diff --git a/docker/telemetry/alloy/config.alloy b/docker/telemetry/alloy/config.alloy
index e5d5f613de..a41493e878 100644
--- a/docker/telemetry/alloy/config.alloy
+++ b/docker/telemetry/alloy/config.alloy
@@ -315,15 +315,14 @@ otelcol.connector.spanmetrics "xrpld" {
     explicit {
       // Buckets MUST stay strictly ascending (the connector binary-searches
       // them and silently misbuckets otherwise) and MUST match the two
-      // otel-collector-config*.yaml lists. Sub-MILLISECOND boundaries exist
-      // because most xrpld spans sit far below 1ms (tx.preflight means
-      // ~0.012ms): a 1ms floor put >99.99% of samples in bucket one and pinned
-      // every p95/p99 at a constant 0.95ms.
-      // otel-collector-config*.yaml lists. Sub-second boundaries cover
-      // RPC/tx/ledger spans; 2s-4s resolve second-scale consensus spans
-      // (consensus.round ~3.9s, consensus.establish ~1.9s) that used to pile
-      // into one 1s-5s bucket; 10s/30s give the ledger.acquire catch-up tail
-      // (~17% exceeds 5s) a measurable home so its p95/p99 stop reading +Inf.
+      // otel-collector-config*.yaml lists. Sub-millisecond boundaries exist
+      // because most xrpld spans sit far below 1ms: a 1ms floor puts nearly
+      // every sample in bucket one and pins p95/p99 to that bucket's edge.
+      // Sub-second boundaries cover RPC, tx and ledger spans. The 2s-4s edges
+      // resolve second-scale consensus spans, and 10s/30s give the
+      // ledger.acquire catch-up tail a home so its p95/p99 stop reading +Inf.
+      // Each edge was chosen from a measured duration distribution for the
+      // spans it covers.
       buckets = ["0.01ms", "0.05ms", "0.1ms", "0.25ms", "0.5ms", "1ms", "5ms", "10ms", "25ms", "50ms", "100ms", "250ms", "500ms", "1s", "2s", "3s", "4s", "5s", "10s", "30s"]
     }
   }
diff --git a/docker/telemetry/otel-collector-config.grafanacloud.yaml b/docker/telemetry/otel-collector-config.grafanacloud.yaml
index 01722e383a..4b488a84cd 100644
--- a/docker/telemetry/otel-collector-config.grafanacloud.yaml
+++ b/docker/telemetry/otel-collector-config.grafanacloud.yaml
@@ -78,10 +78,13 @@ processors:
   batch:
     timeout: 1s
     send_batch_size: 100
-  # Box-only (uncommitted): probabilistic 0.5% tail sampling on the
-  # trace-storage branch only. spanmetrics-derived metrics see 100% of spans
-  # (they run on a separate branch) so RED remain exact; Tempo/Cloud stores
-  # ~1/200 of traces to keep long-running node cost bounded.
+  # Probabilistic tail sampling, on the trace-storage branch only. The
+  # spanmetrics branch sees every span, so span-derived RED metrics stay
+  # exact while stored trace volume is cut to bound cost on a long-running
+  # node. Operators who need full traces raise sampling_percentage to 100.
+  # Note: a log line carries a trace id only when its span was sampled at
+  # the SDK, so sampling here drops traces whose ids are already in the
+  # logs, and log-to-trace links for those resolve to nothing.
   tail_sampling:
     decision_wait: 10s
     num_traces: 50000
@@ -177,16 +180,15 @@ connectors:
       unit: ms
       explicit:
         # Buckets MUST stay strictly ascending (the connector binary-searches
-        # them and silently misbuckets otherwise). Sub-MILLISECOND boundaries
-        # (0.01-0.5ms) exist because most xrpld spans are far below 1ms --
-        # tx.preflight means ~0.012ms, tx.preclaim ~0.15ms -- so a 1ms floor
-        # put >99.99% of samples in the first bucket and pinned every p95/p99
-        # at a constant 0.95ms. Sub-second boundaries cover
-        # RPC/tx/ledger spans; the 2s-4s boundaries resolve second-scale
-        # consensus spans (consensus.round ~3.9s, consensus.establish ~1.9s)
-        # that used to pile into a single 1s-5s bucket; 10s/30s give the
-        # ledger.acquire tail (~17% exceeds 5s during catch-up) a measurable
-        # home so its p95/p99 stop reading as +Inf.
+        # them and silently misbuckets otherwise) and MUST match the Alloy
+        # list and the other collector config. Sub-millisecond boundaries
+        # exist because most xrpld spans sit far below 1ms: a 1ms floor puts
+        # nearly every sample in the first bucket and pins p95/p99 to that
+        # bucket's edge. Sub-second boundaries cover RPC, tx and ledger spans.
+        # The 2s-4s edges resolve second-scale consensus spans, and 10s/30s
+        # give the ledger.acquire catch-up tail a home so its p95/p99 stop
+        # reading as +Inf. Each edge was chosen from a measured duration
+        # distribution for the spans it covers.
         buckets:
           [
             0.01ms,
diff --git a/docker/telemetry/otel-collector-config.yaml b/docker/telemetry/otel-collector-config.yaml
index 093d3ddda1..e3d1850fcd 100644
--- a/docker/telemetry/otel-collector-config.yaml
+++ b/docker/telemetry/otel-collector-config.yaml
@@ -168,16 +168,15 @@ connectors:
       unit: ms
       explicit:
         # Buckets MUST stay strictly ascending (the connector binary-searches
-        # them and silently misbuckets otherwise). Sub-MILLISECOND boundaries
-        # (0.01-0.5ms) exist because most xrpld spans are far below 1ms --
-        # tx.preflight means ~0.012ms, tx.preclaim ~0.15ms -- so a 1ms floor
-        # put >99.99% of samples in the first bucket and pinned every p95/p99
-        # at a constant 0.95ms. Sub-second boundaries cover
-        # RPC/tx/ledger spans; the 2s-4s boundaries resolve second-scale
-        # consensus spans (consensus.round ~3.9s, consensus.establish ~1.9s)
-        # that used to pile into a single 1s-5s bucket; 10s/30s give the
-        # ledger.acquire tail (~17% exceeds 5s during catch-up) a measurable
-        # home so its p95/p99 stop reading as +Inf.
+        # them and silently misbuckets otherwise) and MUST match the Alloy
+        # list and the other collector config. Sub-millisecond boundaries
+        # exist because most xrpld spans sit far below 1ms: a 1ms floor puts
+        # nearly every sample in the first bucket and pins p95/p99 to that
+        # bucket's edge. Sub-second boundaries cover RPC, tx and ledger spans.
+        # The 2s-4s edges resolve second-scale consensus spans, and 10s/30s
+        # give the ledger.acquire catch-up tail a home so its p95/p99 stop
+        # reading as +Inf. Each edge was chosen from a measured duration
+        # distribution for the spans it covers.
         buckets:
           [
             0.01ms,