diff --git a/cmd/prometheus/main.go b/cmd/prometheus/main.go
index 5c004d660..0be3dfbc4 100644
--- a/cmd/prometheus/main.go
+++ b/cmd/prometheus/main.go
@@ -239,6 +239,9 @@ func main() {
 	a.Flag("rules.alert.resend-delay", "Minimum amount of time to wait before resending an alert to Alertmanager.").
 		Default("1m").SetValue(&cfg.resendDelay)
 
+	a.Flag("scrape.timestamp-tolerance", "Tolerance applied to scrapes timestamp to improve timestamp compression. Experimental.").
+		Hidden().Default("2ms").DurationVar(&scrape.ScrapeTimestampTolerance)
+
 	a.Flag("alertmanager.notification-queue-capacity", "The capacity of the queue for pending Alertmanager notifications.").
 		Default("10000").IntVar(&cfg.notifier.QueueCapacity)
 
diff --git a/scrape/scrape.go b/scrape/scrape.go
index de248d876..53d573fcc 100644
--- a/scrape/scrape.go
+++ b/scrape/scrape.go
@@ -50,7 +50,7 @@ import (
 // Temporary tolerance for scrape appends timestamps alignment, to enable better
 // compression at the TSDB level.
 // See https://github.com/prometheus/prometheus/issues/7846
-const scrapeTimestampTolerance = 2 * time.Millisecond
+var ScrapeTimestampTolerance = 2 * time.Millisecond
 
 var errNameLabelMandatory = fmt.Errorf("missing metric name (%s label)", labels.MetricName)
 
@@ -1009,14 +1009,14 @@ mainLoop:
 		// increase in TSDB.
 		// See https://github.com/prometheus/prometheus/issues/7846
 		scrapeTime := time.Now()
-		if interval > 100*scrapeTimestampTolerance {
+		if interval > 100*ScrapeTimestampTolerance {
 			// For some reason, a tick might have been skipped, in which case we
 			// would call alignedScrapeTime.Add(interval) multiple times.
 			for scrapeTime.Sub(alignedScrapeTime) >= interval {
 				alignedScrapeTime = alignedScrapeTime.Add(interval)
 			}
 			// Align the scrape time if we are in the tolerance boundaries.
-			if scrapeTime.Sub(alignedScrapeTime) <= scrapeTimestampTolerance {
+			if scrapeTime.Sub(alignedScrapeTime) <= ScrapeTimestampTolerance {
 				scrapeTime = alignedScrapeTime
 			}
 		}