Skip to content

armgit / xerj

Public Imported from GitHub
0

AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.

blog: the rc.80 gates post — wins and losses

xerj-org2f36d6c88928
diff --git a/landing/blog/index.html b/landing/blog/index.html
index 7528737c3..a0ed0ebbe 100644
--- a/landing/blog/index.html
+++ b/landing/blog/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>XERJ.ai — Blog: the engineering log</title>
-<meta name="description" content="Measured results with the losses left in: the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
+<meta name="description" content="Measured results with the losses left in: the rc.80 gates post, the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
<link rel="preconnect" href="https://fonts.googleapis.com">
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
<link href="https://fonts.googleapis.com/css2?family=Big+Shoulders+Display:wght@400;700;800;900&family=Inter:wght@400;500;600;700&family=IBM+Plex+Sans:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500;700&display=swap" rel="stylesheet">
@@ -16,7 +16,7 @@
<meta property="og:site_name" content="XERJ.ai">
<meta property="og:type" content="website">
<meta property="og:title" content="XERJ.ai — Blog: the engineering log">
-<meta property="og:description" content="Measured results with the losses left in: the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
+<meta property="og:description" content="Measured results with the losses left in: the rc.80 gates post, the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
<meta property="og:url" content="https://xerj.org/blog/">
<meta property="og:image" content="https://xerj.org/og/xerj-card.png">
<meta property="og:image:width" content="1200">
@@ -25,7 +25,7 @@
<meta property="og:locale" content="en_US">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="XERJ.ai — Blog: the engineering log">
-<meta name="twitter:description" content="Measured results with the losses left in: the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
+<meta name="twitter:description" content="Measured results with the losses left in: the rc.80 gates post, the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.">
<meta name="twitter:image" content="https://xerj.org/og/xerj-card.png">
<script type="application/ld+json">
{
@@ -59,11 +59,11 @@
"@context": "https://schema.org",
"@type": "CollectionPage",
"name": "XERJ.ai — Blog: the engineering log",
- "description": "Measured results with the losses left in: the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.",
+ "description": "Measured results with the losses left in: the rc.80 gates post, the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.",
"url": "https://xerj.org/blog/",
"image": "https://xerj.org/og/xerj-card.png",
"inLanguage": "en",
- "dateModified": "2026-09-27",
+ "dateModified": "2026-09-30",
"isPartOf": {
"@id": "https://xerj.org/#website"
},
@@ -126,6 +126,61 @@
<!-- Reverse-chronological by hand while there is one post; new posts PREPEND.
Move to a generator only past ~10 posts. -->
<ol class="blog-list">
+ <li>
+ <a class="bl-card" href="/blog/the-rc80-gates">
+ <!-- Bar widths are normalized WITHIN each row against that row's own
+ max (the card mixes units — F1, a count, an ECE — so one global
+ denominator would be meaningless). Neutral bar = the bar or the
+ before, accent bar = the measured value, verbatim from the post:
+ /_ask 0.9975 vs bar 0.9 · map 400s 20/30 -> 0/30 · ECE 0.3109
+ raw -> 0.0088 isotonic · judge SciFact 0.7045 -> 0.5967 (the row
+ we lose, kept in). -->
+ <div class="bl-preview" aria-hidden="true">
+ <div class="blp-kicker">RELEASE GATES · TEN FEATURES · EIGHT GATES</div>
+ <div class="blp-row">
+ <div class="blp-ds">/_ASK F1</div>
+ <div class="blp-pair">
+ <span class="blp-num">0.9975 <i>measured</i></span><span class="blp-bar a" style="width:100%"></span>
+ <span class="blp-num b">0.9000 <i>the bar</i></span><span class="blp-bar b" style="width:90%"></span>
+ </div>
+ </div>
+ <div class="blp-row">
+ <div class="blp-ds">MAP 400S</div>
+ <div class="blp-pair">
+ <span class="blp-num">20/30 <i>guessed</i></span><span class="blp-bar b" style="width:100%"></span>
+ <span class="blp-num">0/30 <i>map-informed</i></span><span class="blp-bar a" style="width:2%; min-width:3px"></span>
+ </div>
+ </div>
