diff --git a/AGENTS.md b/AGENTS.md
index 33ebb4c0..d95bb268 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -141,10 +141,11 @@ exercised a live Windows role.
   appropriate. Raw PDH values can be cumulative numerators, ticks or counts,
   even when the Windows counter name says average, percentage or seconds.
   Convert units explicitly; expose the base/operation count needed for ratios.
-  The raw value of a `PERF_AVERAGE_TIMER` counter is in ticks of its time
-  base: in new code, divide it by `pdh.Collector.Frequency` instead of a fixed
-  factor. The disk and `smbclient` collectors still use the fixed 1e-7
-  (10 MHz) factor.
+  With `pdh.CounterTypeRaw`, `internal/pdh` converts the time types to seconds:
+  the 100 ns timers and `PERF_COUNTER_100NS_QUEUELEN_TYPE` by dividing by
+  `pdh.TicksPerSecond`, and `PERF_AVERAGE_TIMER` by dividing by the counter's
+  own time base (its `,secondvalue` stays the operation count). Don't scale
+  these again in a collector.
   Never publish an invalid sample as zero: that can look like a counter reset.
 - Keep the typed PDH collectors and `perfdata` tags. `internal/pdh` passes
   instance names through unchanged, so a `#N` in a name always comes from
diff --git a/docs/collector.hyperv.md b/docs/collector.hyperv.md
index e1bb14ce..5e379372 100644
--- a/docs/collector.hyperv.md
+++ b/docs/collector.hyperv.md
@@ -369,7 +369,7 @@ The `vm` label holds the instance name of the `Hyper-V Virtual SMB` performance
 
 The `Latency`, `Lower Latency`, `Queue Length` and `Lower Queue Length` performance counters
 are averages that Windows computes from two samples. Their raw values only grow: the sum of
-all IO latencies in 100ns ticks, and the raw queue length values are the same as the raw
+all IO latencies in ticks of the counter time base (usually 10 MHz), and the raw queue length values are the same as the raw
 latency values. windows_exporter publishes them as the `*_io_latency_seconds_total` counters:
 
 ```promql
diff --git a/docs/collector.performancecounter.md b/docs/collector.performancecounter.md
index 6ed78c3a..00dfcf63 100644
--- a/docs/collector.performancecounter.md
+++ b/docs/collector.performancecounter.md
@@ -218,16 +218,24 @@ keep counting up. If the multiplier is the reciprocal of an integer, like `0.001
 the value is divided by that integer instead, which avoids rounding artifacts
 such as `0.009000000000000001`. If not specified, the value is exported as is.
 
-With the object `type` `raw`, the default, some counters are already converted
-to seconds, so do not scale them again: `PERF_100NSEC_TIMER` and
-`PERF_PRECISION_100NS_TIMER` counters such as `% Idle Time` are exported in
-seconds, and `PERF_ELAPSED_TIME` counters such as `Process\Elapsed Time` as
-seconds since the start. `PERF_100NSEC_TIMER_INV` counters such as
-`% Processor Time` are not converted and arrive in 100 ns ticks of inactive time
-(for `% Processor Time`, the idle time); a multiplier of `1e-7` converts them to
-seconds. `PERF_AVERAGE_TIMER` counters such as `Avg. Disk sec/Read` count ticks
-of the performance timer, whose frequency depends on the host, so a fixed
-multiplier does not fit them; use the `formatted` type for them.
+With the object `type` `raw`, the default, time counters are already converted
+to seconds, so do not scale them again:
+
+- the 100 ns timers (`PERF_100NSEC_TIMER`, `PERF_100NSEC_TIMER_INV`,
+  `PERF_100NSEC_MULTI_TIMER`, `PERF_100NSEC_MULTI_TIMER_INV` and
+  `PERF_PRECISION_100NS_TIMER`), such as `% Idle Time`;
+- `PERF_COUNTER_100NS_QUEUELEN_TYPE` counters, such as `Avg. Data Queue Length`
+  of `SMB Client Shares`, as queue length times seconds, so their rate is the
+  average queue length;
+- `PERF_AVERAGE_TIMER` counters, such as `Avg. Disk sec/Read`, as the summed
+  duration in seconds, using the time base of the counter. Their second value
+  is the number of operations;
+- `PERF_ELAPSED_TIME` counters, such as `Process\Elapsed Time`, as seconds since
+  the start.
+
+The raw value of an `_INV` counter is the inactive time: for `% Processor Time`,
+that is the idle time, not the busy time. `MULTI_TIMER` values are summed over
+all units of the counter, so they can grow faster than wall time.
 
 This key is optional.
 
diff --git a/docs/upgrade/0.32.md b/docs/upgrade/0.32.md
index f9076f1b..aab18db2 100644
--- a/docs/upgrade/0.32.md
+++ b/docs/upgrade/0.32.md
@@ -931,6 +931,9 @@ These metrics keep their names, but 0.31 reported wrong values. Graphs show a st
 #### performancecounter and PDH-based collectors
 
 - In the `performancecounter` collector, counters without an explicit `type` now always get the type of the underlying Windows counter. In 0.31 a mixed object could report counters as gauges, or the other way around, and the type could change between scrapes. If Prometheus scrapes with OpenMetrics, a counter that 0.31 reported as a gauge is now stored with a `_total` suffix, which the compatibility rules can't cover.
+- Raw `PERF_100NSEC_TIMER_INV`, `PERF_100NSEC_MULTI_TIMER` and `PERF_100NSEC_MULTI_TIMER_INV` counters, such as `% Processor Time` read with the `raw` type, are exported in seconds, like the other 100 ns timers. In 0.31 they were exported in 100 ns ticks, so values read this way are 10,000,000 times smaller; remove any query that divided them by `1e7`. The raw value of an `_INV` counter such as `% Processor Time` is the inactive (idle) time, not the busy time; `PERF_100NSEC_MULTI_TIMER(_INV)` values are summed over all units of the counter.
+- Raw `PERF_AVERAGE_TIMER` counters, such as `Avg. Disk sec/Read`, and raw `PERF_COUNTER_100NS_QUEUELEN_TYPE` counters, such as `Avg. Data Queue Length`, read with the `performancecounter` collector are exported in seconds and as counters. In 0.31 they were exported in ticks and as gauges. `PERF_AVERAGE_TIMER` uses the time base of the counter, which is 10 MHz on most hosts. The latencies of the `physical_disk`, `smbclient` and `hyperv` collectors now use it too; on a host with another time base, 0.31 reported them wrong.
+- Values converted from 100 ns ticks to seconds, in the `performancecounter`, `cpu`, `process`, `physical_disk`, `logical_disk`, `smbclient`, `container` and `hyperv` collectors, are now rounded correctly and can differ from 0.31 in the last digit. A counter that doesn't move across the upgrade can drop by that last digit once, which `rate()` and `increase()` count as a reset for one window.
 - Instances without a valid sample are left out. In 0.31 they were reported as `0`, which looked like a counter reset.
 - An explicit instance list such as `instances: ["C:", "D:"]` now returns one series per instance. In 0.31 all listed instances were merged into one series.
 - When Windows reports the same instance name more than once, for example every `svchost` process of the `Process` object, the `performancecounter`, `netframework`, `hyperv` and `exchange` (ADAccess processes) collectors publish each occurrence as a separate series. The first keeps the name, later ones get the suffix `#1`, `#2` and so on (`exchange` keeps its `_2`, `_3` label suffixes), which depends on the order of the sample and does not identify the instance across scrapes. The other PDH-based collectors read objects that Windows names uniquely; if one still reports a name twice, they publish the first instance with valid data and log a warning. In 0.31 the occurrences overwrote each other.
diff --git a/internal/collector/adfs/AGENTS.md b/internal/collector/adfs/AGENTS.md
index 90550bcf..d32f912f 100644
--- a/internal/collector/adfs/AGENTS.md
+++ b/internal/collector/adfs/AGENTS.md
@@ -21,10 +21,9 @@ one.
