diff --git a/docs/spectator/core/meters/counter.md b/docs/spectator/core/meters/counter.md
index 02696fcc..ae04b2a9 100644
--- a/docs/spectator/core/meters/counter.md
+++ b/docs/spectator/core/meters/counter.md
@@ -10,10 +10,7 @@ to reason about the measurement and allows for aggregating the counter across in
 In Atlas, the `:per-step` operator can be used to convert them back into a count-per-step on a
 graph.
 
-!!! Note
-    For high performance code, such as incrementing in a tight loop that lasts less than a
-    reporting interval, increment a local variable and add the final value to the counter after
-    the loop has completed.
+For high-volume use, see [Performance Tips](../performance.md).
 
 ## Languages
 
diff --git a/docs/spectator/core/performance.md b/docs/spectator/core/performance.md
new file mode 100644
index 00000000..673f77a9
--- /dev/null
+++ b/docs/spectator/core/performance.md
@@ -0,0 +1,95 @@
+# Performance Tips
+
+A few patterns are worth keeping in mind when instrumenting hot paths. These apply across
+all Spectator client libraries; language-specific helpers are noted where they exist.
+
+## Cache the meter reference
+
+`registry.counter("server.numRequests")` (and equivalents for timer, gauge, etc.) performs
+a lookup in the registry on every call. For meters that don't have dynamic tag values,
+look the meter up once and reuse the reference:
+
+```java
+// Good: lookup happens once.
+private final Counter requests = registry.counter("server.numRequests");
+
+public void handle() {
+    requests.increment();
+}
+```
+
+```java
+// Avoid in hot paths: lookup on every call.
+public void handle() {
+    registry.counter("server.numRequests").increment();
+}
+```
+
+For meters where some tag values are dynamic (e.g. a status code), cache the base `Id` and
+derive per-request meters with `with_tag` / `withTag`:
+
+```java
+private final Id requestsId = registry.createId("server.numRequests");
+
+public void handle(Response res) {
+    registry.counter(requestsId.withTag("status", res.statusCode())).increment();
+}
+```
+
+## Avoid instrumentation in tight loops
+
+If you need to update a meter inside a tight loop where each iteration is cheap, the
+instrumentation overhead can dominate. Accumulate locally and apply the delta once:
+
+```java
+long localCount = 0;
+for (Item item : items) {
+    if (item.isFoo()) localCount++;
+}
+requests.increment(localCount);
+```
+
+Java's [BatchUpdater](#java-batchupdater) automates this pattern for `Counter`,
+`Timer`, and `DistributionSummary`.
+
+## Prefer basic meters over percentile variants
+
+[Percentile Timer](meters/timer.md#percentile-timer) and
+[Percentile Distribution Summary](meters/dist-summary.md#percentile-distribution-summary)
+maintain a set of bucket counters. Storage cost is up to ~300x that of the basic Timer or
+Distribution Summary. Use them only for one or two key indicators per application, set an
+appropriate range, and keep tag cardinality bounded.
+
+## Keep tag cardinality bounded
+
+Every distinct combination of tag values produces a new time series. Avoid putting
+high-cardinality values in tags — user IDs, request IDs, raw paths, etc. Use the
+[Cardinality Limiter](../lang/java/patterns/cardinality-limiter.md) (Java) or apply a
+similar mapping in other languages to cap the value set for any tag that could grow
+unbounded.
+
+## Java: BatchUpdater
+
+For very high-volume updates within a single thread, `Counter`, `Timer`, and
+`DistributionSummary` expose a `batchUpdater(batchSize)` method that buffers updates and
+flushes them as a single operation. Trade-off: updates are delayed by up to `batchSize`
+events before they appear on the underlying meter.
+
+```java
+try (Counter.BatchUpdater updater = requests.batchUpdater(1000)) {
+    for (Item item : items) {
+        process(item);
+        updater.increment();
+    }
+}
+```
+
+The updater is `AutoCloseable`; the try-with-resources block guarantees a final flush. See
+the [Counter](../lang/java/meters/counter.md), [Timer](../lang/java/meters/timer.md), and
+[Distribution Summary](../lang/java/meters/dist-summary.md) pages for the per-meter
+signatures.
+
+The spectatord-backed clients (C++, Go, Node.js, Python) do not need an equivalent helper:
+they buffer protocol lines client-side, and spectatord aggregates updates from many sources
+before flushing to Atlas. Each language's usage page has a Line Buffer or Buffers section
+for the relevant tuning knobs.
diff --git a/docs/spectator/lang/java/meters/counter.md b/docs/spectator/lang/java/meters/counter.md
index e2146f9c..ab47eaa1 100644
--- a/docs/spectator/lang/java/meters/counter.md
+++ b/docs/spectator/lang/java/meters/counter.md
@@ -48,3 +48,21 @@ events happen together.
   }
 }
 ```
+
+## Batch Updates
+
+For very high-volume updates within a single thread, `Counter` exposes a `batchUpdater` that
+buffers updates and flushes them as a single operation. The trade-off is that updates are
+delayed by up to `batchSize` events before they appear on the underlying counter.
+
+```java
+try (Counter.BatchUpdater updater = insertCounter.batchUpdater(1000)) {
+  for (Object obj : objs) {
+    impl.insert(obj);
+    updater.increment();
+  }
+}
+```
+
+The updater is `AutoCloseable`; the try-with-resources block guarantees a final flush. See
+[Performance Tips](../../../core/performance.md) for the cross-cutting guidance.
diff --git a/docs/spectator/lang/java/meters/dist-summary.md b/docs/spectator/lang/java/meters/dist-summary.md
index 33d0f327..062ed3c7 100644
--- a/docs/spectator/lang/java/meters/dist-summary.md
+++ b/docs/spectator/lang/java/meters/dist-summary.md
@@ -26,10 +26,17 @@ Then call record when an event occurs:
 ```
 **Note:** If the amount recorded is less than 0 the value will be dropped.
 