+ <div class="blp-row">
+ <div class="blp-ds">FIQA ECE</div>
+ <div class="blp-pair">
+ <span class="blp-num">0.3109 <i>raw</i></span><span class="blp-bar b" style="width:100%"></span>
+ <span class="blp-num">0.0088 <i>isotonic</i></span><span class="blp-bar a" style="width:2.8%"></span>
+ </div>
+ </div>
+ <div class="blp-row">
+ <div class="blp-ds">JUDGE</div>
+ <div class="blp-pair">
+ <span class="blp-num">0.7045 <i>hybrid</i></span><span class="blp-bar a" style="width:100%"></span>
+ <span class="blp-num b">0.5967 <i>+ judge</i> <span class="blp-chip">&minus;0.108</span></span><span class="blp-bar b" style="width:84.7%"></span>
+ </div>
+ </div>
+ <p class="blp-foot">3 PASS · 3 FAIL · 1 SPLIT · four issues filed (#1091&ndash;#1094) · runs 2026-09-30</p>
+ </div>
+ <div class="bl-meta">
+ <p class="bl-date"><time datetime="2026-09-30">2026-09-30</time> · RELEASE GATES · ENGINEERING</p>
+ <h2 class="bl-title">The rc.80 gates: three passed outright, three failed, one split</h2>
+ <p class="bl-excerpt">The ten-item program shipped, and every gate that could carry a number was run in
+ public. <strong>/_ask</strong> scored macro-F1 <strong>0.9975</strong> at p50 1.129 ms;
+ <strong>xerj_map</strong> took unknown-field 400s from <strong>20/30 to 0/30</strong>; calibration took
+ FiQA ECE from <strong>0.3109 to 0.0088</strong>. The local judge lost everywhere it was measured, the
+ flywheel covered <strong>19.72%</strong> against an 80% bar, and the decide head missed its accuracy
+ bars &mdash; those rows are in the post too, and they filed four issues.</p>
+ <p class="bl-byline">AUTHOR · XERJ ENGINEERING TEAM · PUBLISHED · 2026-09-30</p>
+ <span class="bl-read">READ THE POST &rarr;</span>
+ </div>
+ </a>
+ </li>
<li>
<a class="bl-card" href="/blog/one-vulnerability-three-names">
<!-- Bar widths share ONE denominator, 4,107 envelopes (the max across
diff --git a/landing/blog/the-rc80-gates.html b/landing/blog/the-rc80-gates.html
new file mode 100644
index 000000000..585691cfd
--- /dev/null
+++ b/landing/blog/the-rc80-gates.html
@@ -0,0 +1,403 @@
+<!DOCTYPE html>
+<html lang="en" data-theme="day">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>The rc.80 gates: three passed outright, three failed, one split</title>
+<meta name="description" content="Ten features, eight measured gates: /_ask macro-F1 0.9975, xerj_map 20/30 to 0/30, calibration ECE 0.0088 pass; the judge, the flywheel and the decide head fail their bars, numbers published.">
+<link rel="preconnect" href="https://fonts.googleapis.com">
+<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
+<link href="https://fonts.googleapis.com/css2?family=Big+Shoulders+Display:wght@400;700;800;900&family=Inter:wght@400;500;600;700&family=IBM+Plex+Sans:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500;700&display=swap" rel="stylesheet">
+<link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'><rect width='100' height='100' fill='%230b0b0d'/><text x='50' y='68' font-size='60' font-family='Big Shoulders Display,sans-serif' font-weight='900' fill='%23ffc400' text-anchor='middle'>X</text></svg>">
+<link rel="stylesheet" href="/style.css?v=if9">
+<link rel="stylesheet" href="/blog.css?v=1">
+<!-- seo:begin — generated by scripts/seo/fix_heads.py · do not hand-edit -->
+<link rel="canonical" href="https://xerj.org/blog/the-rc80-gates">
+<meta property="og:site_name" content="XERJ.ai">
+<meta property="og:type" content="article">
+<meta property="og:title" content="The rc.80 gates: three passed outright, three failed, one split">
+<meta property="og:description" content="Ten features, eight measured gates: /_ask macro-F1 0.9975, xerj_map 20/30 to 0/30, calibration ECE 0.0088 pass; the judge, the flywheel and the decide head fail their bars, numbers published.">
+<meta property="og:url" content="https://xerj.org/blog/the-rc80-gates">
+<meta property="og:image" content="https://xerj.org/og/xerj-card.png">
+<meta property="og:image:width" content="1200">
+<meta property="og:image:height" content="630">
+<meta property="og:image:alt" content="XERJ.ai — the search engine built for AI agents">
+<meta property="og:locale" content="en_US">
+<meta name="twitter:card" content="summary_large_image">
+<meta name="twitter:title" content="The rc.80 gates: three passed outright, three failed, one split">
