[ci-scan-feedback] Strengthen do-not-disable detection and cross-pipeline dedup

[github-actions]

Triggering signals

• (PR [ci-scan] Skip stackoverflowtester under interpreter mode (refs #127899) #128737, @janvorli: "I don't think we should disable the test without trying to understand the problem." @BrzVlad: "I'm also very aggressive with fixing these pipelines without disabling any tests.", [ci-scan] Skip stackoverflowtester under interpreter mode (refs #127899) #128737 (comment))
• (Issue [ci-scan] Build break: Crossgen2 OOM on HugeArray1.dll (windows x86 R2R_CG2) #128697 closed not_planned, duplicate of [ci-scan] Build break: Crossgen2 OOM compiling HugeArray1.dll on x86 R2R #128531 — same Crossgen2 OOM HugeArray1.dll signature filed under different pipeline definitions 108 vs 124, [ci-scan] Build break: Crossgen2 OOM on HugeArray1.dll (windows x86 R2R_CG2) #128697)

Proposed edits

• (.github/workflows/ci-failure-scan.md:Step 4.7, add 6 rejection phrases and a heuristic catch-all for MEMBER investigation intent — addresses PR [ci-scan] Skip stackoverflowtester under interpreter mode (refs #127899) #128737 rejection pattern)
• (.github/workflows/ci-failure-scan.md:Step 4.0, add cross-definition dedup check so same signature on different pipeline defs is not filed as separate KBEs — addresses [ci-scan] Build break: Crossgen2 OOM on HugeArray1.dll (windows x86 R2R_CG2) #128697 duplicate)

Expected behavior change

The next scanner run will skip test-disable PRs when any MEMBER has commented with investigation intent (e.g., "trying to understand", "I'm fixing", "without disabling") on the linked KBE, even when the exact legacy phrase list doesn't match. It will also avoid filing duplicate KBEs when the same failure signature appears across multiple pipeline definitions in the same run.

Generated by CI Outer-Loop Failure Scanner — Feedback · ● 20.6M · ◷

---

Note

This was originally intended as a pull request, but the git push operation failed.

Workflow Run: View run details and download patch artifact

The patch file is available in the agent artifact in the workflow run linked above.

To create a pull request with the changes:

# Download the artifact from the workflow run
gh run download 26703914774 -n agent -D /tmp/agent-26703914774

# Create a new branch
git checkout -b ci-scan-feedback/step4.7-phrases-and-crossdef-dedup-aa374e63740f11e5

# Apply the patch (--3way handles cross-repo patches where files may already exist)
git am --3way /tmp/agent-26703914774/aw-ci-scan-feedback-step4.7-phrases-and-crossdef-dedup.patch

# Push the branch to origin
git push origin ci-scan-feedback/step4.7-phrases-and-crossdef-dedup-aa374e63740f11e5

# Create the pull request
gh pr create --title '[ci-scan-feedback] Strengthen do-not-disable detection and cross-pipeline dedup' --base main --head ci-scan-feedback/step4.7-phrases-and-crossdef-dedup-aa374e63740f11e5 --repo dotnet/runtime

Show patch preview (48 of 48 lines)

From 29c8000e5ba18432fbf9118c4203b4750f777ce4 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]" <github-actions[bot]@users.noreply.github.com>
Date: Sun, 31 May 2026 05:17:11 +0000
Subject: [PATCH] ci-failure-scan: strengthen do-not-disable detection and
 cross-def dedup

- Step 4.7: Add 6 new rejection phrases ('should not disable', 'without
  disabling', 'trying to understand', 'I'm fixing', 'I will fix', 'looking
  into this', 'taking a look') plus a heuristic catch-all for MEMBER
  investigation intent.
- Step 4.0: Add cross-definition dedup check so the same signature on
  different pipeline definitions is not filed as separate KBEs.

Signals: PR #128737 closed by @janvorli/@BrzVlad (prefer investigation),
issue #128697 duplicate of #128531 (same OOM signature, different pipeline def).

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---
 .github/workflows/ci-failure-scan.md | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/ci-failure-scan.md b/.github/workflows/ci-failure-scan.md
index 02270d6b..395b6e71 100644
--- a/.github/workflows/ci-failure-scan.md
+++ b/.github/workflows/ci-failure-scan.md
@@ -224,6 +224,8 @@ For each `(definition_id, phase, queue, stress_mode, signature)` produced by Ste
 
 Cache filed signatures in `/tmp/gh-aw/agent/filed.tsv` as `<key>\t<aw_id>` where `key = <definition_id>|<queue>|<stress_mode>|<signature_norm>`. `<signature_norm>` is the signature with tab/newline/CR characters stripped — needed because raw signatures are copied verbatim from logs and may contain whitespace that would corrupt the TSV. On match, record `skipped: dup of filed-issue #aw_<id> earlier in this run` and stop. Append after every Branch A emission.
 
+Additionally, perform a **cross-definition dedup check**: search the cache for any row whose `<signature_norm>` matches the current signature regardless of `<definition_id>`. If found, the same failure shape is already covered by a KBE filed earlie
... (truncated)