feat: PR filter — compare models on the same set of PRs

UI: a repo#pr multi-select (labeled with how many models ran each PR)
scopes the whole table — runs, minutes, findings, points — to the chosen
PRs, so a model with 2 runs can be fairly compared against one with 60.
API: GET /scoreboard accepts ?repo= and ?pr= (repeatable or comma-list).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

store_test.go

@@ -46,7 +46,7 @@ func TestConsensusAndGrade(t *testing.T) {
 		t.Fatal(err)
 	}
 
-	board, err := s.Scoreboard()
+	board, err := s.Scoreboard(ScoreboardFilter{})
 	if err != nil {
 		t.Fatal(err)
 	}
@@ -88,7 +88,7 @@ func TestLatestGradeWins(t *testing.T) {
 	if err := s.AddGrade(Grade{FindingID: id, IsReal: false}); err != nil { // re-graded as a false positive
 		t.Fatal(err)
 	}
-	board, _ := s.Scoreboard()
+	board, _ := s.Scoreboard(ScoreboardFilter{})
 	m := board[0]
 	if m.Confirmed != 0 || m.FalsePositive != 1 || m.BySeverity["critical"] != 0 {
 		t.Errorf("after re-grade: confirmed=%d fp=%d critical=%d, want 0/1/0", m.Confirmed, m.FalsePositive, m.BySeverity["critical"])
@@ -116,6 +116,58 @@ func TestGradeValidation(t *testing.T) {
 	}
 }
 
+// TestScoreboardFilter: narrowing to repo/PRs drops runs and findings outside
+// the selection, so a model with many extra runs is compared on the same work.
+func TestScoreboardFilter(t *testing.T) {
+	s := testStore(t)
+
+	// fable reviewed only PR 1; veteran reviewed PRs 1 and 2 (and another repo).
+	for _, r := range []Run{
+		{RunID: "f-1", Repo: "steve/x", PR: 1, Model: "fable", Provider: "p", DurationSecs: 60},
+		{RunID: "v-1", Repo: "steve/x", PR: 1, Model: "veteran", Provider: "p", DurationSecs: 120},
+		{RunID: "v-2", Repo: "steve/x", PR: 2, Model: "veteran", Provider: "p", DurationSecs: 120},
+		{RunID: "v-3", Repo: "steve/y", PR: 1, Model: "veteran", Provider: "p", DurationSecs: 120},
+	} {
+		if err := s.AddRun(r); err != nil {
+			t.Fatal(err)
+		}
+	}
+	ids, err := s.AddReports([]ReportIn{
+		{Repo: "steve/x", PR: 1, Lens: "security", File: "a.go", Line: 1, Title: "shared", Model: "fable", Provider: "p", RunID: "f-1"},
+		{Repo: "steve/x", PR: 1, Lens: "security", File: "a.go", Line: 1, Title: "shared", Model: "veteran", Provider: "p", RunID: "v-1"},
+		{Repo: "steve/x", PR: 2, Lens: "security", File: "b.go", Line: 2, Title: "pr2 only", Model: "veteran", Provider: "p", RunID: "v-2"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err := s.AddGrade(Grade{FindingID: ids[0], IsReal: true, Severity: "high"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := s.AddGrade(Grade{FindingID: ids[2], IsReal: true, Severity: "critical"}); err != nil {
+		t.Fatal(err)
+	}
+
+	board, err := s.Scoreboard(ScoreboardFilter{Repo: "steve/x", PRs: []int{1}})
+	if err != nil {
+		t.Fatal(err)
+	}
+	byModel := map[string]ModelStat{}
+	for _, m := range board {
+		byModel[m.Model] = m
+	}
+	v := byModel["veteran"]
+	if v.Runs != 1 || v.Minutes != 2 {
+		t.Errorf("veteran runs=%d minutes=%v, want 1 run / 2 min (PR 2 and steve/y excluded)", v.Runs, v.Minutes)
+	}
+	if v.Findings != 1 || v.Confirmed != 1 || v.BySeverity["critical"] != 0 || v.BySeverity["high"] != 1 {
+		t.Errorf("veteran findings=%d confirmed=%d by_severity=%v, want only the PR-1 finding", v.Findings, v.Confirmed, v.BySeverity)
+	}
+	fbl := byModel["fable"]
+	if fbl.Runs != 1 || fbl.Findings != 1 || fbl.Confirmed != 1 {
+		t.Errorf("fable runs=%d findings=%d confirmed=%d, want 1/1/1", fbl.Runs, fbl.Findings, fbl.Confirmed)
+	}
+}
+
 // TestFindingIDLocationKeyed: id depends on location, not wording; line matters.
 func TestFindingIDLocationKeyed(t *testing.T) {
 	a := findingID("r", 1, "security", "a.go", 10)