Datasets:
Tasks:
Tabular Classification
Formats:
parquet
Languages:
English
Size:
< 1K
Tags:
economics
quantitative-finance
causal-inference
macroeconomics
housing-economics
market-microstructure
License:
| from __future__ import annotations | |
| import json | |
| from pathlib import Path | |
| import pyarrow as pa # type: ignore[import-untyped] | |
| import pytest | |
| from microstructure.data.book import DepthDelta, deltas_table | |
| from microstructure.data.quality import ( | |
| IncrementalQualityValidator, | |
| validate_batches, | |
| validate_table, | |
| ) | |
| from microstructure.data.schemas import table_from_records | |
| from microstructure.data.synthetic import generate_synthetic_market | |
| def test_clean_synthetic_tables_pass_and_validation_does_not_mutate() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=25, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=14, | |
| ) | |
| trade_before = data.trades.to_pylist() | |
| book_before = data.book_observations.to_pylist() | |
| trades = validate_table(data.trades, "trades") | |
| books = validate_table(data.book_observations, "book_observations") | |
| assert not trades.has_errors | |
| assert not books.has_errors | |
| assert data.trades.to_pylist() == trade_before | |
| assert data.book_observations.to_pylist() == book_before | |
| def test_trade_quality_reports_duplicate_and_nonpositive_values_without_repair() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=3, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=15, | |
| ) | |
| records = data.trades.to_pylist() | |
| records[1]["trade_id"] = records[0]["trade_id"] | |
| records[1]["price_ticks"] = -1 | |
| records[1]["price"] = -0.01 | |
| records[2]["quantity_lots"] = 0 | |
| records[2]["quantity"] = 0.0 | |
| invalid = table_from_records("trades", records) | |
| before = invalid.to_pylist() | |
| report = validate_table(invalid, "trades") | |
| rule_ids = {finding.rule_id for finding in report.findings} | |
| assert report.has_errors | |
| assert "trade.duplicate" in rule_ids | |
| assert "trade.nonpositive_price" in rule_ids | |
| assert "trade.nonpositive_quantity" in rule_ids | |
| assert invalid.to_pylist() == before | |
| def test_trade_identity_and_clock_state_are_scoped_by_venue() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=2, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=18, | |
| ) | |
| records = data.trades.to_pylist() | |
| records[1]["venue"] = "second_venue" | |
| records[1]["trade_id"] = records[0]["trade_id"] | |
| cross_venue = table_from_records("trades", records) | |
| report = validate_table(cross_venue, "trades") | |
| assert "trade.duplicate" not in {finding.rule_id for finding in report.findings} | |
| def test_book_quality_reports_sequence_gap_crossed_book_and_clock_discontinuity() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=3, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=16, | |
| ) | |
| records = data.book_observations.to_pylist() | |
| records[1]["sequence_start"] = 4 | |
| records[1]["sequence_end"] = 4 | |
| records[1]["best_bid_ticks"] = records[1]["best_ask_ticks"] + 1 | |
| records[1]["best_bid"] = records[1]["best_ask"] + records[1]["tick_size"] | |
| records[1]["spread"] = records[1]["best_ask"] - records[1]["best_bid"] | |
| records[1]["mid_price"] = (records[1]["best_bid"] + records[1]["best_ask"]) / 2 | |
| records[1]["microprice"] = records[1]["mid_price"] | |
| records[2]["received_ts_ns"] = records[1]["received_ts_ns"] - 1 | |
| records[2]["available_ts_ns"] = max(records[2]["event_ts_ns"], records[2]["received_ts_ns"]) | |
| invalid = table_from_records("book_observations", records) | |
| report = validate_table(invalid, "book_observations") | |
| rule_ids = {finding.rule_id for finding in report.findings} | |
| assert "sequence.missing_range" in rule_ids | |
| assert "book.crossed_or_locked" in rule_ids | |
| assert "temporal.receive_clock_reversal" in rule_ids | |
