quantskills/agent-quantspace/store

Store

skills/store owns reusable file storage for market data and generic research artifacts. The public project uses explicit strategy universes and does not maintain a central universe registry.

DataManager

from skills.store.data_manager import DataManager, DataQualityReport, validate_ohlcv
from skills.store.workspace import resolve_workspace_paths

The data root comes from QUANTSPACE_DATA_ROOT, otherwise the repository data/ directory.

resolve_workspace_paths() centralizes QUANTSPACE_WORKSPACE_ROOT, QUANTSPACE_DATA_ROOT, and QUANTSPACE_REPORTS_ROOT without creating any directories. Runtime entrypoints use it instead of modifying sys.path or independently guessing the repository root.

Supported layout:

data/market/<frequency>/<symbol>.parquet
data/adj_factor/<symbol>.parquet
data/factors/<namespace>/
data/correlation/
data/backtest/
data/models/
data/export/

The subdirectory below data/market/ is a storage directory/data-set key. Its base part is the real bar frequency, while an optional _adj suffix records the price-adjustment state:

Therefore data/market/1d_adj/ does not represent a frequency named 1d_adj; its freq is still 1d, and 1d_adj is only the directory name indicating adjusted 1d data. The same convention applies to other bar intervals: <freq>_adj stores adjusted bars whose actual frequency is <freq>.

DataManager names its directory-selector parameter frequency. Pass the full directory key to read_symbol / read_symbols / save_symbol (for example, frequency="1d_adj") so it resolves the intended path. Do not reuse that suffixed value as the semantic freq or pass it to an upstream market-data API; use freq="1d" there.

Main methods:

• read_symbol, read_symbols, save_symbol
• import_symbol_csv, import_combined_csv, list_symbols
• save_factor, read_factor, factor_namespace_dir, factor_filename
• save_factor_correlation, read_factor_correlation
• save_backtest_run, read_backtest_summary, read_backtest_run
• list_models, read_model_metadata

read_symbols returns a MultiIndex (symbol, eob) panel and reports every missing symbol in one FileNotFoundError.

DataManagerArtifactStore (skills.factor_mining.adapters.store) persists factor-mining research artifacts under data/factors/<namespace>/artifacts/<kind>/<artifact_id>.json (path segments are validated; resolved paths must stay under that namespace). It reuses save_factor / factor_filename for explicit wide-pivot caches keyed by a content-addressed cache_key. DataManager.namespaced_artifact_dir(namespace, *parts) is the generic helper that returns (and creates) data/factors/<namespace>/<parts...>/; there is no second catalog.

from skills.store.data_manager import DataManager
panel = DataManager().read_symbols(
    ["SHSE.510300", "SHSE.510500"],
    frequency="1d_adj",
)

Boundary Rules

• Keep market and generic research files here.
• Pass strategy universes as explicit symbol lists owned by the caller.
• Treat factor/backtest/model subdirectory names as artifact namespaces, not

centrally managed instrument universes.

• Do not add strategy identity schemas or a second experiment catalog.

CLI command reference

<!-- BEGIN GENERATED: qs commands -->

Transport: --input-json FILE|-, --preview N, and --log-level LEVEL; flags override JSON. Use --print-template [FIELD] for fillable templates.

Shared parameter groups

OutputOptions

uv run qs store info

Inspect the local data store inventory.

Priority: P0. Side effects: read-only.

Output data fields: data_root, exists, frequencies, adj_factors, namespaces.

Inspect schemas: uv run qs schema store.info (add --part input|output or --file-field FIELD).

uv run qs store list

List stored market symbols, adjustment factors, or models.

Priority: P0. Side effects: read-only.

Output data fields: kind, items, count, truncated, next_offset.

Inspect schemas: uv run qs schema store.list (add --part input|output or --file-field FIELD).

uv run qs store read

Read stored symbols into a (symbol, eob) panel.

Priority: P0. Side effects: writes-output.

Shared parameters: OutputOptions.

Output data fields: symbols, frequency, rows, start, end, columns, missing, preview.

Inspect schemas: uv run qs schema store.read (add --part input|output or --file-field FIELD).

uv run qs store import

Import market bars from CSV into data/market.

Priority: P1. Side effects: writes-store.

Output data fields: symbols, rows, saved, dry_run, frequency, planned_paths, existing.

Inspect schemas: uv run qs schema store.import (add --part input|output or --file-field FIELD).

uv run qs store validate

Validate OHLCV quality; exit 1 when a check fails.

Priority: P1. Side effects: read-only.

Output data fields: passed, n_rows, nan_count, inf_count, negative_price_count, negative_volume_count, high_lt_low_count, non_monotonic_index, details.

Inspect schemas: uv run qs schema store.validate (add --part input|output or --file-field FIELD).

uv run qs store check-adjusted

Scan adjusted bars for zero-volume price jumps; exit 1 when found.

Priority: P1. Side effects: writes-output.

Shared parameters: OutputOptions.

Output data fields: symbols, missing, issues, preview.

Inspect schemas: uv run qs schema store.check-adjusted (add --part input|output or --file-field FIELD).

uv run qs store query

Run DuckDB SQL over stored Parquet files (needs --extra query).

Priority: P2. Side effects: writes-output.

Shared parameters: OutputOptions.

Output data fields: rows, columns, preview.

Inspect schemas: uv run qs schema store.query (add --part input|output or --file-field FIELD).

<!-- END GENERATED: qs commands -->