site: contributors column, star refresh, and two chart fixes

- Roster: replace the near-constant Class column with a Contributors column that
  credits outside submitters/maintainers (@moonming on apisix/aisix, @geekspeng on
  higress once it lands). Data lives in a SITE-ONLY site/contributors.json, never in
  definition.json: a manifest schema change would perturb the frozen ENGINE_PIN and
  force a full 14-gateway re-run. This does not.
- Charts: cost-per-million no longer renders a below_resolution cell (no measurable
  throughput -> undefined cost) as $0.0000, which read as the cheapest gateway on the
  board; it now drops out honestly like any unmeasured cell.
- Charts: Peak memory metric called an undefined chooserCellMemory() and threw, so the
  chart never updated on selection; point it at memoryFor(), the reader every other
  memory surface uses.
- Refresh the committed GitHub star snapshot for all 14 gateways.

Author: MattJackson

gateways/stars.json

@@ -1,72 +1,72 @@
 {
  "agentgateway": {
-  "stars": 4466,
+  "stars": 4556,
   "first_commit": "2025-03-18",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "aisix": {
-  "stars": 114,
+  "stars": 121,
   "first_commit": "2026-04-17",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "apisix": {
-  "stars": 17016,
+  "stars": 17037,
   "first_commit": "2019-04-10",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "bifrost": {
-  "stars": 7483,
+  "stars": 7583,
   "first_commit": "2025-03-19",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
- "busbar": {
-  "stars": 118,
+ "busbar-151": {
+  "stars": 116,
   "first_commit": "2026-05-29",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "gomodel": {
-  "stars": 1073,
+  "stars": 1082,
   "first_commit": "2025-12-13",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "helicone": {
   "stars": 624,
   "first_commit": "2025-04-01",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "kong": {
-  "stars": 44017,
+  "stars": 44043,
   "first_commit": "2014-11-17",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "litellm-python": {
-  "stars": 56946,
+  "stars": 57322,
   "first_commit": "2023-07-27",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "litellm-rust": {
-  "stars": 56946,
+  "stars": 57322,
   "first_commit": "2023-07-27",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "one-api": {
-  "stars": 36517,
+  "stars": 36590,
   "first_commit": "2023-04-22",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "plano": {
-  "stars": 7013,
+  "stars": 7016,
   "first_commit": "2024-07-09",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "portkey": {
-  "stars": 12790,
+  "stars": 12837,
   "first_commit": "2023-08-23",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  },
  "tensorzero": {
-  "stars": 11721,
+  "stars": 11719,
   "first_commit": "2024-07-16",
-  "as_of": "2026-08-21"
+  "as_of": "2026-08-26"
  }
 }

site/app.js

@@ -1202,7 +1202,13 @@ const CHART_METRICS = [
     get: (g, st) => mval((chooserCellPerf(g, st) || {}).rps_per_dollar) },
   { id: "permillion", label: "Cost per million requests", unit: "USD", log: true, desc: false,
     note: "Instance cost to serve a million requests at the selected bound. Lower is better.",
-    get: (g, st) => mval((chooserCellPerf(g, st) || {}).cost_per_million_usd) },
+    // Cost is instance $/hr ÷ throughput, so a below_resolution cell (no rate the rig could weigh)
+    // makes cost UNDEFINED, not zero: coercing mval's below_resolution→0 here renders $0.0000, which
+    // reads as the CHEAPEST gateway on the board when in truth it sustained no measurable throughput -
+    // the exact inverse of the metric. Drop it like any unmeasured cell (it shows in the "not shown"
+    // note) instead. This is why cost cannot share mval's magnitude coercion with added-latency/CPU.
+    get: (g, st) => { const e = (chooserCellPerf(g, st) || {}).cost_per_million_usd;
+      return (isEnvelope(e) && e.reason === "below_resolution") ? null : mval(e); } },
   { id: "lat", label: "Added latency (p99)", unit: "µs", log: true, desc: false,
     note: "Gateway p99 minus direct-to-mock p99 at concurrency 1. Lower is better.",
     get: (g, st) => mval((chooserCellPerf(g, st) || {}).added_latency_p99_us) },
@@ -1211,7 +1217,7 @@ const CHART_METRICS = [
     get: (g, st) => { const r = frontierAt(frontierOf(chooserCellPerf(g, st)), selectedBound(st)); return r ? mval(r.rps) : null; } },
   { id: "rss", label: "Peak memory", unit: "MiB", log: false, desc: false,
     note: "Highest resident memory observed while the fixed load ran on the chosen cell. Lower is better.",
-    get: (g, st) => mval((chooserCellMemory(g, st) || {}).peak_rss_mib) },
+    get: (g, st) => mval((memoryFor(g, st) || {}).peak_rss_mib) },
 ];
 
