.pack("<q", 10**9) + b[26:], "E_IDAT_FORMAT"),  # TOC offset outside file
+        (lambda b: b[:8], "E_IDAT_FORMAT"),  # tiny
+    ],
+)
+def test_malformed(tmp_path, mutate, code) -> None:  # type: ignore[no-untyped-def]
+    p = tmp_path / "bad.idat"
+    p.write_bytes(mutate(encode_idat(_synth())))
+    with pytest.raises(FormatError) as e:
+        read_idat(p)
+    assert e.value.code == code
+
+
+def test_pair_mismatch(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    a, b = tmp_path / "a.idat", tmp_path / "b.idat"
+    a.write_bytes(encode_idat(_synth(n=50)))
+    b.write_bytes(encode_idat(_synth(n=49)))
+    with pytest.raises(FormatError) as e:
+        check_pair(read_idat(a), read_idat(b))
+    assert e.value.code == "E_IDAT_PAIR_MISMATCH"
+
+
+@pytest.mark.parametrize(
+    "name,expected",
+    [
+        ("207895980019_R03C01_Grn.idat", ("207895980019", "R03C01", "Grn")),
+        ("207895980019_r03c01_red.idat.gz", ("207895980019", "R03C01", "Red")),
+        ("5723646052_R02C02_GRN.IDAT", ("5723646052", "R02C02", "Grn")),
+        ("4019585376_B_Red.idat", None),
+        ("notes.txt", None),
+    ],
+)
+def test_filename(name, expected) -> None:  # type: ignore[no-untyped-def]
+    assert parse_idat_filename(name) == expected
+
+
+@pytest.mark.parametrize(
+    "raw,iso",
+    [
+        ("12/27/2023 5:53:34 PM", "2023-12-27T17:53:34"),
+        ("1/2/2020 12:00:01 AM", "2020-01-02T00:00:01"),
+        ("2020-01-02 13:14:15", "2020-01-02T13:14:15"),
+        ("", ""),
+        ("garbage", ""),
+    ],
+)
+def test_run_time(raw, iso) -> None:  # type: ignore[no-untyped-def]
+    assert parse_run_time(raw) == iso
+
+
+def test_duplicate_addresses_rejected(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    from simurg.formats.idat import has_duplicates
+
+    assert not has_duplicates(np.array([1, 5, 9], np.int32))
+    assert not has_duplicates(np.array([9, 1, 5], np.int32))
+    assert has_duplicates(np.array([9, 1, 9], np.int32))
+    s = _synth(n=5)
+    s.address_ids = np.array([5, 1, 5, 7, 8], np.int32)
+    p = tmp_path / "dup.idat"
+    p.write_bytes(encode_idat(s))
+    with pytest.raises(FormatError):
+        read_idat(p)
@@ -0,0 +1,160 @@
+"""Sample-sheet matrix — spec EDGE-007 (IN-001 … IN-014)."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from simurg.core.errors import SimurgError
+from simurg.formats.samplesheet import output_stem, parse_samplesheet
+
+HEAD = "[Header]\nInvestigator Name,X\n[Manifests]\nA,foo.bpm\n[Data]\n"
+
+
+def sheet(tmp_path: Path, body: str, name: str = "s.csv", raw: bytes | None = None) -> Path:
+    p = tmp_path / name
+    if raw is not None:
+        p.write_bytes(raw)
+    else:
+        p.write_text(body, encoding="utf-8")
+    return p
+
+
+def code_of(tmp_path: Path, body: str) -> str:
+    with pytest.raises(SimurgError) as e:
+        parse_samplesheet(sheet(tmp_path, body))
+    return e.value.code
+
+
+def test_basic_sections_and_passthrough(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(
+        sheet(
+            tmp_path,
+            HEAD + "Sample_ID,SentrixBarcode_A,SentrixPosition_A,Gender,Tissue\na,203790110013,r04c01,Female,blood\n",
+        )
+    )
+    assert s.rows[0].sample_id == "203790110013_R04C01"
+    assert s.rows[0].reported_sex == "F"
+    assert s.passthrough_columns() == ["Sample_ID", "Gender", "Tissue"]
+    assert s.header == {"Investigator Name": "X"}
+
+
+def test_sectionless(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(sheet(tmp_path, "SentrixBarcode_A,SentrixPosition_A\n203790110013,R01C01\n"))
+    assert len(s.rows) == 1
+
+
+def test_bom_crlf_quotes(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    raw = ('[Data]\r\nSentrixBarcode_A,SentrixPosition_A,Note\r\n203790110013,R01C01,"a, b"\r\n').encode()
+    s = parse_samplesheet(sheet(tmp_path, "", raw=raw))
+    assert s.rows[0].values["Note"] == "a, b"
+
+
+def test_trailing_empty_columns(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(sheet(tmp_path, "[Data]\nSentrixBarcode_A,SentrixPosition_A,,\n203790110013,R01C01,,\n"))
+    assert s.columns == ["SentrixBarcode_A", "SentrixPosition_A"]
+    assert [c for c, _ in s.warnings] == ["W_SHEET_EMPTY_COLUMN"]
+    assert "" not in s.rows[0].values
+
+
+def test_manifests_section_with_values_ignored(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    body = (
+        "[Header],,\nDate,1,\n[Manifests],GSA.bpm,\nA,,\n[Data],,\nSentrixBarcode_A,SentrixPosition_A,x\n"
