feat(metrics): exclude out-of-scope runs from KPIs via (skip-metrics) marker

[ignaciosantise]

Why

The rn row on the E2E KPI dashboard hit P95 PR feedback of 62m (vs ~30m usual): one hung Android E2E job on a draft PR ran to its 60m timeout and, with only n=8 PR runs in the window, dominated the P95. More broadly, work-in-progress and ad-hoc runs shouldn't count toward KPIs at all:

• Draft-PR runs — WIP, not real feedback.
• Manual (workflow_dispatch) runs off the default branch — ad-hoc dev testing; the SDK repos have no branch filter, so these land in pass/flake today.
• Scheduled / default-branch runs — these should count and are unaffected.

How

Workflows stamp a (skip-metrics) marker into their run-name for runs that are out of scope (draft PR, or workflow_dispatch off the default branch). It surfaces on the run as display_title. This PR filters those out centrally in fetch_runs, so every KPI script inherits it.

The exclusion decision lives in each workflow — where draft/branch context is free — which keeps the metric side a plain string match with no extra API calls.

Correctness notes

• Pagination preserved: filtering happens after the raw page size is counted, so a page full of skipped runs won't stop paging early.
• Non-marked runs untouched: schedule/push/ready-PR/default-branch runs never carry the marker.
• Opt-in per repo: only repos that emit the marker are affected.

Rollout

Requires each workflow to emit the marker. In progress:

• reown-com/react-native-examples (merged the draft-skip; marker swap to (skip-metrics) follow-up pending)
• ci(e2e-pay): skip E2E on draft PRs; keep draft & off-default manual runs out of KPIs reown-com/reown-kotlin#411
• ci(e2e-pay): skip E2E on draft PRs; keep draft & off-default manual runs out of KPIs reown-com/reown-swift#325
• ci(e2e-pay): skip E2E on draft PRs; add balance-check job; keep off-KPI runs out of metrics reown-com/reown_flutter#392

Until a repo emits the marker, its behavior is unchanged.

🤖 Generated with Claude Code

[claude]

Claude finished @ignaciosantise's task in 2m 0s —— View job

---

Review complete

• Read changed file
• Analyze diff
• Post findings

---

✅ No issues found

The change is correct and well-structured. A few observations worth confirming:

• Pagination is sound: count is computed from the raw $response (line 52) before the jq filter emits output (line 53), so the < 100 early-exit check keys off the full page size — drafts can't mask a full page and cause premature stop.
• Null safety: (.display_title // "") correctly passes non-PR runs (schedule/push) through untouched, since their display_title won't carry the marker.
• Empty-response guard: the existing [[ -z "$response" ]] break (line 42) prevents the filter jq from receiving a blank line, so no parse error risk.
• Convention dependency: filtering relies on workflows consistently emitting (Draft) (exact casing/spacing) in run-name. Non-conforming repos are unaffected, which matches the opt-in rollout described in the PR.

[Copilot]

Pull request overview

This PR updates the shared metrics helper in metrics/scripts/lib.sh so KPI scripts exclude workflow runs that are explicitly marked as out-of-scope via a (skip-metrics) marker in the run name (surfaced by the API as display_title). This centralizes KPI scoping without requiring additional API calls.

Changes:

• Filter out runs whose .display_title contains (skip-metrics) in fetch_runs.
• Preserve pagination behavior by computing the raw page size (count) before applying the filter.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.