 /* LOG SCALE IS NOT A PREFERENCE ON SOME OF THESE, IT IS THE ONLY HONEST AXIS.
@@ -4838,7 +4844,9 @@ const ROSTER_KEY = {
   lastrun: (g) => { const d = gatewayLastRun(g); return d ? d.getTime() : null; }, // newer = larger ms
   age: (g) => (g.first_commit ? new Date(g.first_commit).getTime() : null), // older = smaller ms
   stars: (g) => (g.stars == null ? null : g.stars),
-  cls: (g) => (g.cls || "Gateway").toLowerCase(),
+  // First-party rows (no contributor) sort LAST in either direction, so the outside contributors -
+  // the whole point of the column - cluster at the top when a reader sorts by it.
+  contrib: (g) => { const cs = g.contributed_by || []; return cs.length ? cs[0].handle.toLowerCase() : null; },
 };
 const rosterRows = (gateways) => {
   const key = ROSTER_KEY[rosterSort.col] || ROSTER_KEY.name;
@@ -4858,6 +4866,22 @@ const rosterRows = (gateways) => {
 /* Star counts render compact: 12345 -> "12.3k", below 1000 the full int. Null (no
    snapshot entry) stays null; the cell renders it muted. */
 const fmtStars = (v) => (v == null ? null : v >= 1000 ? `${(v / 1000).toFixed(1)}k` : String(Math.round(v)));
+/* The Contributors cell: the outside people who submitted or maintain this entrant, each a
+   `@handle` chip linking to their GitHub profile (url pre-validated to https-only in gen-data, so
+   it is safe raw in href; a null url renders the chip as plain text). A first-party entrant has no
+   contributor and renders a muted dash - the column exists to show that several rows do, so the
+   bench is visibly not all first-party. */
+function contribCell(g) {
+  const cs = g.contributed_by || [];
+  if (!cs.length) return `<span class="muted" title="first-party entrant">&mdash;</span>`;
+  return cs.map((c) => {
+    const chip = `@${esc(c.handle)}`;
+    const title = esc(`${c.name || c.handle} contributed this entrant`);
+    return c.url
+      ? `<a class="contrib-link" href="${c.url}" target="_blank" rel="noopener noreferrer" title="${title}">${chip}</a>`
+      : `<span class="contrib-link" title="${title}">${chip}</span>`;
+  }).join(" ");
+}
 /* Project age from the repo's first-commit date, in ONE simple floored unit: "11+ years",
    "7+ months", "3+ weeks". Context for the star counts and scores - a decade-old project and a
    three-week-old one earn them differently. Null (no snapshot) renders muted. */
@@ -5055,7 +5079,7 @@ function renderGateways() {
       <td class="lastrun">${lastRunTxt ? `${runModeCell(g)}<span title="last benchmarked ${esc(lastRun.toISOString().slice(0, 16).replace("T", " "))} UTC">${esc(lastRunTxt)}</span>` : `<span class="muted">n/a</span>`}</td>
       <td class="age">${age ? `<span title="first commit ${esc(g.first_commit)}">${esc(age)}</span>` : `<span class="muted">n/a</span>`}</td>
       <td class="stars">${stars != null ? esc(stars) : `<span class="muted">n/a</span>`}</td>
-      <td class="cls">${esc(g.cls || "Gateway")}</td>
+      <td class="contrib">${contribCell(g)}</td>
     </tr>`;
   }).join("");
   // Row click opens the per-gateway drawer (same as the perf tabs) — /gateways rows are clickable too.

site/contributors.json

@@ -0,0 +1,11 @@
+{
+  "apisix": [
+    { "handle": "moonming", "name": "Ming Wen", "url": "https://github.com/moonming" }
+  ],
+  "aisix": [
+    { "handle": "moonming", "name": "Ming Wen", "url": "https://github.com/moonming" }
+  ],
+  "higress": [
+    { "handle": "geekspeng", "name": "geekspeng", "url": "https://github.com/geekspeng" }
+  ]
+}

site/gen-data.mjs

@@ -272,6 +272,13 @@ function resolvedSnapshot(key) {
 // absent key degrades to null; the site renders those muted.
 const starsSnap = readJson(join(gatewaysDir, "stars.json")) || {};
 