   `PERF_AVERAGE_TIMER` (`AverageTimer32`). The raw value is the summed
   duration in ticks of the counter's time base (`PdhGetCounterTimeBase`,
   10 MHz on the test host, same as `QueryPerformanceFrequency`); the base
-  (`,secondvalue`) is the number of operations. `pdh.CounterTypeRaw` does not
-  convert this type; `internal/pdh` reads its time base, and the collector
-  divides the raw value by `pdh.Collector.Frequency` to get seconds. Do not
-  hardcode a scale factor.
+  (`,secondvalue`) is the number of operations. `internal/pdh` divides the
+  raw value by the time base, so the collector gets seconds. Do not scale it
+  again.
 - Sample after about one hour: `Average Config Database Query Time` raw
   11521176, base 306, which is 1.152 s in total, 3.8 ms per query.
   `Federation Metadata Requests` was 1, all authentication counters 0.
diff --git a/internal/collector/adfs/adfs.go b/internal/collector/adfs/adfs.go
index 6e3ebcb1..05e53308 100644
--- a/internal/collector/adfs/adfs.go
+++ b/internal/collector/adfs/adfs.go
@@ -409,19 +409,13 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric, _ time.Duration) error
 		return fmt.Errorf("failed to collect ADFS metrics: %w", types.ErrNoDataUnexpected)
 	}
 
-	// The query times are PERF_AVERAGE_TIMER counters: the raw value is the
-	// summed duration in ticks of the counter's time base.
-	artifactDBFrequency, _ := c.perfDataCollector.Frequency(counterArtifactDBQueryTime)
-	configDBFrequency, _ := c.perfDataCollector.Frequency(counterConfigDBQueryTime)
-
-	c.collectMetrics(ch, c.perfDataObject[0], artifactDBFrequency, configDBFrequency)
+	c.collectMetrics(ch, c.perfDataObject[0])
 
 	return nil
 }
 
-// collectMetrics sends the metrics of data. A query time is only sent if the
-// frequency of its counter is known.
-func (c *Collector) collectMetrics(ch chan<- prometheus.Metric, data perfDataCounterValues, artifactDBFrequency, configDBFrequency int64) {
+// collectMetrics sends the metrics of data.
+func (c *Collector) collectMetrics(ch chan<- prometheus.Metric, data perfDataCounterValues) {
 	ch <- prometheus.MustNewConstMetric(
 		c.adLoginConnectionFailures,
 		prometheus.CounterValue,
@@ -662,13 +656,13 @@ func (c *Collector) collectMetrics(ch chan<- prometheus.Metric, data perfDataCou
 		data.ArtifactDBQueries,
 	)
 
-	if artifactDBFrequency > 0 {
-		ch <- prometheus.MustNewConstMetric(
-			c.artifactDBQueryTime,
-			prometheus.CounterValue,
-			data.ArtifactDBQueryTimeTicks/float64(artifactDBFrequency),
-		)
-	}
+	// PERF_AVERAGE_TIMER: the summed query time, converted to seconds by
+	// internal/pdh; the second value counts the queries.
+	ch <- prometheus.MustNewConstMetric(
+		c.artifactDBQueryTime,
+		prometheus.CounterValue,
+		data.ArtifactDBQueryTimeSeconds,
+	)
 
 	ch <- prometheus.MustNewConstMetric(
 		c.configDBFailures,
@@ -682,13 +676,13 @@ func (c *Collector) collectMetrics(ch chan<- prometheus.Metric, data perfDataCou
 		data.ConfigDBQueries,
 	)
 
-	if configDBFrequency > 0 {
-		ch <- prometheus.MustNewConstMetric(
-			c.configDBQueryTime,
-			prometheus.CounterValue,
-			data.ConfigDBQueryTimeTicks/float64(configDBFrequency),
-		)
-	}
+	// PERF_AVERAGE_TIMER: the summed query time, converted to seconds by
+	// internal/pdh; the second value counts the queries.
+	ch <- prometheus.MustNewConstMetric(
+		c.configDBQueryTime,
+		prometheus.CounterValue,
+		data.ConfigDBQueryTimeSeconds,
+	)
 
 	ch <- prometheus.MustNewConstMetric(
 		c.federationMetadataRequests,
diff --git a/internal/collector/adfs/adfs_internal_test.go b/internal/collector/adfs/adfs_internal_test.go
index 047d9c5b..1afdbb9d 100644
--- a/internal/collector/adfs/adfs_internal_test.go
+++ b/internal/collector/adfs/adfs_internal_test.go
@@ -25,100 +25,51 @@ import (
 	"github.com/stretchr/testify/require"
 )
 
-// TestCollectDatabaseQueryTimes checks that the raw ticks of the database
-// query time counters are converted to seconds with the counter's time base,
-// and that the query counts are exported.
+// TestCollectDatabaseQueryTimes checks that the database query times, which
+// internal/pdh converts to seconds, and the query counts are exported.