+        "203790110013,R01C01,1\n"
+    )
+    s = parse_samplesheet(sheet(tmp_path, body))
+    assert s.header == {"Date": "1"} and len(s.rows) == 1
+
+
+@pytest.mark.parametrize(
+    "body,code",
+    [
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A,Sample_Name,sample_name\n1,2,3,4\n", "E_SHEET_DUP_COLUMN"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A,Sex Estimate\n203790110013,R01C01,M\n", "E_SHEET_RESERVED"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A,call_rate\n203790110013,R01C01,1\n", "E_SHEET_RESERVED"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A,x\n203790110013,R01C01," + "a" * 501 + "\n", "E_SHEET_TOO_LONG"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A\n2.04753E+11,R01C01\n", "E_SHEET_BARCODE_FORMAT"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A\n203790110013,R1C1\n", "E_SHEET_BARCODE_FORMAT"),
+        ("[Data]\nSentrixBarcode_A;SentrixPosition_A\n203790110013;R01C01\n", "E_SHEET_DELIMITER"),
+        ('{"samples": []}', "E_SHEET_DELIMITER"),
+        ("[Data]\nSample_ID,Foo\n1,2\n", "E_SHEET_NO_KEY_COLUMNS"),
+        ("[Data]\nSentrixBarcode_A,SentrixPosition_A\n", "E_SHEET_EMPTY"),
+        ("[Header]\nA,B\n", "E_SHEET_EMPTY"),
+        (
+            "[Data]\nSentrixBarcode_A,SentrixPosition_A,Sample_Name\n203790110013,R01C01,a\n203790110013,R02C01,\n",
+            "E_SHEET_NAME_EMPTY",
+        ),
+        (
+            "[Data]\nSentrixBarcode_A,SentrixPosition_A,Sample_Name\n203790110013,R01C01,203790110013_R02C01\n"
+            "203790110013,R02C01,b\n",
+            "E_SHEET_NAME_COLLISION",
+        ),
+        (
+            "[Data]\nSentrixBarcode_A,SentrixPosition_A,Sample_Name\n203790110013,R01C01,a\n203790110013,R02C01,a\n",
+            "E_SHEET_NAME_COLLISION",
+        ),
+        (
+            "[Data]\nSentrixBarcode_A,SentrixPosition_A\n203790110013,R01C01\n203790110013,r01c01\n",
+            "E_SHEET_DUP_SAMPLE",
+        ),
+    ],
+)
+def test_error_matrix(tmp_path, body, code) -> None:  # type: ignore[no-untyped-def]
+    assert code_of(tmp_path, body) == code
+
+
+def test_barcode_decimal_repaired(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(sheet(tmp_path, "[Data]\nSentrixBarcode_A,SentrixPosition_A\n204753010023.0,R01C01\n"))
+    assert s.rows[0].barcode == "204753010023"
+    assert s.warnings[0][0] == "W_SHEET_BARCODE_REPAIRED"
+
+
+def test_scientific_notation_hint(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    with pytest.raises(SimurgError) as e:
+        parse_samplesheet(sheet(tmp_path, '[Data]\nSentrixBarcode_A,SentrixPosition_A\n"2,04753E+11",R01C01\n'))
+    assert e.value.hint and "0 decimals" in e.value.hint
+
+
+def test_alias_keys_and_rootfolder(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(
+        sheet(tmp_path, "[Header]\nRootFolder,idats\n[Data]\nbeadChipName,sampleSectionName\n203790110013,R01C01\n")
+    )
+    assert s.barcode_column == "beadChipName"
+    assert s.root_folder == tmp_path / "idats"
+
+
+def test_sex_columns(tmp_path) -> None:  # type: ignore[no-untyped-def]
+    s = parse_samplesheet(
+        sheet(
+            tmp_path,
+            "[Data]\nSentrixBarcode_A,SentrixPosition_A,Gender,Sex\n"
+            "203790110013,R01C01,Male,F\n203790110013,R02C01,,\n",
+        )
+    )
+    assert [r.reported_sex for r in s.rows] == ["F", "U"]
+    assert any(c == "W_SHEET_SEX_CONFLICT" for c, _ in s.warnings)
+
+
+def test_turkish_case_folding_is_locale_independent(tmp_path) -> None:  # type: ignore[no-untyped-def]  # EDGE-001
+    s = parse_samplesheet(sheet(tmp_path, "[Data]\nSENTRIXBARCODE_A,SentrixPosition_A,İl\n203790110013,R01C01,x\n"))
+    assert s.barcode_column == "SENTRIXBARCODE_A"
+
+
+def test_ten_thousand_rows_fast(tmp_path) -> None:  # type: ignore[no-untyped-def]  # PERF-007 (parse part)
+    import time
+
+    lines = ["[Data]", "SentrixBarcode_A,SentrixPosition_A,Sample_Name"]
+    for i in range(10_000):
+        lines.append(f"{200000000000 + i // 24},R{(i % 24) + 1:02d}C01,s{i}")
+    p = sheet(tmp_path, "\n".join(lines) + "\n")
+    t = time.perf_counter()
+    s = parse_samplesheet(p)
+    assert len(s.rows) == 10_000
+    assert time.perf_counter() - t < 2.0
+
+
+def test_output_stem() -> None:
+    assert output_stem("NA 12878/ü") == "NA_12878__"
+    assert output_stem("..hidden") == "hidden"
+    assert len(output_stem("x" * 500)) == 200