RFC 039: Pandas-familiar exploration API

[dannymeijer]

Use this issue to track InQL RFC 039, proposed in PR #60 at docs/rfcs/039_pandas_familiar_exploration_api.md.

Area

• Specification (RFCs)
• Documentation

Summary

RFC 039 defines a pandas-familiar exploration API for InQL while keeping Prism, Substrait IR, and adapters as InQL's semantic path.

Motivation

Many users expect familiar dataframe-style exploration affordances. InQL can support useful pandas-like spellings where they map cleanly to declarative query semantics, while rejecting row indexing, mutation, eager Series behavior, and other pandas semantics that do not fit InQL's model.

Proposal sketch

• Support dictionary-like column access such as data["column"].
• Support projection lists such as data[["a", "b"]].
• Support predicate filtering such as data[predicate].
• Add familiar aliases such as where, assign, groupby, sort_values, and head where semantics are explicit.
• Keep unsupported pandas behavior rejected rather than approximated.

Alternatives considered

• Keep only InQL-native query APIs.
• Attempt full pandas compatibility.
• Add .loc / .iloc and Series-like semantics in the initial scope.

Impact / compatibility

This should be additive. The API must remain a frontend over InQL planning and should not bypass Prism or make pandas semantics authoritative.

Implementation notes (optional)

Review PR #60 for the draft. Existing exploratory spike code remains separate from the RFC text.

Checklist

• I checked for an existing RFC or issue covering this.
• I can describe how this impacts existing code and how to migrate (if needed).