 func TestCollectDatabaseQueryTimes(t *testing.T) {
 	t.Parallel()
 
 	data := perfDataCounterValues{
-		ArtifactDBQueryTimeTicks: 2_500_000,
-		ArtifactDBQueries:        5,
-		ConfigDBQueryTimeTicks:   11_521_176,
-		ConfigDBQueries:          306,
+		ArtifactDBQueryTimeSeconds: 0.25,
+		ArtifactDBQueries:          5,
+		ConfigDBQueryTimeSeconds:   1.1521176,
+		ConfigDBQueries:            306,
 	}
 
-	for _, tc := range []struct {
-		name                string
-		artifactDBFrequency int64
-		configDBFrequency   int64
-		expected            map[string]float64
-		absent              []string
-	}{
-		{
-			name:                "10 MHz",
-			artifactDBFrequency: 10_000_000,
-			configDBFrequency:   10_000_000,
-			expected: map[string]float64{
-				"windows_adfs_db_artifact_query_time_seconds_total": 0.25,
-				"windows_adfs_db_artifact_queries_total":            5,
-				"windows_adfs_db_config_query_time_seconds_total":   1.1521176,
-				"windows_adfs_db_config_queries_total":              306,
-			},
-		},
-		{
-			name:                "other frequency",
-			artifactDBFrequency: 2_500_000,
-			configDBFrequency:   3_000_000,
-			expected: map[string]float64{
-				"windows_adfs_db_artifact_query_time_seconds_total": 1,
-				"windows_adfs_db_config_query_time_seconds_total":   3.840392,
-			},
-		},
-		{
-			name: "unknown frequency",
-			expected: map[string]float64{
-				"windows_adfs_db_artifact_queries_total": 5,
-				"windows_adfs_db_config_queries_total":   306,
-			},
-			absent: []string{
-				"windows_adfs_db_artifact_query_time_seconds_total",
-				"windows_adfs_db_config_query_time_seconds_total",
-			},
-		},
-	} {
-		t.Run(tc.name, func(t *testing.T) {
-			t.Parallel()
+	expected := map[string]float64{
+		"windows_adfs_db_artifact_query_time_seconds_total": 0.25,
+		"windows_adfs_db_artifact_queries_total":            5,
+		"windows_adfs_db_config_query_time_seconds_total":   1.1521176,
+		"windows_adfs_db_config_queries_total":              306,
+	}
 
-			c := New(nil)
-			c.buildDescriptors()
+	c := New(nil)
+	c.buildDescriptors()
 
-			registry := prometheus.NewPedanticRegistry()
-			require.NoError(t, registry.Register(metricCollector{
-				collector:           c,
-				data:                data,
-				artifactDBFrequency: tc.artifactDBFrequency,
-				configDBFrequency:   tc.configDBFrequency,
-			}))
+	registry := prometheus.NewPedanticRegistry()
+	require.NoError(t, registry.Register(metricCollector{collector: c, data: data}))
 
-			families, err := registry.Gather()
-			require.NoError(t, err)
+	families, err := registry.Gather()
+	require.NoError(t, err)
 
-			byName := make(map[string]*dto.MetricFamily, len(families))
-			for _, family := range families {
-				byName[family.GetName()] = family
-			}
+	byName := make(map[string]*dto.MetricFamily, len(families))
+	for _, family := range families {
+		byName[family.GetName()] = family
+	}
 
-			for name, value := range tc.expected {
-				family, ok := byName[name]
-				require.True(t, ok, "missing metric %s", name)
-				require.Equal(t, dto.MetricType_COUNTER, family.GetType(), name)
-				require.Len(t, family.GetMetric(), 1, name)
-				require.InDelta(t, value, family.GetMetric()[0].GetCounter().GetValue(), 1e-9, name)
-			}
-
-			for _, name := range tc.absent {
-				require.NotContains(t, byName, name)
-			}
-		})
+	for name, value := range expected {
+		family, ok := byName[name]
+		require.True(t, ok, "missing metric %s", name)
+		require.Equal(t, dto.MetricType_COUNTER, family.GetType(), name)
+		require.Len(t, family.GetMetric(), 1, name)
+		require.InDelta(t, value, family.GetMetric()[0].GetCounter().GetValue(), 0, name)
 	}
 }
 
 type metricCollector struct {
-	collector           *Collector
-	data                perfDataCounterValues
-	artifactDBFrequency int64
-	configDBFrequency   int64
+	collector *Collector
+	data      perfDataCounterValues
 }
 
 func (c metricCollector) Describe(ch chan<- *prometheus.Desc) {
@@ -126,5 +77,5 @@ func (c metricCollector) Describe(ch chan<- *prometheus.Desc) {
 }
 
 func (c metricCollector) Collect(ch chan<- prometheus.Metric) {
-	c.collector.collectMetrics(ch, c.data, c.artifactDBFrequency, c.configDBFrequency)
+	c.collector.collectMetrics(ch, c.data)
 }
diff --git a/internal/collector/adfs/types.go b/internal/collector/adfs/types.go
index 0f840319..fc5dd216 100644
--- a/internal/collector/adfs/types.go
+++ b/internal/collector/adfs/types.go
@@ -17,19 +17,12 @@
 
 package adfs
 
-// PERF_AVERAGE_TIMER counters. The raw value is the summed query time in ticks
-// of the counter's time base, the second value the number of queries.
-const (
-	counterArtifactDBQueryTime = "Average Artifact Database Query Time"
-	counterConfigDBQueryTime   = "Average Config Database Query Time"
-)
-
 type perfDataCounterValues struct {
 	AdLoginConnectionFailures                      float64 `perfdata:"AD Login Connection Failures"`
 	ArtifactDBFailures                             float64 `perfdata:"Artifact Database Connection Failures"`
-	ArtifactDBQueryTimeTicks                       float64 `perfdata:"Average Artifact Database Query Time"`
+	ArtifactDBQueryTimeSeconds                     float64 `perfdata:"Average Artifact Database Query Time"`
 	ArtifactDBQueries                              float64 `perfdata:"Average Artifact Database Query Time,secondvalue"`
-	ConfigDBQueryTimeTicks                         float64 `perfdata:"Average Config Database Query Time"`
+	ConfigDBQueryTimeSeconds                       float64 `perfdata:"Average Config Database Query Time"`
 	ConfigDBQueries                                float64 `perfdata:"Average Config Database Query Time,secondvalue"`
 	CertificateAuthentications                     float64 `perfdata:"Certificate Authentications"`
 	ConfigDBFailures                               float64 `perfdata:"Configuration Database Connection Failures"`
diff --git a/internal/collector/container/container.go b/internal/collector/container/container.go
index baf5829f..a0e151ed 100644
--- a/internal/collector/container/container.go
+++ b/internal/collector/container/container.go
@@ -671,7 +671,7 @@ func (c *Collector) collectHCSContainer(ch chan<- prometheus.Metric, containerDe
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeTotal,
 		prometheus.CounterValue,
-		float64(containerStats.Processor.TotalRuntime100ns)*pdh.TicksToSecondScaleFactor,
+		float64(containerStats.Processor.TotalRuntime100ns)/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
@@ -679,7 +679,7 @@ func (c *Collector) collectHCSContainer(ch chan<- prometheus.Metric, containerDe
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeUser,
 		prometheus.CounterValue,
-		float64(containerStats.Processor.RuntimeUser100ns)*pdh.TicksToSecondScaleFactor,
+		float64(containerStats.Processor.RuntimeUser100ns)/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
@@ -687,7 +687,7 @@ func (c *Collector) collectHCSContainer(ch chan<- prometheus.Metric, containerDe
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeKernel,
 		prometheus.CounterValue,
-		float64(containerStats.Processor.RuntimeKernel100ns)*pdh.TicksToSecondScaleFactor,
+		float64(containerStats.Processor.RuntimeKernel100ns)/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
@@ -939,7 +939,7 @@ func (c *Collector) collectJobContainer(ch chan<- prometheus.Metric, containerID
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeTotal,
 		prometheus.CounterValue,
-		(float64(jobInfo.BasicInfo.TotalKernelTime)+float64(jobInfo.BasicInfo.TotalUserTime))*pdh.TicksToSecondScaleFactor,
+		(float64(jobInfo.BasicInfo.TotalKernelTime)+float64(jobInfo.BasicInfo.TotalUserTime))/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
@@ -947,7 +947,7 @@ func (c *Collector) collectJobContainer(ch chan<- prometheus.Metric, containerID
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeUser,
 		prometheus.CounterValue,
-		float64(jobInfo.BasicInfo.TotalUserTime)*pdh.TicksToSecondScaleFactor,
+		float64(jobInfo.BasicInfo.TotalUserTime)/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
@@ -955,7 +955,7 @@ func (c *Collector) collectJobContainer(ch chan<- prometheus.Metric, containerID
 	ch <- prometheus.MustNewConstMetric(
 		c.runtimeKernel,
 		prometheus.CounterValue,
-		float64(jobInfo.BasicInfo.TotalKernelTime)*pdh.TicksToSecondScaleFactor,
+		float64(jobInfo.BasicInfo.TotalKernelTime)/pdh.TicksPerSecond,
 
 		containerInfo.id, containerInfo.namespace, containerInfo.pod, containerInfo.container,
 	)
diff --git a/internal/collector/cpu/AGENTS.md b/internal/collector/cpu/AGENTS.md
index b48fb208..298f9114 100644
--- a/internal/collector/cpu/AGENTS.md
+++ b/internal/collector/cpu/AGENTS.md
@@ -18,9 +18,9 @@ Hyper-V guest with 12 logical processors in one processor group.
   so `Collect` holds the previous value of `time_seconds_total` and
   `cstate_seconds_total` for drops below `maxTimeDrop` (1 s).
