Reproducible research

fincore's enhanced layer makes external-data analysis reproducible: a fetched frame is wrapped in a DataSnapshot that freezes its source, request interval, as-of timestamp, price-adjustment convention, and a SHA256 of the data — without recording secret configuration.

import pandas as pd

from fincore.data.snapshots import DataSnapshot

# -- snapshot
snapshot = DataSnapshot.from_frame(
    frame=pd.DataFrame({"close": [10.0]}),
    provider="fixture",
    requested_start="2024-01-01",
    requested_end="2024-01-02",
    as_of="2024-01-03T00:00:00Z",
)
# -- snapshot

manifest = snapshot.to_manifest()
print(manifest["content_sha256"])  # 64-hex SHA256 of the data

The manifest carries provenance only — no API keys, tokens, raw returns, or absolute local paths. Reports add a second layer: with return_result=True and audit_manifest=True, create_strategy_report writes a sidecar JSON recording the code commit, dependency versions, per-input shapes and hashes, and the sanitized resolved structured performance disclosure (calculation convention, units, frequency, sample period, data quality, fee/cashflow treatment, benchmark, risk-free convention and annualization). It still does not copy raw input values into the manifest; free-form report HTML is rendered separately (see Risk validation and the report API).

Provider access is provider_required: inject a client for offline tests, and a broken optional SDK surfaces as a controlled DependencyError that names the required extra. See Core Concepts for the provider contract.