diff --git a/TODO.md b/TODO.md
index 88f10bf..a96d6cc 100644
--- a/TODO.md
+++ b/TODO.md
@@ -151,6 +151,23 @@ and the p90 tail fell from ~470ms to ~330ms. The proxy was ruled out early
       run can also finish before the client JVM has a session to flush, and
       then only the host profile is collected, with nothing said about the
       missing one.
+- [ ] **Game-summary images cost ~18% of the EDT's working time.** Measured on
+      the first spectated profile (2026-08-07): `BoardView$1.gamePhaseChange`
+      calls `ImageIO.write` on every phase change, on the EDT, because we set
+      `GameSummaryBoardView` in `megamek/clientsettings.xml.template`. That is
+      1541 of 8383 running EDT samples going into PNGs nobody has looked at -
+      and the sibling `GameSummaryMinimap` is the one already doubted above,
+      since `MinimapEnabled` is false.
+
+      It is also a second PNG encoder: the summary goes through the JDK's
+      `com.sun.imageio.plugins.png.PNGImageWriter`, not Suramadu's native ZPNG,
+      so the two compete for the same EDT.
+
+      Caveat before acting: a bot-vs-bot match changes phase far more often per
+      minute than a human one, so 18% is the fast end. The fix is to decide
+      whether these images are wanted at all - `container/exit/finalize.sh`
+      collects them - and if they are, whether they can be written off the EDT.
+
 - [ ] **Frame tearing.** A white or half-drawn band at the board's top edge,
       phase-independent, survives patches 0006 and 0007 (0006 confirmed
       effective on the 2026-08-05 retest). MegaMek's paint paths audit clean:
diff --git a/tests/profile-summary.py b/tests/profile-summary.py
index 8125ff8..a8c100b 100755
--- a/tests/profile-summary.py
+++ b/tests/profile-summary.py
@@ -27,8 +27,30 @@ from collections import Counter
 # matches, because they nest - an encode happens inside a paint.
 BUCKETS = (
     ("paint", ("BoardView", "paintComponent", "RepaintManager")),
-    ("encode+hash", ("ZPNG", "libzpng", "EncoderBase", "xxhash")),
+    # Everything that turns pixels into PNG bytes, including zlib's own
+    # internals. Naming only the entry points undercounted this by nearly 3x:
+    # the encoders hand off to deflate and the samples land in longest_match
+    # and compress_block, which named neither ZPNG nor PNG.
+    (
+        "png+deflate",
+        (
+            "ZPNG",
+            "libzpng",
+            "EncoderBase",
+            "PNGImageWriter",
+            "ImageIO",
+            "deflate",
+            "longest_match",
+            "compress_block",
+            "slide_hash",
+        ),
+    ),
+    # A subset of the above, called out because it is not frame traffic at all:
+    # MegaMek's GameSummaryBoardView preference writes a board PNG through
+    # ImageIO on every phase change, on the EDT. Nobody has ever looked at one.
+    ("  of which game-summary images", ("ImageIO", "PNGImageWriter")),
     ("directdraw", ("ImageConst", "DrawConstantPool", "DrawInstructionFactory")),
+    ("hash", ("xxhash", "Coords.hashCode")),
     ("gc/alloc", ("GCTaskThread", "G1", "CollectedHeap")),
 )
 
diff --git a/tests/shell/test-profile-summary.sh b/tests/shell/test-profile-summary.sh
index 43835b4..73feedc 100755
--- a/tests/shell/test-profile-summary.sh
+++ b/tests/shell/test-profile-summary.sh
@@ -40,8 +40,8 @@ check "  and the rest is working" grep -q "100    running" <<<"$out"
 # The buckets nest: an encode happens inside a paint, so 100% paint and 70%
 # encode is the correct reading of this fixture, not a double count.
 check "  everything running was under paint" grep -q "100.0%  paint" <<<"$out"
-check "  and the encode share is of the working samples" \
-  grep -q "70.0%  encode+hash" <<<"$out"
+check "  and the png share is of the working samples" \
+  grep -q "70.0%  png+deflate" <<<"$out"
 check "  a thread that is not the EDT is left out" \
   bash -c '! grep -q "ThreadPoolExecutor" <<<"$1"' _ "$out"
 
@@ -52,6 +52,14 @@ check "--thread '' counts every thread" \
   bash -c 'python3 "$1/tests/profile-summary.py" --thread "" "$2" | grep -q "1,400    waiting"' \
   _ "$ROOT" "$TMP/p.collapsed"
 
+# zlib internals name neither ZPNG nor PNG, and missing them undercounted the
+# encode share by nearly 3x on the first real profile.
+printf 'java/awt/EventDispatchThread.run;megamek/BoardView.drawHex;deflate_slow;longest_match 40\n' \
+  >> "$TMP/p.collapsed"
+check "  zlib internals count as encode, not as something else" \
+  bash -c 'python3 "$1/tests/profile-summary.py" "$2" | grep -q "78.6%  png+deflate"' \
+  _ "$ROOT" "$TMP/p.collapsed"
+
 # A profile with nothing running says so rather than dividing by zero.
 printf 'java/awt/EventDispatchThread.run;java/awt/EventQueue.getNextEvent 10\n' \
   > "$TMP/idle.collapsed"