 - `% Processor Time` and `% Priority Time` are `PERF_100NSEC_TIMER_INV`. The
-  raw value is cumulative idle (non-busy) time in 100 ns ticks, not busy time,
-  and `internal/pdh` does not scale this type to seconds. In the sample, the
-  raw value of `% Processor Time` equals `% Idle Time`.
+  raw value is cumulative idle (non-busy) time in 100 ns ticks, not busy time.
+  `internal/pdh` scales it to seconds like the other 100 ns timers. In the
+  sample, the raw value of `% Processor Time` equals `% Idle Time`.
 - `% Processor Utility`, `% Processor Performance` and `% Privileged Utility`
   are `PERF_AVERAGE_BULK`. Raw and base are both cumulative. The displayed
   value is delta raw / delta base and is already a percentage (it can exceed
diff --git a/internal/collector/gpu/gpu.go b/internal/collector/gpu/gpu.go
index 7bb31507..c33003b9 100644
--- a/internal/collector/gpu/gpu.go
+++ b/internal/collector/gpu/gpu.go
@@ -512,7 +512,7 @@ func (c *Collector) collectGpuEngineMetrics(ch chan<- prometheus.Metric, scrape
 		ch <- scrape.MustNewConstMetric(
 			c.gpuEngineRunningTime,
 			prometheus.CounterValue,
-			data.RunningTime/10_000_000,
+			data.RunningTime/pdh.TicksPerSecond,
 			instance.Pid, instance.Luid, device.ID, instance.Phys, instance.Eng, instance.Engtype,
 		)
 	}
diff --git a/internal/collector/hyperv/hyperv_virtual_storage_device.go b/internal/collector/hyperv/hyperv_virtual_storage_device.go
index 8ddba046..4f7c6228 100644
--- a/internal/collector/hyperv/hyperv_virtual_storage_device.go
+++ b/internal/collector/hyperv/hyperv_virtual_storage_device.go
@@ -205,19 +205,20 @@ func (c *Collector) publishVirtualStorageDevice(ch chan<- prometheus.Metric, scr
 			data.Name,
 		)
 
-		// The raw value of a PERF_AVERAGE_TIMER counter is the sum of all IO latencies in
-		// 100ns ticks; the operation count it is averaged over is Throughput.
+		// The value of a PERF_AVERAGE_TIMER counter is the sum of all IO latencies,
+		// converted to seconds by internal/pdh; the operation count it is averaged
+		// over is Throughput.
 		ch <- scrape.MustNewConstMetric(
 			c.virtualStorageDeviceIOLatency,
 			prometheus.CounterValue,
-			data.VirtualStorageDeviceLatency*pdh.TicksToSecondScaleFactor,
+			data.VirtualStorageDeviceLatency,
 			data.Name,
 		)
 
 		ch <- scrape.MustNewConstMetric(
 			c.virtualStorageDeviceLowerIOLatency,
 			prometheus.CounterValue,
-			data.VirtualStorageDeviceLowerLatency*pdh.TicksToSecondScaleFactor,
+			data.VirtualStorageDeviceLowerLatency,
 			data.Name,
 		)
 	}
diff --git a/internal/collector/logical_disk/AGENTS.md b/internal/collector/logical_disk/AGENTS.md
index a983e322..3d319e32 100644
--- a/internal/collector/logical_disk/AGENTS.md
+++ b/internal/collector/logical_disk/AGENTS.md
@@ -19,11 +19,9 @@ multi-instance. These notes were recorded on a Windows Server 2025
   were all 9858832; the write counters were all 11319205, and
   `Avg. Disk sec/Transfer` (21178037) is their sum. The raw value is the
   cumulative busy time in 100 ns ticks. `internal/pdh` converts the
-  `PERF_PRECISION_100NS_TIMER` counters to seconds. The queue length and
-  `PERF_AVERAGE_TIMER` values arrive in ticks and need the 1e-7 factor; the
-  counter time base is 10 MHz here. The queue length counters duplicate the
-  busy time, so the collector no longer reads them; the average queue length is
-  `rate(windows_logical_disk_{read,write}_seconds_total[5m])`.
+  `PERF_PRECISION_100NS_TIMER` counters to seconds. The queue length counters
+  duplicate the busy time, so the collector no longer reads them; the average
+  queue length is `rate(windows_logical_disk_{read,write}_seconds_total[5m])`.
 - `Avg. Disk sec/Read`, `/Write` and `/Transfer` are `PERF_AVERAGE_TIMER`.
   Their base is the operation count and equals `Disk Reads/sec` (3073),
   `Disk Writes/sec` (1698) and their sum (4771). Their raw values duplicate the
diff --git a/internal/collector/physical_disk/AGENTS.md b/internal/collector/physical_disk/AGENTS.md
index 02aa17fd..d75e9680 100644
--- a/internal/collector/physical_disk/AGENTS.md
+++ b/internal/collector/physical_disk/AGENTS.md
@@ -17,8 +17,8 @@ multi-instance. These notes were recorded on a Windows Server 2025
   value is the cumulative busy time in 100 ns ticks.
 - `internal/pdh` converts the `PERF_PRECISION_100NS_TIMER` counters
   (`% Disk Read Time`, `% Disk Write Time`, `% Idle Time`) to seconds.
-  `PERF_AVERAGE_TIMER` values arrive in ticks of the 10 MHz counter time base
-  and need the 1e-7 factor.
+  It also converts the `PERF_AVERAGE_TIMER` values (`Avg. Disk sec/*`) to
+  seconds with the counter time base, which was 10 MHz on JOK-PC.
 - The second raw value of `% Idle Time` is the sample time as a UTC FILETIME;
   PDH divides by its delta. Even with that divisor the idle delta of a busy
   disk exceeded the elapsed time by up to 43% between 1 s samples on a
diff --git a/internal/collector/physical_disk/physical_disk.go b/internal/collector/physical_disk/physical_disk.go
index fec57aaa..6d587674 100644
--- a/internal/collector/physical_disk/physical_disk.go
+++ b/internal/collector/physical_disk/physical_disk.go
@@ -295,21 +295,21 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric, _ time.Duration) error
 		ch <- prometheus.MustNewConstMetric(
 			c.readLatency,
 			prometheus.CounterValue,
-			data.AvgDiskSecPerRead*pdh.TicksToSecondScaleFactor,
+			data.AvgDiskSecPerRead,
 			disk_number,
 		)
 
 		ch <- prometheus.MustNewConstMetric(
 			c.writeLatency,
 			prometheus.CounterValue,
-			data.AvgDiskSecPerWrite*pdh.TicksToSecondScaleFactor,
+			data.AvgDiskSecPerWrite,
 			disk_number,
 		)
 
 		ch <- prometheus.MustNewConstMetric(
 			c.readWriteLatency,
 			prometheus.CounterValue,
-			data.AvgDiskSecPerTransfer*pdh.TicksToSecondScaleFactor,
+			data.AvgDiskSecPerTransfer,
 			disk_number,
 		)
 	}
diff --git a/internal/collector/process/process.go b/internal/collector/process/process.go
index 6ed734b9..0e5efdb6 100644
--- a/internal/collector/process/process.go
+++ b/internal/collector/process/process.go
@@ -42,11 +42,10 @@ const Name = "process"
 const (
 	// windowsEpoch is the FILETIME of the Unix epoch, in 100ns intervals since 1601.
 	windowsEpoch int64 = 116444736000000000
-	// ticksPerSecond is the number of 100ns intervals per second.
