bench: consolidate stream throughput columns

.changeset/calm-geese-share.md

@@ -0,0 +1,4 @@
+---
+---
+
+Remove low-value throughput columns from the CI stream benchmark report.

.github/scripts/render-benchmark-comment.mjs

@@ -787,9 +787,8 @@ function pairedSortKey(a, b) {
 
 /**
  * The stream-scenario table: one row per stream scenario, with the columns
- * streams actually want — writer-side achieved and reader-side delivered
- * sustained rates (steady window: first/last 10% of chunks trimmed), CRTT
- * percentiles, and the median worst delivery stall (CDV max positive).
+ * streams actually want — CRTT percentiles and the median worst delivery
+ * stall (CDV max positive).
  * Deltas vs main are plain percentages; deliberately NO 🔴/🟢 marks — targets
  * attach in a later PR once a baseline exists. Rates read higher-is-better,
  * latencies lower-is-better, so directional marks would need per-column
@@ -813,14 +812,14 @@ function renderStreamTable(result) {
   const lines = [
     '**Streams**',
     '',
-    '| Scenario | wr c/s | rd c/s | wr KiB/s | rd KiB/s | CRTT 1st | p75 | p90 | p99 | CDV max | iters |',
-    '|----------|-------:|-------:|---------:|---------:|---------:|----:|----:|----:|--------:|------:|',
+    '| Scenario | CRTT 1st | p75 | p90 | p99 | CDV max | iters |',
+    '|----------|---------:|----:|----:|----:|--------:|------:|',
   ];
   for (const row of rows) {
     const s = row.stream;
     const b = row.baselineStream ?? {};
     lines.push(
-      `| ${row.scenario} | ${cell(s.wrCps, b.wrCps)} | ${cell(s.rdCps, b.rdCps)} | ${cell(s.wrKiBps, b.wrKiBps)} | ${cell(s.rdKiBps, b.rdKiBps)} | ${cell(s.firstMs, b.firstMs)} | ${cell(row.p75, row.baselineP75)} | ${cell(row.p90, row.baselineP90)} | ${cell(row.p99, row.baselineP99)} | ${cell(s.cdvMaxMs, b.cdvMaxMs)} | ${s.iterations} |`
+      `| ${row.scenario} | ${cell(s.firstMs, b.firstMs)} | ${cell(row.p75, row.baselineP75)} | ${cell(row.p90, row.baselineP90)} | ${cell(row.p99, row.baselineP99)} | ${cell(s.cdvMaxMs, b.cdvMaxMs)} | ${s.iterations} |`
     );
   }
   return lines.join('\n');
@@ -995,7 +994,7 @@ function renderFooter(entries) {
   const smallprint = [
     ...(hasStreamTable
       ? [
-          '<sub>**Streams**: writer/reader sustained rates (steady window, 10% trimmed each side), first-chunk RTT (the stream-open path, before any buffering/backpressure), CRTT percentiles, and worst delivery stall (CDV max). Cells are medians across iterations; per-run values in the artifacts. No \ud83d\udd34/\ud83d\udfe2 marks until targets attach.</sub>',
+          '<sub>**Streams**: first-chunk RTT (the stream-open path, before any buffering/backpressure), CRTT percentiles, and worst delivery stall (CDV max). Cells are medians across iterations; per-run values in the artifacts. No \ud83d\udd34/\ud83d\udfe2 marks until targets attach.</sub>',
           '',
         ]
       : []),

.github/scripts/render-benchmark-comment.test.js

@@ -706,7 +706,7 @@ function crttResult({ avg = 120, hist }) {
   });
 }
 
-test('renders stream scenarios in their own table with rate columns', async () => {
+test('renders stream scenarios in their own table without rate columns', async () => {
   const { renderComment, extractHistory } = await loadModule();
   const hist = crttHist([
     [59, 1400],
@@ -728,18 +728,12 @@ test('renders stream scenarios in their own table with rate columns', async () =
   assert.doesNotMatch(body, /\| \*\*stream\*\* \|/);
   assert.match(
     body,
-    /\| Scenario \| wr c\/s \| rd c\/s \| wr KiB\/s \| rd KiB\/s \| CRTT 1st \| p75 \| p90 \| p99 \| CDV max \| iters \|/
-  );
-  // Rate cells with plain vs-main deltas, latency cells from percentile
-  // baselines, and NO red/green marks anywhere in the stream table.
-  assert.match(
-    body,
-    /\| chunk RTT \(llm\) \| 100 \(\u00b10%\) \| 99\.4 \(\+10%\) \| 6\.1 \(\u00b10%\) \|/
-  );
-  assert.match(
-    body,
-    /\| replay eve-test \(2x\) \| 297 \(\u00b10%\) \| 288 \(\u00b10%\) \| 742 \(\u00b10%\) \| 719 \(\u00b10%\) \|/
+    /\| Scenario \| CRTT 1st \| p75 \| p90 \| p99 \| CDV max \| iters \|/
   );
+  // Latency cells retain their vs-main deltas, and the stream table has no
+  // red/green marks.
+  assert.match(body, /\| chunk RTT \(llm\) \| 96 \(\u00b10%\) \|/);
+  assert.match(body, /\| replay eve-test \(2x\) \| 118 \(\u00b10%\) \|/);
   assert.match(body, /\| 141 \(\u00b10%\) \| 10 \|/);
   const streamsSection = body.slice(
     body.indexOf('**Streams**'),
@@ -763,7 +757,7 @@ test('renders stream scenarios in their own table with rate columns', async () =
   // the internal 'stream' id never leaks into it.
   assert.match(body, /\*\*CRTT\*\*: chunk round-trip time/);
   assert.match(body, /\*\*CDV\*\*: chunk delay variation/);
-  assert.match(body, /\*\*Streams\*\*: writer\/reader sustained rates/);
+  assert.match(body, /\*\*Streams\*\*: first-chunk RTT/);
   // History block: per-run arrays and sparkline payloads stripped, medians
   // and baseline annotations kept.
   const history = extractHistory(body);
@@ -780,7 +774,7 @@ test('renders stream scenarios in their own table with rate columns', async () =
     history,
     commit: 'ffffff1234567890',
   });
-  assert.match(rerendered, /\| chunk RTT \(llm\) \| 100/);
+  assert.match(rerendered, /\| chunk RTT \(llm\) \| 96/);
   assert.doesNotMatch(rerendered, /CRTT drill-down/);
 });
 
@@ -794,7 +788,7 @@ test('renders the stream table without deltas when main has no baseline', async
   });
   assert.match(
     body,
-    /\| chunk RTT \(llm\) \| 100 \| 99\.4 \| 6\.1 \| 6 \| 96 \| 188 \| 438 \| 1229 \| 141 \| 10 \|/
+    /\| chunk RTT \(llm\) \| 96 \| 188 \| 438 \| 1229 \| 141 \| 10 \|/
   );
   assert.doesNotMatch(body, /%\)/);
   assert.match(body, /No `main` baseline yet/);