+## Batch Updates
+
+For very high-volume updates within a single thread, `DistributionSummary` exposes a
+`batchUpdater` that buffers updates and flushes them as a single operation. See
+[Performance Tips](../../../core/performance.md) for the cross-cutting guidance and
+[`Counter.batchUpdater`](counter.md#batch-updates) for an annotated example.
+
 ## Percentile Distribution Summary
 
-concept and storage-cost caveats. Use a regular [Distribution Summary](dist-summary.md) unless
-percentile estimates are required.
+Use a regular [Distribution Summary](#distribution-summary) unless percentile estimates are
+required.
 
 Create one via the static builder:
 
diff --git a/docs/spectator/lang/java/meters/timer.md b/docs/spectator/lang/java/meters/timer.md
index f71fa72c..db454f8a 100644
--- a/docs/spectator/lang/java/meters/timer.md
+++ b/docs/spectator/lang/java/meters/timer.md
@@ -105,11 +105,18 @@ backend, the unit will be seconds.
 [System.nanoTime()]: https://docs.oracle.com/en/java/javase/25/docs/api/java.base/java/lang/System.html#nanoTime()
 [System.currentTimeMillis()]: https://docs.oracle.com/en/java/javase/25/docs/api/java.base/java/lang/System.html#currentTimeMillis()
 
+## Batch Updates
+
+For very high-volume updates within a single thread, `Timer` exposes a `batchUpdater` that
+buffers updates and flushes them as a single operation. See
+[Performance Tips](../../../core/performance.md) for the cross-cutting guidance and
+[`Counter.batchUpdater`](counter.md#batch-updates) for an annotated example.
+
 ## Percentile Timer
 
 **Note**: Percentile timers generate a metric per bucket in the histogram. Create instances
 once per ID and reuse them as needed. Avoid adding tags with high cardinality as that increases
-the cardinality of the metric. If at all possible, use a [Timer](timer.md) instead.
+the cardinality of the metric. If at all possible, use a [Timer](#timer) instead.
 
 To get started, create an instance using the [Registry](../registry/overview.md):
 
diff --git a/mkdocs.yml b/mkdocs.yml
index c40e28c2..115a7a91 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -224,6 +224,7 @@ nav:
       - Gauge: spectator/core/meters/gauge.md
       - Max Gauge: spectator/core/meters/max-gauge.md
       - Timer: spectator/core/meters/timer.md
+    - Performance Tips: spectator/core/performance.md
   - Patterns:
     - Age Gauge: spectator/patterns/age-gauge.md
     - Monotonic Counter: spectator/patterns/monotonic-counter.md