+	// ticksPerSecond is the number of 100ns intervals per second. Dividing by it,
+	// like internal/pdh does, keeps the CPU time values bit-identical to the
+	// Process counter set.
 	ticksPerSecond = 1e7
-	// secondsPerTick converts 100ns intervals to seconds. Multiplying by it, like the perflib
-	// reader did, keeps the CPU time values bit-identical to the Process counter set.
-	secondsPerTick = 1 / ticksPerSecond
 )
 
 type Config struct {
@@ -486,14 +485,14 @@ func (c *Collector) collectProcess(ch chan<- prometheus.Metric, scrape *labelcac
 	ch <- scrape.MustNewConstMetric(
 		c.cpuTimeTotal,
 		prometheus.CounterValue,
-		float64(process.kernelTime)*secondsPerTick,
+		float64(process.kernelTime)/ticksPerSecond,
 		name, pid, "privileged",
 	)
 
 	ch <- scrape.MustNewConstMetric(
 		c.cpuTimeTotal,
 		prometheus.CounterValue,
-		float64(process.userTime)*secondsPerTick,
+		float64(process.userTime)/ticksPerSecond,
 		name, pid, "user",
 	)
 
diff --git a/internal/collector/smbclient/AGENTS.md b/internal/collector/smbclient/AGENTS.md
index df9c4e26..fc4cae2b 100644
--- a/internal/collector/smbclient/AGENTS.md
+++ b/internal/collector/smbclient/AGENTS.md
@@ -13,12 +13,16 @@ host has connected to. These notes were recorded on a Windows Server 2025
 - `Current Data Queue Length` is the only gauge ("the current number of read
   or write requests outstanding"). The `/sec` counters are cumulative counts.
 - `Avg. Data/Read/Write Queue Length` are `PERF_COUNTER_100NS_QUEUELEN_TYPE`:
-  the raw value is accumulated queue time in 100 ns ticks. `internal/pdh` does
-  not convert this type, so the collector applies the 1e-7 factor.
+  the raw value is accumulated queue time in 100 ns ticks, which
+  `internal/pdh` converts to seconds.
 - `Avg. sec/Read`, `Avg. sec/Write` and `Avg. sec/Data Request` are
-  `PERF_AVERAGE_TIMER`. The first value is the summed latency in ticks of the
-  10 MHz counter time base and the base is the request count; the collector
-  reads only the first value.
+  `PERF_AVERAGE_TIMER`. The first value is the summed latency, which
+  `internal/pdh` converts to seconds with the counter time base, and the base
+  is the request count; the collector reads only the first value. The raw
+  sum is 32 bits, so the collector converts it back to ticks with
+  `pdh.Collector.Frequency`, extends it past the 2^32 wrap with
+  `utils.Counter32` and divides again. Keep that round trip: extending the
+  seconds directly truncates them and adds 2^32 seconds per wrap.
 - `Data Bytes/sec`, `Read Bytes/sec` and `Write Bytes/sec` are bytes
   (`PERF_COUNTER_BULK_COUNT`).
 
diff --git a/internal/collector/smbclient/smbclient.go b/internal/collector/smbclient/smbclient.go
index 46a90214..ef4470ce 100644
--- a/internal/collector/smbclient/smbclient.go
+++ b/internal/collector/smbclient/smbclient.go
@@ -20,6 +20,7 @@ package smbclient
 import (
 	"fmt"
 	"log/slog"
+	"math"
 	"strings"
 	"time"
 
@@ -111,6 +112,12 @@ func restartSignals(data *perfDataCounterValues) [numRestartSignals]float64 {
 	}
 }
 
+// timerCounters32 names the PERF_AVERAGE_TIMER counters in the order of
+// fields32.
+//
+//nolint:gochecknoglobals
+var timerCounters32 = [numTimers32]string{"Avg. sec/Read", "Avg. sec/Write", "Avg. sec/Data Request"}
+
 // fields32 returns the 32-bit counters: first the PERF_AVERAGE_TIMER
 // numerators, then the request counters (PERF_COUNTER_COUNTER).
 func fields32(data *perfDataCounterValues) [numTimers32 + numCounts32]*float64 {
@@ -275,6 +282,16 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric, _ time.Duration) error
 
 	c.generation++
 
+	// internal/pdh converts the timer sums to seconds; extendCounters32 needs
+	// their ticks.
+	var frequencies [numTimers32]float64
+
+	for i, name := range timerCounters32 {
+		if frequency, ok := c.perfDataCollector.Frequency(name); ok {
+			frequencies[i] = float64(frequency)
+		}
+	}
+
 	for _, data := range c.perfDataObject {
 		serverValue, shareValue, ok := parseInstanceName(data.Name)
 		if !ok {
@@ -289,50 +306,50 @@ func (c *Collector) Collect(ch chan<- prometheus.Metric, _ time.Duration) error
 			continue
 		}
 
-		c.extendCounters32(&data)
+		c.extendCounters32(&data, frequencies)
 
-		// Request time spent on queue. Convert from ticks to seconds.
+		// Request time spent on queue, in seconds.
 		ch <- prometheus.MustNewConstMetric(
 			c.requestQueueSecsTotal,
 			prometheus.CounterValue,
-			data.AvgDataQueueLength*pdh.TicksToSecondScaleFactor,
+			data.AvgDataQueueLength,
 			serverValue, shareValue,
 		)
 
-		// Read time spent on queue. Convert from ticks to seconds.
+		// Read time spent on queue, in seconds.
 		ch <- prometheus.MustNewConstMetric(
 			c.readRequestQueueSecsTotal,
 			prometheus.CounterValue,
-			data.AvgReadQueueLength*pdh.TicksToSecondScaleFactor,
+			data.AvgReadQueueLength,
 			serverValue, shareValue,
 		)
 
 		ch <- prometheus.MustNewConstMetric(
 			c.readSecsTotal,
 			prometheus.CounterValue,
-			data.AvgSecPerRead*pdh.TicksToSecondScaleFactor,
+			data.AvgSecPerRead,
 			serverValue, shareValue,
 		)
 
 		ch <- prometheus.MustNewConstMetric(
 			c.writeSecsTotal,
 			prometheus.CounterValue,
-			data.AvgSecPerWrite*pdh.TicksToSecondScaleFactor,
+			data.AvgSecPerWrite,
 			serverValue, shareValue,
 		)
 
 		ch <- prometheus.MustNewConstMetric(
 			c.requestSecs,
 			prometheus.CounterValue,
-			data.AvgSecPerDataRequest*pdh.TicksToSecondScaleFactor,
+			data.AvgSecPerDataRequest,
 			serverValue, shareValue,
 		)
 
-		// Write time spent on queue. Convert from  ticks to seconds.
+		// Write time spent on queue, in seconds.
 		ch <- prometheus.MustNewConstMetric(
 			c.writeRequestQueueSecsTotal,
 			prometheus.CounterValue,
-			data.AvgWriteQueueLength*pdh.TicksToSecondScaleFactor,
+			data.AvgWriteQueueLength,
 			serverValue, shareValue,
 		)
 
@@ -458,11 +475,18 @@ func (c *Collector) pruneCounters32() {
 	}
 }
 
-// extendCounters32 replaces the raw 32-bit counters with values that keep
-// counting past 2^32.
-func (c *Collector) extendCounters32(data *perfDataCounterValues) {
+// extendCounters32 replaces the 32-bit counters with values that keep
+// counting past 2^32. The timer sums arrive in seconds and wrap at 2^32 ticks
+// of their time base in frequencies, so they are extended as ticks.