| def test_book_quality_rejects_float_values_inconsistent_with_exact_scales() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=2, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=17, | |
| ) | |
| records = data.book_observations.to_pylist() | |
| records[0]["best_bid"] += records[0]["tick_size"] / 2.0 | |
| inconsistent = table_from_records("book_observations", records) | |
| report = validate_table(inconsistent, "book_observations") | |
| assert "book.price_scale_mismatch" in {finding.rule_id for finding in report.findings} | |
| def test_zero_depth_quantity_is_valid_delete_but_negative_quantity_is_error() -> None: | |
| common = { | |
| "venue": "binance_spot", | |
| "symbol": "BTCUSDT", | |
| "event_ts_ns": 1_000, | |
| "received_ts_ns": 1_100, | |
| "available_ts_ns": 1_100, | |
| "availability_basis": "local_receive_time", | |
| "capture_seq": 1, | |
| "continuity_id": "session-1", | |
| "first_update_id": 101, | |
| "last_update_id": 101, | |
| "previous_update_id": None, | |
| "asks": (), | |
| "tick_size": 0.01, | |
| "lot_size": 0.001, | |
| "source_artifact_id": "fixture", | |
| } | |
| delete = DepthDelta(bids=((100, 0),), **common) | |
| negative = DepthDelta( | |
| bids=((100, -1),), | |
| **{**common, "event_ts_ns": 2_000, "received_ts_ns": 2_100, "available_ts_ns": 2_100}, | |
| ) | |
| table = deltas_table([delete, negative]) | |
| report = validate_table(table, "depth_deltas") | |
| quantity_findings = [ | |
| finding for finding in report.findings if finding.rule_id == "depth.negative_quantity" | |
| ] | |
| assert len(quantity_findings) == 1 | |
| assert quantity_findings[0].row_index == 1 | |
| def test_depth_delta_quality_reports_gap_stale_and_previous_id_mismatch() -> None: | |
| common = { | |
| "venue": "binance_spot", | |
| "symbol": "BTCUSDT", | |
| "availability_basis": "local_receive_time", | |
| "continuity_id": "session-1", | |
| "bids": ((10_000, 1),), | |
| "asks": (), | |
| "tick_size": 0.01, | |
| "lot_size": 0.001, | |
| "source_artifact_id": "fixture", | |
| } | |
| deltas = [ | |
| DepthDelta( | |
| **common, | |
| event_ts_ns=1_000, | |
| received_ts_ns=1_100, | |
| available_ts_ns=1_100, | |
| capture_seq=1, | |
| first_update_id=101, | |
| last_update_id=102, | |
| previous_update_id=None, | |
| ), | |
| DepthDelta( | |
| **common, | |
| event_ts_ns=2_000, | |
| received_ts_ns=2_100, | |
| available_ts_ns=2_100, | |
| capture_seq=2, | |
| first_update_id=105, | |
| last_update_id=106, | |
| previous_update_id=99, | |
| ), | |
| DepthDelta( | |
| **common, | |
| event_ts_ns=3_000, | |
| received_ts_ns=3_100, | |
| available_ts_ns=3_100, | |
| capture_seq=3, | |
| first_update_id=104, | |
| last_update_id=105, | |
| previous_update_id=None, | |
| ), | |
| ] | |
| report = validate_table(deltas_table(deltas), "depth_deltas") | |
| rule_ids = {finding.rule_id for finding in report.findings} | |
| assert "sequence.missing_range" in rule_ids | |
| assert "sequence.previous_id_mismatch" in rule_ids | |
| assert "sequence.stale_or_duplicate" in rule_ids | |
| def test_incremental_trade_state_crosses_batch_boundaries_with_global_indexes() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=3, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=41, | |
| ) | |
| records = data.trades.to_pylist() | |
| first_event = int(records[0]["event_ts_ns"]) | |
| first_received = int(records[0]["received_ts_ns"]) | |
| records[1]["trade_id"] = records[0]["trade_id"] | |
| records[1]["event_ts_ns"] = first_event - 1 | |
| records[1]["received_ts_ns"] = first_received - 1 | |
| records[1]["available_ts_ns"] = first_received | |
| records[2]["event_ts_ns"] = first_event + 1_000 | |
| records[2]["received_ts_ns"] = first_received + 1_000 | |
| records[2]["available_ts_ns"] = first_received + 1_000 | |
| batches = [table_from_records("trades", [record]) for record in records] | |
| report = validate_batches(batches, "trades", max_silence_ns=100) | |
| by_rule = {finding.rule_id: finding for finding in report.findings} | |
| assert report.rows_checked == 3 | |
