release: v2.2.0

[TexasCoding]

Cuts v2.2.0 — the response-side spec drift hardening stack (#157).

Version bumps

• pyproject.toml: 2.1.0 → 2.2.0
• kalshi/__init__.py: 2.1.0 → 2.2.0

Semver: minor

All changes since v2.1.0 are additive or relaxing:

• 65 new optional fields on response models (no breaks to existing field shapes).
• le=32 constraint dropped on 6 subaccount request fields (only callers passing > 32 were affected, and those round-tripped to the server anyway).
• WS envelope/helper models gained extra="allow" (relaxes tolerance, never tightens).
• Test-infra warn→fail flip is internal-only — SDK users see no behavior change.

No API breaks. No migration required.

Release scope

Area	Change
REST response models	+38 fields across 13 models (Market, Order, Fill, Event, EventMetadata, Settlement, Trade, IncentiveProgram, RFQ, Quote, OrderGroup, GetOrderGroupResponse, CreateOrderGroupResponse)
WS payload models	+33 fields across 11 payloads (Unix-ms *_ts_ms timestamps, outcome_side/book_side direction encoding, MVE linkage, RFQ/Quote context echoes)
Drift detection	Additive drift now hard-fails CI; unmapped WS models hard-fail; required-but-optional stays warn-only (~204 entries, separate policy decision)
WS infrastructure	ErrorPayload registered in WS_CONTRACT_MAP; envelopes + helpers all use extra="allow"
Bugfix	Dropped le=32 cap on 6 subaccount request fields (demo allocates above 32; spec has no upper bound)

PRs in this release: #165 (PR0), #166 (PR1), #167 (PR2), #168 (PR3+4+5), #169 (#164 fix).

Docs touched

• CHANGELOG.md — finalized ## Unreleased → ## 2.2.0 — 2026-05-19 with a release-summary headline. The Added/Changed/Fixed entries written incrementally across feat(models): backfill v3.18.0 fields on Market / Order / Fill (PR1) #166–fix(models): relax subaccount-number request constraint (drop le=32) (#164) #169 stay as-is.
• ROADMAP.md — added v2.2.0 to Shipped; dropped the two "Next milestone" items now done by this release (response-side drift detection, WS envelope extra=allow); added two new follow-ups (REST sub-model mapping pass; required-but-optional drift policy).
• CLAUDE.md — active milestone → post-v2.2.

Verification

$ uv run python -c "import kalshi; print(kalshi.__version__)"
2.2.0

$ uv run pytest tests/ --ignore=tests/integration -q
2023 passed, 35 warnings in 127.76s

$ uv run pytest tests/integration/test_subaccounts.py::TestSubaccountsSync::test_transfer_between_subaccounts
1 passed in 2.30s          # against live demo, confirms #164 fix end-to-end

$ uv run mypy kalshi/
Success: no issues found in 76 source files

$ uv run ruff check .
All checks passed!

The 35 warnings are all test_required_drift / test_ws_required_drift — the required-but-optional axis stays warn-only per the issue stack's explicit out-of-scope note.

Release workflow

Once squash-merged, push tag v2.2.0 to trigger .github/workflows/release.yml:

1. Verify tag matches pyproject.toml version (gate).
2. Build sdist + wheel via uv build.
3. twine check the artifacts.
4. Publish to PyPI via trusted publishing (OIDC + sigstore attestations).
5. Create GitHub release with the ## 2.2.0 — ... CHANGELOG section as notes.

No manual PyPI upload needed.

[claude]

Code Review — v2.2.0 Release Cut

This is a clean, minimal release PR. The five changed files are all docs and version bumps; the actual code shipped in #165–#169. Review notes below are ordered by importance.

---

Correctness

Field-count discrepancy in the PR body (informational)

The PR body's summary table says +38 fields across 13 models (REST) + +33 fields across 11 payloads (WS) = 71 fields / 24 models, but the CHANGELOG and ROADMAP both land on "65 new optional fields across 16 models". The discrepancy is only in the release-notes prose, not in the code, so there is no functional impact — but one of the two counts is wrong. If the CHANGELOG/ROADMAP figure of 65/16 is canonical (it appears to be, since it was written more carefully as the release was assembled), the table numbers in the PR description should have been reconciled before merge. Worth a quick pass before this PR is used as source material for the GitHub release notes.

CHANGELOG date vs. merge date

The changelog is stamped 2026-05-19 but today is 2026-05-20. If merge happens today the date will be one day off. Minor, but release notes are permanent — consider updating to match the actual merge date.

---

Version bumps

Both pyproject.toml and kalshi/__init__.py are bumped from 2.1.0 → 2.2.0. They are consistent with each other and with the PR title. ✓

---

Semver claim

Claimed as minor. The PR body enumerates: new optional fields (additive), relaxed le=32 constraint (broadening), extra="allow" on WS models (relaxing), test-infra warn→fail flip (internal). All of these are either additive or relaxing — none are API breaks. The minor bump is correct. ✓

---

Documentation changes

CHANGELOG.md — Upgrading ## Unreleased to ## 2.2.0 — 2026-05-19 is the right pattern. The summary paragraph is clearer and more complete than the pre-PR stub. ✓

ROADMAP.md — The shipped v2.2.0 entry is well-summarised. Removing the two completed "Next milestone" items and promoting new follow-ups (REST sub-model mapping pass, required-but-optional drift policy) keeps the backlog honest. The reformatting of deferred items with bold labels improves scannability. ✓

CLAUDE.md — Milestone updated from post-v2.0 to post-v2.2. ✓

---

Release workflow

The post-merge tag instruction (push tag v2.2.0 to trigger release.yml) is correct. The CI gate that verifies the tag matches pyproject.toml provides a safety net against mismatched publishes. ✓

---

Summary

The PR is release-ready modulo two low-severity documentation notes:

1. Reconcile the 65/71 field-count discrepancy before the GitHub release notes are auto-generated from this PR.
2. Update the CHANGELOG date to 2026-05-20 if merging today, or merge it today and leave it — pick one and be consistent.

No code changes to review; logic shipped in #165–#169.