+func (c *Collector) extendCounters32(data *perfDataCounterValues, frequencies [numTimers32]float64) {
 	fields := fields32(data)
 
+	for i, frequency := range frequencies {
+		if frequency > 0 {
+			*fields[i] = math.Round(*fields[i] * frequency)
+		}
+	}
+
 	signals := restartSignals(data)
 
 	counters, ok := c.counters32[data.Name]
@@ -493,6 +517,12 @@ func (c *Collector) extendCounters32(data *perfDataCounterValues) {
 	for i, field := range fields {
 		*field = counters.counters[i].Update(*field, reset)
 	}
+
+	for i, frequency := range frequencies {
+		if frequency > 0 {
+			*fields[i] /= frequency
+		}
+	}
 }
 
 // parseInstanceName splits an SMB Client Shares instance name such as
diff --git a/internal/collector/smbclient/smbclient_internal_test.go b/internal/collector/smbclient/smbclient_internal_test.go
index 1c1dc076..ab278931 100644
--- a/internal/collector/smbclient/smbclient_internal_test.go
+++ b/internal/collector/smbclient/smbclient_internal_test.go
@@ -53,29 +53,36 @@ func TestExtendCounters32(t *testing.T) {
 
 	c := &Collector{counters32: make(map[string]*shareCounters32)}
 
+	// internal/pdh delivers the timer sums in seconds of a 10 MHz time base.
+	frequencies := [numTimers32]float64{1e7, 1e7, 1e7}
+
 	sample := func(data perfDataCounterValues) perfDataCounterValues {
 		c.generation++
 
 		data.Name = `\\server\share`
-		c.extendCounters32(&data)
+		c.extendCounters32(&data, frequencies)
 		c.pruneCounters32()
 
 		return data
 	}
 
-	got := sample(perfDataCounterValues{AvgSecPerRead: 1 << 20, ReadRequestsPerSec: 7, ReadBytesPerSec: 100})
-	require.Equal(t, float64(1<<20), got.AvgSecPerRead) //nolint:testifylint
-	require.Equal(t, 7.0, got.ReadRequestsPerSec)       //nolint:testifylint
+	got := sample(perfDataCounterValues{AvgSecPerRead: (1 << 20) / 1e7, ReadRequestsPerSec: 7, ReadBytesPerSec: 100})
+	require.Equal(t, (1<<20)/1e7, got.AvgSecPerRead) //nolint:testifylint
+	require.Equal(t, 7.0, got.ReadRequestsPerSec)    //nolint:testifylint
 
 	// A timer sum can wrap from anywhere: 2^32 ticks are about 430 s of
 	// summed request time.
-	got = sample(perfDataCounterValues{AvgSecPerRead: 1 << 10, ReadRequestsPerSec: 9, ReadBytesPerSec: 200})
-	require.Equal(t, float64(math.MaxUint32)+1+(1<<10), got.AvgSecPerRead) //nolint:testifylint
-	require.Equal(t, 9.0, got.ReadRequestsPerSec)                          //nolint:testifylint
+	got = sample(perfDataCounterValues{AvgSecPerRead: (1 << 10) / 1e7, ReadRequestsPerSec: 9, ReadBytesPerSec: 200})
+	require.Equal(t, (float64(math.MaxUint32)+1+(1<<10))/1e7, got.AvgSecPerRead) //nolint:testifylint
+	require.Equal(t, 9.0, got.ReadRequestsPerSec)                                //nolint:testifylint
+
+	// Fractions of a second survive the round trip through ticks.
+	got = sample(perfDataCounterValues{AvgSecPerRead: 0.1234567, ReadRequestsPerSec: 10, ReadBytesPerSec: 300})
+	require.Equal(t, (float64(math.MaxUint32)+1+1234567)/1e7, got.AvgSecPerRead) //nolint:testifylint
 
 	// A drop of a 64-bit byte counter shows a reconnect, even if DataBytes grew.
-	got = sample(perfDataCounterValues{AvgSecPerRead: 5, ReadRequestsPerSec: 1, ReadBytesPerSec: 50, DataBytesPerSec: 1000})
-	require.Equal(t, 5.0, got.AvgSecPerRead)      //nolint:testifylint
+	got = sample(perfDataCounterValues{AvgSecPerRead: 5 / 1e7, ReadRequestsPerSec: 1, ReadBytesPerSec: 50, DataBytesPerSec: 1000})
+	require.Equal(t, 5/1e7, got.AvgSecPerRead)    //nolint:testifylint
 	require.Equal(t, 1.0, got.ReadRequestsPerSec) //nolint:testifylint
 
 	// A metadata-only share has no byte counter to show the reconnect; a
diff --git a/internal/headers/kernel32/kernel32.go b/internal/headers/kernel32/kernel32.go
index 4f77b18f..f1c4d3b2 100644
--- a/internal/headers/kernel32/kernel32.go
+++ b/internal/headers/kernel32/kernel32.go
@@ -32,6 +32,7 @@ var (
 	procKernelLocalFileTimeToFileTime    = modkernel32.NewProc("LocalFileTimeToFileTime")
 	procOpenJobObject                    = modkernel32.NewProc("OpenJobObjectW")
 	procIsProcessInJob                   = modkernel32.NewProc("IsProcessInJob")
+	procQueryPerformanceFrequency        = modkernel32.NewProc("QueryPerformanceFrequency")
 
 	// GetDiskSpaceInformationW is optional and checked by GetDiskSpaceInformation.
 	load = lazyproc.Once(modkernel32,
@@ -39,6 +40,7 @@ var (
 		procKernelLocalFileTimeToFileTime,
 		procOpenJobObject,
 		procIsProcessInJob,
+		procQueryPerformanceFrequency,
 	)
 )
 
@@ -94,3 +96,17 @@ func LocalFileTimeToFileTime(localFileTime, utcFileTime *windows.Filetime) uint3
 
 	return uint32(ret)
 }
+
+// QueryPerformanceFrequency returns the frequency of the performance counter
+// in counts per second.
+// 📑 https://learn.microsoft.com/en-us/windows/win32/api/profileapi/nf-profileapi-queryperformancefrequency
+func QueryPerformanceFrequency() (int64, error) {
+	var frequency int64
+
+	ret, _, err := procQueryPerformanceFrequency.Call(uintptr(unsafe.Pointer(&frequency)))
+	if ret == 0 {
+		return 0, err
+	}
+
+	return frequency, nil
+}
diff --git a/internal/pdh/collector.go b/internal/pdh/collector.go
index ea9c56a6..5328110b 100644
--- a/internal/pdh/collector.go
+++ b/internal/pdh/collector.go
@@ -31,6 +31,7 @@ import (
 	"sync"
 	"unsafe"
 
+	"github.com/prometheus-community/windows_exporter/internal/headers/kernel32"
 	"github.com/prometheus-community/windows_exporter/internal/osversion"
 	"github.com/prometheus/client_golang/prometheus"
 	"golang.org/x/sys/windows"
@@ -487,10 +488,10 @@ func newCollector[T any](logger *slog.Logger, query *Query, resultType CounterTy
 					continue
 				}
 			case PERF_AVERAGE_TIMER:
-				// Only [Collector.Frequency] uses the time base of an average
-				// timer, so a missing one is not an error. Collect retries it.
-				if ret := GetCounterTimeBase(counterHandle, &counter.Frequency); ret != ErrorSuccess {
-					counter.Frequency = 0
+				// The time base of an average timer is the performance counter
+				// frequency, so use that if PDH has none.