| assert by_rule["trade.duplicate"].row_index == 1 | |
| assert by_rule["trade.duplicate"].details["first_row"] == 0 | |
| assert by_rule["temporal.out_of_order_event_time"].row_index == 1 | |
| assert by_rule["temporal.out_of_order_event_time"].details["previous_row"] == 0 | |
| assert by_rule["temporal.receive_clock_reversal"].row_index == 1 | |
| assert by_rule["temporal.long_silence"].row_index == 2 | |
| def test_incremental_book_sequence_gap_is_detected_across_batches() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=2, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=42, | |
| ) | |
| records = data.book_observations.to_pylist() | |
| records[1]["sequence_start"] = int(records[0]["sequence_end"]) + 2 | |
| records[1]["sequence_end"] = records[1]["sequence_start"] | |
| report = validate_batches( | |
| ( | |
| table_from_records("book_observations", [records[0]]), | |
| table_from_records("book_observations", [records[1]]), | |
| ), | |
| "book_observations", | |
| ) | |
| gaps = [finding for finding in report.findings if finding.rule_id == "sequence.missing_range"] | |
| assert len(gaps) == 1 | |
| assert gaps[0].row_index == 1 | |
| assert gaps[0].details == { | |
| "expected_sequence": int(records[0]["sequence_end"]) + 1, | |
| "observed_start": records[1]["sequence_start"], | |
| "missing_start": int(records[0]["sequence_end"]) + 1, | |
| "missing_end": int(records[0]["sequence_end"]) + 1, | |
| } | |
| def test_incremental_depth_sequence_and_previous_hint_cross_batches() -> None: | |
| common = { | |
| "venue": "binance_spot", | |
| "symbol": "BTCUSDT", | |
| "availability_basis": "local_receive_time", | |
| "continuity_id": "session-1", | |
| "bids": ((10_000, 1),), | |
| "asks": (), | |
| "tick_size": 0.01, | |
| "lot_size": 0.001, | |
| "source_artifact_id": "fixture", | |
| } | |
| first = DepthDelta( | |
| **common, | |
| event_ts_ns=1_000, | |
| received_ts_ns=1_100, | |
| available_ts_ns=1_100, | |
| capture_seq=1, | |
| first_update_id=101, | |
| last_update_id=102, | |
| previous_update_id=None, | |
| ) | |
| second = DepthDelta( | |
| **common, | |
| event_ts_ns=2_000, | |
| received_ts_ns=2_100, | |
| available_ts_ns=2_100, | |
| capture_seq=2, | |
| first_update_id=105, | |
| last_update_id=106, | |
| previous_update_id=99, | |
| ) | |
| report = validate_batches( | |
| (deltas_table([first]), deltas_table([second])), | |
| "depth_deltas", | |
| ) | |
| rules = {finding.rule_id: finding for finding in report.findings} | |
| assert rules["sequence.missing_range"].row_index == 1 | |
| assert rules["sequence.previous_id_mismatch"].row_index == 1 | |
| def test_incremental_state_is_isolated_by_venue_across_batches() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=2, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=43, | |
| ) | |
| records = data.trades.to_pylist() | |
| records[1]["venue"] = "second_venue" | |
| records[1]["trade_id"] = records[0]["trade_id"] | |
| records[1]["event_ts_ns"] = int(records[0]["event_ts_ns"]) - 1 | |
| records[1]["received_ts_ns"] = int(records[0]["received_ts_ns"]) - 1 | |
| records[1]["available_ts_ns"] = records[0]["available_ts_ns"] | |
| report = validate_batches( | |
| (table_from_records("trades", [record]) for record in records), | |
| "trades", | |
| ) | |
| rule_ids = {finding.rule_id for finding in report.findings} | |
| assert "trade.duplicate" not in rule_ids | |
| assert "temporal.out_of_order_event_time" not in rule_ids | |
| assert "temporal.receive_clock_reversal" not in rule_ids | |
| class _OneShotBatches: | |
| def __init__(self, batches: list[pa.Table | pa.RecordBatch]) -> None: | |
| self._batches = batches | |
| self.iterations = 0 | |
| def __iter__(self): # type: ignore[no-untyped-def] | |
| self.iterations += 1 | |
| if self.iterations > 1: | |
| raise AssertionError("batch iterable was consumed more than once") | |
| yield from self._batches | |