+// OUTSIDE CONTRIBUTORS, keyed by gateway. Editorial credit, kept in a SITE-ONLY file the engine
+// never parses - a contributor field in definition.json would change the engine's Manifest schema,
+// and the engine is a frozen instrument (ENGINE_PIN): a schema change there is a board-wide re-run.
+// This file is not the instrument, so crediting @moonming and @geekspeng costs no re-measurement.
+// Shape: { "<gatewayKey>": [ { "handle": "...", "name": "...", "url": "https://..." }, ... ] }.
+const contributorsBook = readJson(join(HERE, "contributors.json")) || {};
+
 // snapshotDefinitionsByKey: gateway key -> the metric-definitions map off the SAME snapshot that
 // became that gateway's g.matrix (see the capture below). Kept OUT of the gateway object itself -
 // the per-gateway records are the public bundle shape and definitions are a board-level projection
@@ -284,6 +291,17 @@ const allGateways = gatewayKeys.map((key) => {
     display: meta.display || key,
     lang: meta.lang || "Other",
     cls: meta.cls || "Gateway",
+    // Outside contributors (from the site-only contributorsBook), sanitised the same way g.repo is:
+    // app.js renders handle/name esc()'d but interpolates url RAW into an href, so only https://
+    // profile URLs survive - a bad/missing url drops to null and the cell renders the handle as
+    // plain text rather than a link.
+    contributed_by: (Array.isArray(contributorsBook[key]) ? contributorsBook[key] : [])
+      .filter((c) => c && typeof c.handle === "string" && c.handle.trim())
+      .map((c) => ({
+        handle: c.handle,
+        name: (typeof c.name === "string" && c.name.trim()) ? c.name : c.handle,
+        url: (typeof c.url === "string" && /^https:\/\/[^\s"'<>]+$/.test(c.url)) ? c.url : null,
+      })),
     // Only accept an https:// repo URL. app.js interpolates g.repo RAW into href="${g.repo}" at four
     // render sites (display is esc()'d, href is not), so a manifest GW_REPO like
     // `x" onfocus=alert(...) autofocus="` or a `javascript:` scheme would inject on the public board.

site/index.html

@@ -101,7 +101,7 @@ <h2 class="hero-word">Reproducible, honest benchmarks for the AI tooling you act
     <p class="fineprint">Every gateway on the board, in alphabetical order. Language, the exact version measured (hover for the full build/image digest), when it was last benchmarked, project age, stars (snapshot), and each project's own self-description. The other tabs measure how they perform.</p>
     <div class="table-scroll">
       <table id="gateways-table">
-        <thead><tr><th data-sort="name">Gateway</th><th data-sort="lang">Language</th><th class="build" data-sort="version">Version</th><th class="lastrun" data-sort="lastrun">Last benchmarked</th><th class="age" data-sort="age">Project Age</th><th class="stars" data-sort="stars">Stars</th><th data-sort="cls">Class</th></tr></thead>
+        <thead><tr><th data-sort="name">Gateway</th><th data-sort="lang">Language</th><th class="build" data-sort="version">Version</th><th class="lastrun" data-sort="lastrun">Last benchmarked</th><th class="age" data-sort="age">Project Age</th><th class="stars" data-sort="stars">Stars</th><th class="contrib" data-sort="contrib">Contributors</th></tr></thead>
         <tbody></tbody>
       </table>
     </div>

site/style.css

@@ -307,7 +307,13 @@ tbody tr { cursor: pointer; }
 #gateways-table thead th[data-dir="asc"]::after { content: " \25B2"; font-size: .7em; color: var(--fg-mute); }
 #gateways-table thead th[data-dir="desc"]::after { content: " \25BC"; font-size: .7em; color: var(--fg-mute); }
 #gateways-table tbody tr { cursor: pointer; }
-#gateways-table td.cls { color: var(--fg-dim); }
+/* Contributors cell: outside people who submitted/maintain an entrant. A first-party row renders a
+   muted dash; a credited row shows @handle chips linking to GitHub. The link colour is the normal
+   accent so it reads as "click me", and it stops the row-click so the profile opens instead of the
+   drawer. */
+#gateways-table td.contrib { color: var(--fg-dim); white-space: nowrap; }
+#gateways-table td.contrib .contrib-link { color: var(--accent); text-decoration: none; }
+#gateways-table td.contrib a.contrib-link:hover { text-decoration: underline; }
 
 /* drawer (per-gateway slide-over) */
 .backdrop { position: fixed; inset: 0; background: var(--overlay); z-index: 60; }