+<meta name="twitter:description" content="Ten features, eight measured gates: /_ask macro-F1 0.9975, xerj_map 20/30 to 0/30, calibration ECE 0.0088 pass; the judge, the flywheel and the decide head fail their bars, numbers published.">
+<meta name="twitter:image" content="https://xerj.org/og/xerj-card.png">
+<script type="application/ld+json">
+{
+ "@context": "https://schema.org",
+ "@type": "Organization",
+ "@id": "https://xerj.org/#organization",
+ "name": "XERJ",
+ "alternateName": "XERJ.ai",
+ "url": "https://xerj.org/",
+ "logo": "https://xerj.org/brandbook/xerj-mark.svg",
+ "sameAs": [
+ "https://github.com/xerj-org/xerj"
+ ]
+}
+</script>
+<script type="application/ld+json">
+{
+ "@context": "https://schema.org",
+ "@type": "WebSite",
+ "@id": "https://xerj.org/#website",
+ "name": "XERJ.ai",
+ "url": "https://xerj.org/",
+ "inLanguage": "en",
+ "publisher": {
+ "@id": "https://xerj.org/#organization"
+ }
+}
+</script>
+<script type="application/ld+json">
+{
+ "@context": "https://schema.org",
+ "@type": "Article",
+ "headline": "The rc.80 gates: three passed outright, three failed, one split",
+ "description": "Ten features, eight measured gates: /_ask macro-F1 0.9975, xerj_map 20/30 to 0/30, calibration ECE 0.0088 pass; the judge, the flywheel and the decide head fail their bars, numbers published.",
+ "url": "https://xerj.org/blog/the-rc80-gates",
+ "image": "https://xerj.org/og/xerj-card.png",
+ "inLanguage": "en",
+ "datePublished": "2026-09-30",
+ "dateModified": "2026-09-30",
+ "author": {
+ "@id": "https://xerj.org/#organization"
+ },
+ "publisher": {
+ "@id": "https://xerj.org/#organization"
+ },
+ "isPartOf": {
+ "@id": "https://xerj.org/#website"
+ },
+ "mainEntityOfPage": {
+ "@type": "WebPage",
+ "@id": "https://xerj.org/blog/the-rc80-gates"
+ }
+}
+</script>
+<script type="application/ld+json">
+{
+ "@context": "https://schema.org",
+ "@type": "BreadcrumbList",
+ "itemListElement": [
+ {
+ "@type": "ListItem",
+ "position": 1,
+ "name": "Home",
+ "item": "https://xerj.org/"
+ },
+ {
+ "@type": "ListItem",
+ "position": 2,
+ "name": "Blog",
+ "item": "https://xerj.org/blog/"
+ },
+ {
+ "@type": "ListItem",
+ "position": 3,
+ "name": "The rc.80 gates"
+ }
+ ]
+}
+</script>
+<!-- seo:end -->
+</head>
+<body>
+
+<nav class="nav" aria-label="Primary">
+ <a href="/" class="brand">XERJ</a>
+ <a href="/">HOME</a>
+ <a href="/product">PRODUCT</a>
+ <span class="nav-drop"><a href="/use-cases">USE CASES</a></span>
+ <a href="/case-studies">CASE STUDIES</a>
+ <a href="/benchmarks/">BENCHMARKS</a>
+ <a href="/blog/" class="active">BLOG</a>
+ <span class="spacer"></span>
+ <a href="/for-agents">FOR AGENTS</a>
+ <a href="/docs/">DOCS</a>
+ <a href="/docs/recipes/">RECIPES</a>
+ <a href="/playground">DASHBOARDS</a>
+ <a href="https://github.com/xerj-org/xerj" rel="external">GITHUB</a>
+ <span class="theme"><button type="button" data-theme="day" class="active">DAY</button><span class="dash">·</span><button type="button" data-theme="night">NIGHT</button></span>
+</nav>
+
+<main>
+<div class="uc-hero">
+ <p class="blog-crumb"><a href="/blog/">&larr; ALL POSTS · /BLOG</a></p>
+ <div class="uc-eyebrow">RELEASE GATES · TEN FEATURES · EIGHT GATES · RUNS 2026-09-30</div>
+ <h1>THE RC.80 GATES.<br><span class="accent">MEASURED IN PUBLIC, LOSSES LEFT IN.</span></h1>
+ <p class="blog-byline">AUTHOR · XERJ ENGINEERING TEAM · PUBLISHED · 2026-09-30</p>
+
+ <!-- The verdict, first. Every figure below is verbatim from the gate runs
+ cited in §01–§04; all runs are 2026-09-30 on the release build of
+ main @ a179e1ef3 (version string v1.0.0-rc.78). -->
+ <section class="verdict-panel" id="verdict">
+ <div class="vh"><span class="wn">THE VERDICT</span>
+ <span class="wm">split, measured, with the losses left in</span></div>