| def test_validate_batches_consumes_one_shot_iterable_once_and_accepts_record_batches() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=4, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=44, | |
| ) | |
| batches = _OneShotBatches(list(data.trades.to_batches(max_chunksize=1))) | |
| report = validate_batches(batches, "trades", row_chunk_size=1) | |
| assert batches.iterations == 1 | |
| assert report.rows_checked == data.trades.num_rows | |
| assert not report.has_errors | |
| def test_incremental_bounded_preview_keeps_exact_totals_and_streams_all_jsonl( | |
| tmp_path: Path, | |
| ) -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=12, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=45, | |
| ) | |
| records = data.trades.to_pylist() | |
| for record in records: | |
| record["price_ticks"] = -1 | |
| record["price"] = -float(record["tick_size"]) | |
| invalid = table_from_records("trades", records) | |
| before = invalid.to_pylist() | |
| findings_path = tmp_path / "quality" / "findings.jsonl" | |
| report = validate_batches( | |
| invalid.to_batches(max_chunksize=2), | |
| "trades", | |
| max_findings=3, | |
| findings_jsonl_path=findings_path, | |
| row_chunk_size=1, | |
| ) | |
| summary_path = tmp_path / "quality" / "summary.json" | |
| report.write_json(summary_path) | |
| streamed = [json.loads(line) for line in findings_path.read_text().splitlines()] | |
| summary = json.loads(summary_path.read_text()) | |
| assert invalid.to_pylist() == before | |
| assert report.error_count == len(records) | |
| assert report.warning_count == 0 | |
| assert report.has_errors | |
| assert report.findings_truncated | |
| assert len(report.findings) == 3 | |
| assert len(streamed) == len(records) | |
| assert [item["row_index"] for item in streamed] == list(range(len(records))) | |
| assert summary["summary"] == {"errors": len(records), "warnings": 0} | |
| assert summary["findings_preview"] == { | |
| "retained": 3, | |
| "total": len(records), | |
| "truncated": True, | |
| } | |
| assert summary["findings_jsonl_path"] == str(findings_path.resolve()) | |
| assert len(summary["findings"]) == 3 | |
| def test_incremental_findings_publish_atomically_and_preserve_prior_on_abort( | |
| tmp_path: Path, | |
| ) -> None: | |
| findings_path = tmp_path / "quality" / "findings.jsonl" | |
| findings_path.parent.mkdir(parents=True) | |
| findings_path.write_text("prior-complete-evidence\n", encoding="utf-8") | |
| validator = IncrementalQualityValidator( | |
| "trades", | |
| findings_jsonl_path=findings_path, | |
| ) | |
| validator.close() | |
| assert findings_path.read_text(encoding="utf-8") == "prior-complete-evidence\n" | |
| assert not list(findings_path.parent.glob(f".{findings_path.name}.*.tmp")) | |
| def test_incremental_finish_is_idempotent_and_update_after_finish_fails() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=2, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=46, | |
| ) | |
| validator = IncrementalQualityValidator("trades", max_findings=0) | |
| validator.update(data.trades) | |
| first = validator.finish() | |
| assert validator.finish() is first | |
| with pytest.raises(RuntimeError, match="already closed"): | |
| validator.update(data.trades) | |
| def test_validate_batches_matches_unbounded_table_rule_semantics() -> None: | |
| data = generate_synthetic_market( | |
| symbols=("BTCUSDT",), | |
| events_per_symbol=3, | |
| start_ts_ns=1_704_153_600_000_000_000, | |
| seed=47, | |
| ) | |
| records = data.trades.to_pylist() | |
| records[1]["trade_id"] = records[0]["trade_id"] | |
| records[2]["quantity_lots"] = 0 | |
| records[2]["quantity"] = 0.0 | |
| invalid = table_from_records("trades", records) | |
| legacy = validate_table(invalid, "trades") | |
| incremental = validate_batches([invalid], "trades", max_findings=None) | |
| assert incremental.rows_checked == legacy.rows_checked | |
| assert incremental.error_count == legacy.error_count | |
| assert incremental.warning_count == legacy.warning_count | |
| assert incremental.findings == legacy.findings | |