# Release-artifact waivers (#356) — read by scripts/check_release_artifacts.sh.
#
# The release gate fails when a version-stamped derived artifact (the benchmark
# baseline or the blind-eval snapshot) isn't stamped for the release being cut.
# If an artifact genuinely can't be refreshed for a release (e.g. a CI/docs-only
# release with no perf or tool-surface change), record a waiver here instead of
# refreshing — so the skip is deliberate and tracked, never silent.
#
# Format (one per waived artifact):
#   v<version> <benchmark|eval> #<issue> <reason>
# Example:
#   v0.10.2 benchmark #999 CI-only release; no perf-relevant change
#
# Entries are per-release; prune old ones freely. See
# docs/guides/RELEASE_ARTIFACTS.md.

v0.11.0 benchmark #450 CANNOT be captured: #437 made bulk mutations cost ~16s per RFC Message-ID (measured 1 id=15.3s, 10 ids=167s), and the bulk benchmarks use 50 ids x 5 runs, so the suite cannot finish. Refreshing is blocked on #450, not skipped for convenience. NOTE: this release changed perf substantially (#415/#419/#432/#434/#437), so the shipped baseline is knowingly stale and v0.11.1 must not trust it for comparison.
v0.11.0 eval #373 OPENROUTER_API_KEY unavailable on this runner (same cause as v0.10.2). Staler than last release: the tool surface grew 24 -> 25 (get_statistics, #378) and get_thread gained partial/partial_reason fields (#420), so scores are NOT expected to reproduce unchanged. Re-run tracked in #373.

v0.10.2 eval #373 additive tool-surface change (still 24 tools: added optional account/mailbox to save_attachments, content_truncated to get_messages, deprecated gmail_mode); OPENROUTER_API_KEY unavailable on the release runner, scores expected to reproduce, re-run tracked. (Benchmark baseline IS refreshed this release — #364 changed move_messages timing.)
