docs(odfv): document column naming behavior when using multiple FeatureView sources

[anix-lynch]

Problem

The ODFV docs only show examples with a single FeatureView + RequestSource combination, where columns are accessed by feature name directly (e.g., features_df["conv_rate"]).

What's not documented: when multiple FeatureViews are passed as sources, Feast prefixes each column with {feature_view_name}__ to avoid naming conflicts. This causes a silent KeyError when users try features_df["conv_rate"] expecting the behavior shown in the docs.

Real-world failure

Building an ODFV with two FeatureView sources:

@on_demand_feature_view(
    sources=[customer_transaction_features, customer_profile_features],
    schema=[Field(name="risk_recency_score", dtype=Float64)],
)
def risk_recency_score(df: pd.DataFrame) -> pd.DataFrame:
    recency = df["recency_days"]  # KeyError — not found

Actual column name at runtime: customer_transaction_features__recency_days

Nothing in the docs explains this prefix behavior.

Fix

Added a Column Naming in the Input DataFrame section with:

• Table showing column name format by source type
• Defensive access pattern for production code

Changes

• docs/reference/beta-on-demand-feature-view.md — new section before CLI Commands

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.