+				if ret := GetCounterTimeBase(counterHandle, &counter.Frequency); ret != ErrorSuccess || counter.Frequency <= 0 {
+					counter.Frequency, _ = kernel32.QueryPerformanceFrequency()
 				}
 			}
 		}
@@ -569,11 +570,11 @@ func (c *Collector[T]) MetricType(counterName string) (prometheus.ValueType, boo
 }
 
 // Frequency returns the time base in ticks per second of the raw timer counter
-// named counterName. The raw value of a PERF_AVERAGE_TIMER counter is the sum
-// of the measured durations in these ticks; divide it by the frequency to get
-// seconds. It returns false if the counter is unknown, is not such a timer or
-// its time base is not known yet. [Collector.Collect] reads a missing time
-// base again.
+// named counterName. [Collector.Collect] already converts the values of these
+// counters to seconds; a collector needs the time base only to work on the
+// ticks, for example to extend a 32-bit counter past its wrap. It returns
+// false if the counter is unknown, is not such a timer or its time base is not
+// known yet.
 func (c *Collector[T]) Frequency(counterName string) (int64, bool) {
 	if c == nil {
 		return 0, false
@@ -724,7 +725,8 @@ func hasTimeBase(counterType uint32) bool {
 // the collector was created. Early after a reboot, PDH can report no data
 // for it, e.g. for APP_POOL_WAS while WAS is still starting (#2313). Without a
 // time base, an elapsed time counter has no valid value and every instance is
-// omitted.
+// omitted. An average timer falls back to QueryPerformanceFrequency at
+// creation, so it only needs this if that failed too.
 func (c *Collector[T]) loadTimeBase(counter *Counter) {
 	if c.resultType != CounterTypeRaw || !hasTimeBase(counter.Type) || counter.Frequency > 0 {
 		return
@@ -810,9 +812,26 @@ func (c *Collector[T]) setRawValue(row *T, counter *Counter, value RawCounter) b
 		if counter.FieldIndexValue != -1 {
 			c.rows.setValue(row, counter.FieldIndexValue, float64(value.SecondValue-value.FirstValue)/float64(counter.Frequency))
 		}
-	case PERF_100NSEC_TIMER, PERF_PRECISION_100NS_TIMER:
+	case PERF_100NSEC_TIMER, PERF_100NSEC_TIMER_INV, PERF_100NSEC_MULTI_TIMER, PERF_100NSEC_MULTI_TIMER_INV, PERF_PRECISION_100NS_TIMER,
+		PERF_COUNTER_100NS_QUEUELEN_TYPE:
+		// The queue length type sums the queue length times the elapsed 100 ns
+		// ticks, so in seconds its rate is the average queue length.
 		if counter.FieldIndexValue != -1 {
-			c.rows.setValue(row, counter.FieldIndexValue, float64(value.FirstValue)*TicksToSecondScaleFactor)
+			c.rows.setValue(row, counter.FieldIndexValue, float64(value.FirstValue)/TicksPerSecond)
+		}
+	case PERF_AVERAGE_TIMER:
+		// The first value sums the measured durations in ticks of the counter's
+		// time base, the second value counts the operations.
+		if counter.Frequency <= 0 {
+			return false
+		}
+
+		if counter.FieldIndexSecondValue != -1 {
+			c.rows.setValue(row, counter.FieldIndexSecondValue, float64(value.SecondValue))
+		}
+
+		if counter.FieldIndexValue != -1 {
+			c.rows.setValue(row, counter.FieldIndexValue, float64(value.FirstValue)/float64(counter.Frequency))
 		}
 	default:
 		if counter.FieldIndexSecondValue != -1 {
diff --git a/internal/pdh/collector_internal_test.go b/internal/pdh/collector_internal_test.go
index 8c20f08e..d9397ee6 100644
--- a/internal/pdh/collector_internal_test.go
+++ b/internal/pdh/collector_internal_test.go
@@ -385,6 +385,32 @@ func TestSetRawValueElapsedTime(t *testing.T) {
 	require.True(t, c.setRawValue(&values, counter, RawCounter{FirstValue: 1, SecondValue: 2}))
 }
 
+// TestSetRawValue100nsTimers checks that every 100 ns timer type, and the
+// 100 ns queue length type, is exported in seconds. 115960642 * (1 / 1e7) is
+// 11.596064199999999, so the test also catches a multiplication by the
+// reciprocal.
+func TestSetRawValue100nsTimers(t *testing.T) {
+	t.Parallel()
+
+	c := newRowSetTestCollector(false)
+
+	for _, counterType := range []uint32{
+		PERF_100NSEC_TIMER,
+		PERF_100NSEC_TIMER_INV,
+		PERF_100NSEC_MULTI_TIMER,
+		PERF_100NSEC_MULTI_TIMER_INV,
+		PERF_PRECISION_100NS_TIMER,
+		PERF_COUNTER_100NS_QUEUELEN_TYPE,
+	} {
+		var values rowSetValues
+
+		counter := &Counter{Type: counterType, FieldIndexValue: int(unsafe.Offsetof(values.A)), FieldIndexSecondValue: -1}
+
+		require.True(t, c.setRawValue(&values, counter, RawCounter{FirstValue: 115_960_642, SecondValue: 1}))
+		require.InDelta(t, 11.5960642, values.A, 0, "counter type %#x", counterType)
+	}
+}
+
 // TestCollectLoadsMissingTimeBase checks that an elapsed time counter whose time
 // base was not available when the collector was created gets it on a later
 // sample, instead of omitting every instance until a restart (#2313).
@@ -419,17 +445,15 @@ func TestCollectLoadsMissingTimeBase(t *testing.T) {
 	require.Positive(t, dst[0].SystemUpTime)
 }
 
-// TestFrequencyOfAverageTimer checks that the time base of a raw average timer
-// is available, also when it was missing at startup, and that other counters
-// have none.
-func TestFrequencyOfAverageTimer(t *testing.T) {
+// TestAverageTimerInSeconds checks that the raw value of an average timer is
+// converted to seconds with the counter's time base, also when the time base
+// was missing at startup, and that the second value keeps the operation count.
+func TestAverageTimerInSeconds(t *testing.T) {
 	t.Parallel()
 
 	type physicalDisk struct {
-		AvgDiskSecPerRead      float64 `perfdata:"Avg. Disk sec/Read"`
-		AvgDiskSecPerReadBase  float64 `perfdata:"Avg. Disk sec/Read,secondvalue"`
-		DiskReadsPerSec        float64 `perfdata:"Disk Reads/sec"`
-		CurrentDiskQueueLength float64 `perfdata:"Current Disk Queue Length"`
+		AvgDiskSecPerRead     float64 `perfdata:"Avg. Disk sec/Read"`
+		AvgDiskSecPerReadBase float64 `perfdata:"Avg. Disk sec/Read,secondvalue"`
 	}
 
 	collector, err := NewCollector[physicalDisk](slog.New(slog.DiscardHandler), CounterTypeRaw, "PhysicalDisk", []string{"_Total"})
@@ -437,36 +461,45 @@ func TestFrequencyOfAverageTimer(t *testing.T) {
 
 	t.Cleanup(collector.Close)
 
-	frequency, ok := collector.Frequency("Avg. Disk sec/Read")
-	require.True(t, ok)
-	require.Positive(t, frequency)
-
-	for _, name := range []string{"Disk Reads/sec", "Current Disk Queue Length", "unknown"} {
-		_, ok = collector.Frequency(name)
-		require.False(t, ok, name)
-	}
-
-	// As if GetCounterTimeBase failed at startup.
 	idx := slices.IndexFunc(collector.counters, func(counter Counter) bool { return counter.Name == "Avg. Disk sec/Read" })
 	require.NotEqual(t, -1, idx)
 
-	collector.counters[idx].Frequency = 0
+	frequency := collector.counters[idx].Frequency
+	require.Positive(t, frequency)
 
-	_, ok = collector.Frequency("Avg. Disk sec/Read")
-	require.False(t, ok)
+	// As if GetCounterTimeBase failed at startup.
