fix(data): implement market data layer; stop .gitignore swallowing src/data (#1)

[bradsmithmba]

Summary

src/data/ was missing __init__.py, providers.py, models.py, cache.py, and schema.py. Any import of FeatureEngineering — and therefore the entire feature-engineering and regime-detection pipeline — failed at runtime with ImportError.

Closes #1.

Root cause: a .gitignore footgun

This was not simply unwritten code. .gitignore line 165 had an unanchored rule:

data/

An unanchored directory pattern matches a directory of that name at every level, so it matched src/data/ as well as the intended root-level runtime cache directory (config.cache.cache_db_path = data/cache.db). Any module added under src/data/ was silently ignored by git and never committed.

Evidence it was written and then lost:

• tests/test_providers.py and tests/test_cache.py already exist and specify a detailed API for YFinanceProvider, RateLimiter, CacheManager, and the ORM schema.
• requirements.txt already declares yfinance>=0.2.28 and sqlalchemy>=2.0.0.
• The three older modules (data_utils.py, regime_labeler.py, training_data.py) predate the rule, so they remained tracked — which is why the directory looked partially populated.

The fix anchors the rule to the repository root:

/data/

This still ignores the runtime cache (data/cache.db) but no longer touches src/data/. Without this change, any future module added under src/data/ would silently vanish again.

Changes

File	Contents
.gitignore	data/ → /data/ (anchor to repo root)
src/data/models.py	PriceData, OptionChainData dataclasses
src/data/providers.py	YFinanceProvider, RateLimiter, DataProviderError/InvalidSymbol/DataNotAvailable, MarketDataRequest, get_default_provider()
src/data/schema.py	SQLAlchemy ORM: PriceHistory, OptionsData, CacheMetadata
src/data/cache.py	SQLite-backed CacheManager with TTL freshness, hit/miss stats, and cleanup
src/data/__init__.py	Package exports consumed by src/features/base.py

The implementation was written to the contract defined by the pre-existing test files, not invented — the tests are the spec.

Testing

$ pytest tests/test_providers.py tests/test_cache.py -q
21 passed

The exact reproduction from the issue now imports cleanly:

from src.features.base import FeatureEngineering          # OK
from src.data import get_default_provider, CacheManager   # OK
from src.data.regime_labeler import RegimeType            # OK

Broader run across tests/{test_providers,test_cache}.py, tests/features/, and tests/models/ (excluding one unrelated missing module): 262 passed. Almost none of these collected before this change.

Related findings (separate issues, not addressed here)

• The same unanchored-pattern bug also affects models/ (line 168), which silently ignores src/models/. Filed separately.
• src/models/integrated_selector.py is genuinely absent and breaks recommendation_engine import. Filed separately.
• transfer_trainer has a torch API drift (ReduceLROnPlateau constructor arg). Filed separately.

🤖 Generated with Claude Code