From e71eae4b35a7bc707b9412b1d4515191539546df Mon Sep 17 00:00:00 2001
From: Dexterity104 <hatanokanjiro@gmail.com>
Date: Fri, 26 Jun 2026 14:01:08 +0000
Subject: [PATCH] fix(scoring): anchor the draft title pattern to genuine
 markers

---
 src/scoring/pending-pr-scenarios.ts           |  7 +++--
 src/signals/contributor-open-pr-monitor.ts    |  3 +-
 test/unit/contributor-open-pr-monitor.test.ts | 20 +++++++++++++
 test/unit/pending-pr-scenarios.test.ts        | 29 +++++++++++++++++++
 4 files changed, 56 insertions(+), 3 deletions(-)

diff --git a/src/scoring/pending-pr-scenarios.ts b/src/scoring/pending-pr-scenarios.ts
index 67368cf7cd..358f037f32 100644
--- a/src/scoring/pending-pr-scenarios.ts
+++ b/src/scoring/pending-pr-scenarios.ts
@@ -37,6 +37,10 @@ export type ContributorRepoOpenPrSignals = {
 
 const STALE_DAYS = 14;
 
+// Real draft markers only — "[draft]", "Draft:", "Draft -"; the delimiter keeps "Drafting" and
+// "draft-js" from matching. Trailing \s* lets the same pattern also strip the marker for dedup keys.
+export const DRAFT_TITLE_PATTERN = /^(?:\[\s*draft\s*\]|draft(?:\s*:|\s+-))\s*/i;
+
 export async function loadContributorRepoOpenPrSignalRecords(
   env: Env,
   repoFullName: string,
@@ -221,8 +225,7 @@ export function applyPendingPrDetectionToScoreInput(
 
 function isDraftPullRequest(pr: PullRequestRecord): boolean {
   if (pr.isDraft) return true;
-  const title = pr.title.trim();
-  if (/^\[?\s*draft\s*\]?/i.test(title) || /^draft:/i.test(title)) return true;
+  if (DRAFT_TITLE_PATTERN.test(pr.title.trim())) return true;
   return pr.labels.some((label) => label.toLowerCase() === "draft" || label.toLowerCase() === "wip");
 }
 
diff --git a/src/signals/contributor-open-pr-monitor.ts b/src/signals/contributor-open-pr-monitor.ts
index 30769503b2..8989440f5d 100644
--- a/src/signals/contributor-open-pr-monitor.ts
+++ b/src/signals/contributor-open-pr-monitor.ts
@@ -3,6 +3,7 @@ import { sanitizePublicComment } from "../github/commands";
 import {
   classifyOpenPullRequest,
   detectPendingPrScenario,
+  DRAFT_TITLE_PATTERN,
   loadContributorRepoOpenPrSignals,
   type ClassifiedOpenPullRequest,
   type PendingPrScenarioDetection,
@@ -243,7 +244,7 @@ function duplicatePronePullNumbers(openPullRequests: PullRequestRecord[]): Set<n
 function normalizeTitle(title: string): string {
   return title
     .toLowerCase()
-    .replace(/^\[?\s*draft\s*\]?\s*/i, "")
+    .replace(DRAFT_TITLE_PATTERN, "")
     .replace(/^wip:\s*/i, "")
     .replace(/[^a-z0-9]+/g, " ")
     .trim();
diff --git a/test/unit/contributor-open-pr-monitor.test.ts b/test/unit/contributor-open-pr-monitor.test.ts
index 6edde26836..8725814c8b 100644
--- a/test/unit/contributor-open-pr-monitor.test.ts
+++ b/test/unit/contributor-open-pr-monitor.test.ts
@@ -220,6 +220,26 @@ describe("contributor open PR monitor", () => {
     expect(__contributorOpenPrMonitorInternals.duplicatePronePullNumbers([pr({ number: 42, labels: ["wip"] })]).has(42)).toBe(true);
   });
 
+  it("strips only genuine draft markers when normalizing titles for dedup (regression)", () => {
+    const { duplicatePronePullNumbers } = __contributorOpenPrMonitorInternals;
+
+    // "[draft] X" and "X" are the same work, so stripping the marker collapses them into one cluster.
+    const clustered = duplicatePronePullNumbers([
+      pr({ number: 60, title: "[draft] fix parser bug" }),
+      pr({ number: 61, title: "fix parser bug" }),
+    ]);
+    expect(clustered.has(60)).toBe(true);
+    expect(clustered.has(61)).toBe(true);
+
+    // A title that merely starts with the word "draft" keeps it, so it is not mistaken for "tooling".
+    const distinct = duplicatePronePullNumbers([
+      pr({ number: 62, title: "Draft tooling" }),
+      pr({ number: 63, title: "tooling" }),
+    ]);
+    expect(distinct.has(62)).toBe(false);
+    expect(distinct.has(63)).toBe(false);
+  });
+
   it("covers monitor summaries, guidance, next steps, and file heuristics", () => {
     const { nextStepsForClassification, summarizeMonitor, buildMonitorGuidance, missingTestsFromFiles, priorityRank } =
       __contributorOpenPrMonitorInternals;
diff --git a/test/unit/pending-pr-scenarios.test.ts b/test/unit/pending-pr-scenarios.test.ts
index ea8248034d..5aea1593de 100644
--- a/test/unit/pending-pr-scenarios.test.ts
+++ b/test/unit/pending-pr-scenarios.test.ts
@@ -302,6 +302,35 @@ describe("pending PR scenario detection", () => {
     expect(withOverlapFlags.reasons.join(" ")).toMatch(/duplicate|test files/i);
   });
 
+  it("does not treat titles that merely start with the letters 'draft' as drafts (regression)", () => {
+    // These all begin with "draft" but carry no real marker, so they must flow through to merge_ready
+    // instead of being dropped — otherwise pendingMergedPrCount is silently understated. The hyphenated
+    // names ("draft-js", "draft-mode") are the cases an unspaced `draft[-:]` boundary would mis-flag.
+    for (const title of ["Drafting a new feature", "Draftsman tool", "Drafted changes", "draft-js upgrade", "Draft-mode rendering rewrite"]) {
+      expect(
+        classifyOpenPullRequest({
+          pr: pr({ number: 60, title }),
+          roleContext: outsideContributorRole,
+          reviews: [approvedReview(60)],
+          checks: [],
+        }).classification,
+      ).toBe("merge_ready");
+    }
+  });
+
+  it("treats only genuine draft markers in the title as drafts", () => {
+    for (const title of ["[draft] spike", "[ draft ] spike", "Draft: spike", "draft : spike", "Draft - spike", "Draft -spike"]) {
+      expect(
+        classifyOpenPullRequest({
+          pr: pr({ number: 61, title }),
+          roleContext: outsideContributorRole,
+          reviews: [approvedReview(61)],
+          checks: [],
+        }).classification,
+      ).toBe("draft");
+    }
+  });
+
   it("recognizes draft heuristics and excludes pull numbers from detection", () => {
     expect(
       classifyOpenPullRequest({