+	collector.counters[idx].Frequency = 0
 
 	var dst []physicalDisk
 
 	require.NoError(t, collector.Collect(&dst))
+	require.Equal(t, frequency, collector.counters[idx].Frequency)
+	require.Len(t, dst, 1)
+	require.GreaterOrEqual(t, dst[0].AvgDiskSecPerRead, 0.0)
+}
 
-	reloaded, ok := collector.Frequency("Avg. Disk sec/Read")
-	require.True(t, ok)
-	require.Equal(t, frequency, reloaded)
+// TestSetRawValueAverageTimer checks the conversion of an average timer and
+// that a counter without a time base has no valid value.
+func TestSetRawValueAverageTimer(t *testing.T) {
+	t.Parallel()
 
-	var nilCollector *Collector[physicalDisk]
+	c := newRowSetTestCollector(false)
 
-	_, ok = nilCollector.Frequency("Avg. Disk sec/Read")
-	require.False(t, ok)
+	var values rowSetValues
+
+	counter := &Counter{
+		Type:                  PERF_AVERAGE_TIMER,
+		Frequency:             2_500_000,
+		FieldIndexValue:       int(unsafe.Offsetof(values.B)),
+		FieldIndexSecondValue: int(unsafe.Offsetof(values.BSecond)),
+	}
+
+	require.True(t, c.setRawValue(&values, counter, RawCounter{FirstValue: 5_000_000, SecondValue: 7}))
+	require.InDelta(t, 2.0, values.B, 0)
+	require.InDelta(t, 7.0, values.BSecond, 0)
+
+	counter.Frequency = 0
+	require.False(t, c.setRawValue(&values, counter, RawCounter{FirstValue: 1, SecondValue: 1}))
 }
 
 func TestMetricTypeForCounterType(t *testing.T) {
@@ -474,6 +507,10 @@ func TestMetricTypeForCounterType(t *testing.T) {
 
 	require.Equal(t, prometheus.GaugeValue, metricType(CounterTypeRaw, PERF_COUNTER_RAWCOUNT))
 	require.Equal(t, prometheus.CounterValue, metricType(CounterTypeRaw, PERF_COUNTER_BULK_COUNT))
+	require.Equal(t, prometheus.CounterValue, metricType(CounterTypeRaw, PERF_100NSEC_MULTI_TIMER))
+	require.Equal(t, prometheus.CounterValue, metricType(CounterTypeRaw, PERF_100NSEC_MULTI_TIMER_INV))
+	require.Equal(t, prometheus.CounterValue, metricType(CounterTypeRaw, PERF_AVERAGE_TIMER))
+	require.Equal(t, prometheus.CounterValue, metricType(CounterTypeRaw, PERF_COUNTER_100NS_QUEUELEN_TYPE))
 	require.Equal(t, prometheus.GaugeValue, metricType(CounterTypeRaw, 0xFFFFFFFF))
 	require.Equal(t, prometheus.GaugeValue, metricType(CounterTypeFormatted, PERF_COUNTER_BULK_COUNT))
 }
diff --git a/internal/pdh/const.go b/internal/pdh/const.go
index 36b24268..23fad5c7 100644
--- a/internal/pdh/const.go
+++ b/internal/pdh/const.go
@@ -21,8 +21,11 @@ import "github.com/prometheus/client_golang/prometheus"
 
 // Conversion factors.
 const (
-	TicksToSecondScaleFactor       = 1 / 1e7
-	WindowsEpoch             int64 = 116444736000000000
+	// TicksPerSecond is the number of 100 ns ticks per second. Divide by it
+	// rather than multiply by its reciprocal: the quotient is correctly rounded,
+	// the product is not.
+	TicksPerSecond       = 1e7
+	WindowsEpoch   int64 = 116444736000000000
 )
 
 // Based on https://github.com/leoluk/perflib_exporter/blob/master/collector/mapper.go
@@ -74,21 +77,25 @@ const (
 
 //nolint:gochecknoglobals
 var SupportedCounterTypes = map[uint32]prometheus.ValueType{
-	PERF_COUNTER_RAWCOUNT_HEX:       prometheus.GaugeValue,
-	PERF_COUNTER_LARGE_RAWCOUNT_HEX: prometheus.GaugeValue,
-	PERF_COUNTER_RAWCOUNT:           prometheus.GaugeValue,
-	PERF_COUNTER_LARGE_RAWCOUNT:     prometheus.GaugeValue,
-	PERF_COUNTER_DELTA:              prometheus.CounterValue,
-	PERF_COUNTER_COUNTER:            prometheus.CounterValue,
-	PERF_COUNTER_BULK_COUNT:         prometheus.CounterValue,
-	PERF_RAW_FRACTION:               prometheus.GaugeValue,
-	PERF_LARGE_RAW_FRACTION:         prometheus.GaugeValue,
-	PERF_100NSEC_TIMER:              prometheus.CounterValue,
-	PERF_PRECISION_100NS_TIMER:      prometheus.CounterValue,
-	PERF_SAMPLE_FRACTION:            prometheus.GaugeValue,
-	PERF_100NSEC_TIMER_INV:          prometheus.CounterValue,
-	PERF_ELAPSED_TIME:               prometheus.GaugeValue,
-	PERF_SAMPLE_BASE:                prometheus.GaugeValue,
-	PERF_RAW_BASE:                   prometheus.GaugeValue,
-	PERF_LARGE_RAW_BASE:             prometheus.GaugeValue,
+	PERF_COUNTER_RAWCOUNT_HEX:        prometheus.GaugeValue,
+	PERF_COUNTER_LARGE_RAWCOUNT_HEX:  prometheus.GaugeValue,
+	PERF_COUNTER_RAWCOUNT:            prometheus.GaugeValue,
+	PERF_COUNTER_LARGE_RAWCOUNT:      prometheus.GaugeValue,
+	PERF_COUNTER_DELTA:               prometheus.CounterValue,
+	PERF_COUNTER_COUNTER:             prometheus.CounterValue,
+	PERF_COUNTER_BULK_COUNT:          prometheus.CounterValue,
+	PERF_RAW_FRACTION:                prometheus.GaugeValue,
+	PERF_LARGE_RAW_FRACTION:          prometheus.GaugeValue,
+	PERF_100NSEC_TIMER:               prometheus.CounterValue,
+	PERF_PRECISION_100NS_TIMER:       prometheus.CounterValue,
+	PERF_SAMPLE_FRACTION:             prometheus.GaugeValue,
+	PERF_100NSEC_TIMER_INV:           prometheus.CounterValue,
+	PERF_100NSEC_MULTI_TIMER:         prometheus.CounterValue,
+	PERF_100NSEC_MULTI_TIMER_INV:     prometheus.CounterValue,
+	PERF_ELAPSED_TIME:                prometheus.GaugeValue,
+	PERF_AVERAGE_TIMER:               prometheus.CounterValue,
+	PERF_COUNTER_100NS_QUEUELEN_TYPE: prometheus.CounterValue,
+	PERF_SAMPLE_BASE:                 prometheus.GaugeValue,
+	PERF_RAW_BASE:                    prometheus.GaugeValue,
+	PERF_LARGE_RAW_BASE:              prometheus.GaugeValue,
 }