diff --git a/bin/result-processor.py b/bin/result-processor.py
index 68114c9..4dd103d 100755
--- a/bin/result-processor.py
+++ b/bin/result-processor.py
@@ -284,7 +284,7 @@ def validate_archive(archive):
     return 0
 
 
-def log_result_directory(result_directory, result_status):
+def log_result_directory(result_directory, result_status, data=None):
     if myglobal.args.mode == "completion":
         # completion mode uses a very terse output data to feed into a
         # bash tab completion engine
@@ -303,6 +303,10 @@ def log_result_directory(result_directory, result_status):
 
         myglobal.log.info("status: %s" % (result_status))
 
+        if data is not None and data.get('partial'):
+            dropped_engines = data.get('dropped-engines', [])
+            myglobal.log.info("partial: yes (%d engine(s) dropped)" % (len(dropped_engines)))
+
     return 0
 
 
@@ -420,7 +424,7 @@ def ls_result_directory(result_directory):
             myglobal.log.debug("result directory '%s' has no rickshaw-run" % (result_directory))
             return 0
 
-    log_result_directory(result_directory, status)
+    log_result_directory(result_directory, status, data)
 
     if myglobal.args.type == "short":
         pass
@@ -691,7 +695,7 @@ def run_results_tag_mode():
 
     data, status = load_rickshaw_run(run_dir)
 
-    log_result_directory(run_dir, status)
+    log_result_directory(run_dir, status, data)
 
     if data is not None:
         if myglobal.args.action == "ls":
diff --git a/docs/how-benchmark-execution-works.md b/docs/how-benchmark-execution-works.md
index d514775..5ee2876 100644
--- a/docs/how-benchmark-execution-works.md
+++ b/docs/how-benchmark-execution-works.md
@@ -480,6 +480,37 @@ no failures is considered passing.
 An iteration is considered failed if it doesn't have enough
 passing samples to meet the `num-samples` requirement.
 
+### Partial runs and dropped engines
+
+A roadblock timeout anywhere in the run — during the per-sample
+benchmark execution phases (`infra-start`, `server-start`,
+`endpoint-start`, `client-start`, `client-stop`, `endpoint-stop`,
+`server-stop`, `infra-stop`) or the post-benchmark phases
+(stopping tools, sending data, endpoint cleanup) — causes
+rickshaw to drop the missing engine(s) from the active follower
+list, e.g. because one crashed (OOM kill) or lost its SSH
+connection. What happens next depends on where the drop occurs:
+a timeout during the per-sample phases aborts all remaining
+samples and iterations for the rest of the run, while a timeout
+during the post-benchmark phases only removes the missing
+engine(s) and lets the run finish normally with the remaining
+engines. Either way, the run can complete with fewer engines
+than it started with, and without this feature that fact would
+otherwise go unrecorded.
+
+When this happens, rickshaw logs a warning for each dropped
+engine (naming the engine and the roadblock it missed) and an
+end-of-run participation summary. The run's `rickshaw-run.json`
+records `partial: true` and a `dropped-engines` list (each entry
+has the engine ID and the roadblock it was dropped at); `crucible
+ls`/`crucible get result` surface this as a `partial: yes (N
+engine(s) dropped)` line alongside the normal `status` line. CDM
+run documents (v9dev+) carry the same `partial`/`dropped-engines`
+fields, so query results can be filtered or flagged accordingly.
+Treat results from a partial run with caution — data from the
+dropped engine(s) is missing or incomplete for the affected
+phases.
+
 ## Re-processing
 
 ### crucible postprocess
diff --git a/docs/how-cdm-works.md b/docs/how-cdm-works.md
index 0f01ab8..a8d32cd 100644
--- a/docs/how-cdm-works.md
+++ b/docs/how-cdm-works.md
@@ -41,7 +41,7 @@ run
 
 | Type | Purpose | Key fields |
 |------|---------|-----------|
-| **run** | A single benchmark execution | run UUID, begin/end times, benchmark name, user, host |
+| **run** | A single benchmark execution | run UUID, begin/end times, benchmark name, user, host, partial, dropped-engines |
 | **iteration** | One parameter combination | iteration UUID, primary metric, primary period, status |
 | **sample** | One execution of an iteration | sample UUID, status (pass/fail) |
 | **period** | A time window within a sample | period UUID, name, begin/end timestamps |
@@ -413,7 +413,7 @@ CDM has evolved through several versions:
 |---------|------------|
 | v7dev | Original version, generic `id` fields |
 | v8dev | Document-specific UUID fields (`iteration-uuid`, `period-uuid`, etc.) |
-| v9dev | Per-month index naming (`cdm-v9dev-metric_data@2026.06`), `metric_def` document type |
+| v9dev | Per-month index naming (`cdm-v9dev-metric_data@2026.06`), `metric_def` document type, `partial`/`dropped-engines` fields on the run doc |
 | v10dev | `default-aggregation` field on metric_desc, per-metric aggregation control (sum/avg/max/min) |
 
 Multiple CDM versions can coexist in the same OpenSearch