docs(bench): name the app settings an in-app row has to match

CHANGELOG.md

@@ -101,6 +101,11 @@ Semantic Versioning.
   so a host row was quietly running a different configuration from the app it is compared against.
   Override with `UBATCH=`; `UBATCH=0` restores the old full-width behaviour. The five maintainer
   rows in the table predate this and are marked as such.
+- **The app path says which settings to change.** An in-app row was being invited without saying
+  that the app ships cold-expert dropping at 75 %, a lossy speedup the CLI protocol does not use, so
+  a telemetry screenshot was not the configuration it was being compared against. The page now names
+  the three settings to match (dropping off, cache auto, context 2048) and says that a phone row
+  cannot fill the storage-rate column.
 - **Two platform limits written down** in [`limitations.md`](docs/limitations.md). macOS has no
   `O_DIRECT` and the engine does not call the `F_NOCACHE` equivalent, so expert reads there go
   through the page cache — while the metrics still say `o_direct=1`, because that field records the

docs/community-benchmarks.md

@@ -37,10 +37,15 @@ engine actually issues, runs the protocol, prints two markdown tables. Paste bot
 ### Android phone
 
 - **Easy path:** install the release APK, load a model from its catalog, generate at least 256
-  tokens, screenshot the telemetry panel. That is a valid row.
+  tokens, screenshot the telemetry panel. Change three settings first, or the row is measuring
+  something else: **cold-expert dropping Off** (it defaults to 75 %, which is a lossy speedup),
+  **cache Auto**, **context 2048**. Everything else already matches the protocol.
 - **Exact path** (comparable with the PC rows): push `bmoe-cli` and run it over adb, see
   [On a phone](#on-a-phone).
 
+Either way the app cannot measure your drive's raw read rate, so a phone row leaves that column
+empty unless you run the storage probe yourself.
+
 ### Apple hardware
 
 | | Status |
@@ -90,10 +95,15 @@ adb shell /data/local/tmp/bmoe-cli -m /data/local/tmp/MODEL.gguf --chatml -n 256
   -p "Write a long detailed essay about the history of computing including its origins its key milestones the people involved and the future directions of the field"
 ```
 
+This command is already the protocol: no dropping, cache auto, the default 2048 context. Three
+things to get right around it:
+
 - Stage the model on `/data/local/tmp` or `/sdcard/Download`. **Not** the app's external files dir
   under `/storage/emulated`: it is FUSE-backed and `O_DIRECT` silently falls back to buffered.
 - Check `o_direct=1` in the telemetry before trusting the number.
 - Phones throttle: run twice, report the second, say if the two differ by more than a few percent.
+- The shell runs under a lower CPU frequency cap than a foreground app on some devices, so an adb
+  row can read a little slower than the same configuration in the app. Say which one you used.
 
 ### Models that line up with the README