[efegokdemir]

What changed

Expose Arc's truncated, truncation_reason, rows_capped and row_cap response fields on QueryResponse, and document how to distinguish failed/truncated results from results limited by policy.

Older servers remain compatible: missing fields default to False or None.

Testing

• uv run pytest tests/unit -q --tb=short (103 passed)
• uv run ruff check .
• uv run ruff format --check .
• uv run mypy src/arc_client --ignore-missing-imports

The regression test failed before adding the model fields and passes with this change.

Related: Basekick-Labs/arc#726

[jallegri]

Reviewed against Arc#726: the model exposes both signals with backward-compatible defaults, preserves the distinction between failed/truncated results and valid policy-capped results, and QueryClient.query() validates the decoded response through this model. I found no code-blocking issues. The PR description reports 103 passing unit tests plus Ruff and mypy. The current GitHub CI run is action_required with no jobs, so the CI result is still outstanding before merge; this review does not treat the author-reported local checks as CI evidence.

[jallegri]

Residual transport work tracked in #9; overall parity analysis: https://github.com/Basekick-Labs/ops/issues/40.

This PR covers JSON/model fields. PR #2's MessagePack decoder reconstructs QueryResponse without rows_capped/row_cap, although both pinned server targets emit them. Combining the model changes alone therefore does not preserve MessagePack cap signals. Please test sync/async public decoding paths and assess Arrow trailer visibility separately. Reuse this PR/#2 where appropriate; #9 tracks the unimplemented residual.

[xe-nvdk]

Closing as superseded: truncated, truncation_reason, rows_capped and row_cap are on main via #25 (src/arc_client/models/query.py), together with the Arrow completion metadata. Thanks for the groundwork, Efe.