v0.16.0 — backend v3.66.2 wire-validation alignment

[maltsev-dev]

@## Summary

SDK v0.16.0 (commit 681ce93) closes the v1/v2 no-reservation fallback in _route_track. Behavior change for callers that invoke track_llm() outside a paired /check scope.

What changed

• _route_track no-smid branch (src/nullrun/runtime.py:3036-3067) drops llm_call events instead of falling back to /track/batch. Backend v3.66.2 closed the v1/v2 no-reservation consume path with per-event type-aware wire validation: any llm_call event in a batch WITHOUT reservation_id is rejected with 503 BUDGET_RECHECK_FAILED (whole-batch fail-CLOSED). The 0.12.0 fallback was amplifying into 503-storm for every call site that forgot to pair track_llm with a prior gate.
• New metric metrics.runtime.dropped_llm_call_no_reservation (src/nullrun/observability/__init__.py:104-109) — operators alert on sustained rate to detect integration bugs (missing gate pairing around track_llm).
• WARNING log (not DEBUG) naming event_type + workflow_id, mirroring the 0.15.2 fail-OPEN observability fix.

Migration

Operations hitting dropped_llm_call_no_reservation > 0 on /health are calling track_llm() outside a paired /check scope. Wrap calls in one of:

• @protect(...) decorator (wraps in with workflow(...) + check_workflow_budget() automatically),
• check_workflow_budget() before track_llm() (explicit two-step),
• with workflow("wf-id"): context manager + check_workflow_budget() inside.

Bare track_llm() calls (no surrounding gate) silently drop the event post-0.16.0 — the call still returns {"allowed": True, ...}, but no cost_events row is written.

Tests

• 2 source-pin regression tests updated to pin the corrected drop behavior (TestRouteTrack.test_llm_call_without_smid_is_dropped + TestEndToEndCaptureFlow::test_block_response_does_not_infect_subsequent_track).
• 1 wire-payload test updated to use tool_call (avoids the drop-on-no-smid).
• 2 model-wire tests opt into NULLRUN_V3_TRACK_DISABLE=1 to keep the focus on the missing-model flag-and-log surface.
• 1571 passed, 7 skipped in 23.89s (matches 0.15.2 baseline). Zero regressions. ruff clean.

Compatibility

• No SDK_MIN_VERSION bump. Backend v3.66.2 ships since 2026-08-18 (commit e262f1c3).
• Wire-format unchanged. No public API change.
• Drop-in replacement for 0.15.2 for callers that always pair track_llm with a prior gate — those observe zero behavior change.

Files

 CHANGELOG.md | +32
pyproject.toml | +1 -1
src/nullrun/observability/__init__.py | +9
src/nullrun/runtime.py | +35 -12
tests/contract/test_llm_call_model_wire.py | +18 -4
tests/test_runtime.py | +9 -4
tests/test_v3_wire_contract.py | +47 -8
uv.lock | +1 -1

Note on commit history

PR was rebased onto master (HEAD 4b6ce56, 0.15.2 squash-merge) — the previous commits on the source branch (35728b9, 1c96654, 9b87d20, 127b003, 0f86c8c, 46c5325) are already in master via the 0.15.2 squash-merge, so the v0.16.0 commit is the only new commit on top of master. The original commit e09813c (pre-rebase) is superseded by 681ce93 (post-rebase) with identical tree content.
@

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!