Aggregate UI E2E coverage for sharded runs

• Status: Accepted
• Date: 2026-01-23
• Context:
  • What problem are we solving?
    • Playwright sharding runs global teardown per shard, causing partial coverage checks to fail.
  • What constraints or forces shape the decision?
    • Keep Playwright invoked via just ui-e2e, avoid new dependencies, and preserve coverage gating.
• Decision:
  • Summary of the choice made.
    • Skip coverage assertions in sharded teardown, write shard-specific coverage files, upload them as artifacts, and run an aggregate coverage check in a dedicated job.
  • Alternatives considered.
    • Disable coverage checks entirely for sharded runs (reduces signal).
    • Keep non-sharded UI E2E only (slower feedback).
• Consequences:
  • Positive outcomes.
    • Sharded UI E2E runs succeed while retaining full coverage enforcement.
  • Risks or trade-offs.
    • Additional workflow job and artifact handling.
• Follow-up:
  • Implementation tasks.
    • Monitor shard duration and artifact sizes.
  • Review checkpoints.
    • Revisit shard count if coverage aggregation becomes slow.

Task record

• Motivation: Fix sharded UI E2E failures while maintaining coverage enforcement.
• Design notes: Shard-specific coverage files with an aggregate coverage check via just ui-e2e-coverage.
• Test coverage summary: just ci, just ui-e2e.
• Observability updates: None (workflow-only change).
• Risk & rollback plan: Revert sharding and coverage aggregation changes if instability persists.
• Dependency rationale: No new dependencies introduced.