ObservedHours states what its zero-coverage fallback defends and pins it: the nominal window, never a near-zero divisor, moot behind the collectors' unobserved guard (#3538 A7 follow-through from #3616 review)

Lite.Tests/FactScorerTests.cs

@@ -1627,6 +1627,20 @@ public void Score_PerHourGates_ReadAMissingPeriodAsOneHour()
         Assert.Equal(0.0, BaseOf(WaitFact("PAGELATCH_UP", 0.01, new() { ["wait_time_ms"] = 10_000 })), precision: 4);
     }
 
+    // A coverage_fraction of exactly 0 (WindowCoverage.Unobserved's stamp) reads as the NOMINAL window, not
+    // as a near-zero divisor: no shipped collector emits a wait fact for an unobserved window, so the branch
+    // only defends 0/0 — and if a fact like that ever arrived, dividing by ~0 would fire on the artifact.
+    // 12,000 ms over a nominal 4 h is 3 s/hr → 0, the pre-#3538 reading; pinned so the choice cannot flip.
+    [Fact]
+    public void Score_PerHourGates_ReadZeroCoverageAsTheNominalWindow_NotAsNearZero()
+    {
+        var fact = WaitFact("PAGELATCH_UP", 0.01, new()
+        {
+            ["wait_time_ms"] = 12_000, ["period_duration_ms"] = FourHoursMs, ["coverage_fraction"] = 0.0
+        });
+        Assert.Equal(0.0, BaseOf(fact), precision: 6);
+    }
+
     // Lineage census (source-text pin): every entry in GetWaitThresholds is preceded by a comment that
     // names either a measured figure (a percentile, a max, "measured") or says "unmeasured". A new entry
     // added without its lineage — the pre-#3538 shape of the whole table — fails here. Entries that share

PerformanceMonitor.Analysis/FactScorer.cs

@@ -193,6 +193,14 @@ private static double ScoreWaitFact(Fact fact)
     /// stamps the period, so only a hand-built fact lacks it; reading it as one hour makes the per-hour
     /// bars read as plain totals (10 s of PAGELATCH_UP, 15 min of THREADPOOL), which is the one reading a
     /// fact with no window can honestly have.</para>
+    ///
+    /// <para>A <c>coverage_fraction</c> of exactly 0 (what <c>WindowCoverage.Unobserved</c> stamps) falls back
+    /// to the NOMINAL window rather than to a near-zero divisor. That branch defends the arithmetic against
+    /// 0/0 and should be moot: both collectors emit no wait fact at all when <c>ObservedDurationMs &lt;= 0</c>,
+    /// so a fact carrying zero coverage and a non-zero wait total is a fact no shipped collector produces. If
+    /// one ever did, dividing by ~0 would turn any trace into an enormous hourly rate and fire on the
+    /// artifact; the nominal reading is the pre-#3538 form — the least-sensitive honest one — and is pinned
+    /// so the choice cannot silently flip.</para>
     /// </summary>
     private static double ObservedHours(Fact fact)
     {