diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
index 1940ade81..4ed3cb02a 100644
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -25,6 +25,10 @@ on:
       - test/**
       - '!**.md'
 
+defaults:
+  run:
+    shell: bash
+
 concurrency:
   # Terminate all previous runs of the same workflow for pull requests
   group: test-${{ github.head_ref || github.run_id }}
@@ -152,10 +156,34 @@ jobs:
             latest_tag=$(git tag --merged HEAD --sort=-creatordate "v*" | head -n1)
           fi
           postgrest-loadtest-against -k ${{ matrix.kind }} "$TARGET_BRANCH" "$latest_tag"
+
+      - name: Report P50
+        # This step checks whether any red cross indicators (:x:) are present in the step summary.
+        # The loadtest reporter writes them when any of individual steps fails the performance
+        # regression threshold.
+        run: |
+          postgrest-loadtest-report -g ${{ matrix.kind }} -p 50 \
+            | tee "$GITHUB_STEP_SUMMARY" \
+            | grep -v ':x:'
+
+      - name: Report P0
+        if: always()
+        run: |
           postgrest-loadtest-report -g ${{ matrix.kind }} -p 0 >> "$GITHUB_STEP_SUMMARY"
-          postgrest-loadtest-report -g ${{ matrix.kind }} -p 50 >> "$GITHUB_STEP_SUMMARY"
+
+      - name: Report P90
+        if: always()
+        run: |
           postgrest-loadtest-report -g ${{ matrix.kind }} -p 90 >> "$GITHUB_STEP_SUMMARY"
+
+      - name: Report P95
+        if: always()
+        run: |
           postgrest-loadtest-report -g ${{ matrix.kind }} -p 95 >> "$GITHUB_STEP_SUMMARY"
+
+      - name: Report CPU/MEM
+        if: always()
+        run: |
           postgrest-loadtest-report-load -g ${{ matrix.kind }} >> "$GITHUB_STEP_SUMMARY"
 
   flake:
diff --git a/nix/tools/loadtest.nix b/nix/tools/loadtest.nix
index 3553044d5..2130bf1a7 100644
--- a/nix/tools/loadtest.nix
+++ b/nix/tools/loadtest.nix
@@ -274,12 +274,28 @@ let
         import sys
         import pandas as pd
 
+
+        def evaluate_change(df):
+            return ((df['head'] / df['main'] - 1) * 100) \
+              .map(lambda r: "{icon} {ratio:.1f} %".format(
+                ratio=r,
+                # Hardcoded failure threshold for CI is 5% here.
+                icon="" if r < 5 else ":x:"
+              ))
+
+
         pd.read_json(sys.stdin) \
           .rename(columns={'latency': sys.argv[1]}) \
           .set_index(sys.argv[1]) \
           .drop(['branch']) \
           .convert_dtypes() \
-          .to_markdown(sys.stdout, floatfmt='.1f')
+          .assign(change=evaluate_change) \
+          .to_markdown(
+            sys.stdout,
+            floatfmt='.1f',
+            colglobalalign='right',
+            colalign=('left',)
+          )
       '';