+ <p>The program behind rc.80 shipped ten features, and every item that could carry a number was filed with a measured gate that binds at ship time. This post is the packaging half of that rule, and the scoreboard is a split. <strong>Three gates passed outright</strong>: <code class="inline">/_ask</code> turned 230 prompt/gold pairs into validated DSL at macro-F1 <strong>0.9975</strong> with zero invalid queries and wall-clock p50 <strong>1.129 ms</strong> against a 300 ms budget; <code class="inline">xerj_map</code> took unknown-field 400s on MCP-shaped DSL from <strong>20/30 guessed to 0/30 map-informed</strong> &mdash; a 100% reduction against a &ge;80% bar; and the calibration layer took FiQA rerank ECE from <strong>0.3109 raw to 0.0088 held-out</strong> against a &le;0.10 bar. <strong>Three failed with the numbers</strong>: the local judge loses to its own first stage everywhere it was measured (SciFact <strong>0.7045 &rarr; 0.5967</strong>), the decision flywheel answers <strong>19.72%</strong> of later traffic at &ge;0.8 confidence against an 80% bar, and the tier-2 decide head misses its accuracy bars by 3 to 59 points. <strong>One split</strong>: stemming gained +0.0160/+0.0179 BM25 nDCG@10 (bar +0.01) and still left 15 zero-hit queries against a bar of &le;10 &mdash; and no stemmer can fix the 15. The failures filed four issues, <strong>#1091&ndash;#1094</strong>. No number below was tuned after its gate was read.</p>
+ </section>
+
+ <p class="uc-lead">One window, ten features, eight gates, all measured on 2026-09-30 against bars that were written down before the work started. The wins and the losses sit in the same tables, every figure links to the run that produced it, and the four defects the gates exposed are filed in the open tracker with the measurements that found them. That is the whole post.</p>
+</div>
+
+<section class="scene-section" style="border-top:0; padding-top:0;">
+ <div class="stat-band">
+ <div class="stat"><div class="v">0.9975</div><div class="k">macro-F1, /_ask</div>
+ <div class="s">230 prompt/gold pairs, 0 invalid DSL, byte-identical across 3 runs. p50 1.129 ms; the bar allowed 300.</div></div>
+ <div class="stat"><div class="v">0/30</div><div class="k">unknown-field 400s, with xerj_map</div>
+ <div class="s">down from 20/30 guessed field names. 100% reduction; the bar asked for 80%. The 5 silent 0-hits became issue #1093.</div></div>
+ <div class="stat"><div class="v">0.0088</div><div class="k">held-out ECE, calibrated</div>
+ <div class="s">from 0.3109 raw on 3,600 held-out FiQA pairs, query-level split. Temperature scaling managed 0.3163 and is published beside it.</div></div>
+ <div class="stat"><div class="v">4</div><div class="k">issues the failures filed</div>
+ <div class="s">#1091 hybrid cost, #1092 analyzer echo, #1093 silent 0-hit, #1094 flywheel freeze. Each carries its measurement.</div></div>
+ </div>
+</section>
+
+<!-- ============================ 01 · THE PROGRAM ============================ -->
+<section class="scene-section" id="program">
+ <div class="kicker"><span class="accent">01</span><span class="dash">·</span><span>THE PROGRAM · TEN FEATURES, EACH WITH A GATE THAT BINDS</span></div>
+ <h2 class="scene" style="font-size:var(--fs-56);">THE RULE WAS WRITTEN<br><span class="accent">BEFORE THE WORK.</span></h2>
+ <p class="wide">The ten-item program to the first official release (<a href="https://github.com/xerj-org/xerj/discussions/1054" rel="external">discussion #1054</a>, filed as issues <a href="https://github.com/xerj-org/xerj/issues/1055" rel="external">#1055</a>&ndash;<a href="https://github.com/xerj-org/xerj/issues/1064" rel="external">#1064</a>) was planned as five items for rc.79, four for rc.80, and the open model for v1.1.0. They landed together in one window &mdash; PRs #1066&ndash;#1090 since the rc.78 cut &mdash; so the gates were run as one suite, and the artifacts call it the rc.80 gate suite. This post keeps that name and states the plan honestly: two planned releases, one measured window.</p>
+
+ <p class="wide">The plan's own packaging rule, verbatim: <em>a <code class="inline">benchmarks/&lt;name&gt;</code> directory with raw results plus a blog post with the losses left in.</em> Three items also carried a loses-does-not-ship rule in their issue text &mdash; the local judge must beat the shipped hybrid arm beyond the run spread or it does not ship; the decide head below its bar ships behind a flag with the numbers published; the FiQA judge row must clear 0.30. This post is where those rules get honoured.</p>
+
+ <div class="vt-wrap">
+ <table class="vt">
+ <thead><tr><th>The program · feature (PR) · its gate</th><th>Verdict</th></tr></thead>
+ <tbody>
+ <tr><td><code class="inline">xerj_map</code> MCP FieldSpec tool (<a href="https://github.com/xerj-org/xerj/pull/1074" rel="external">#1074</a>) &middot; unknown-field 400s down &ge;80%</td>
+ <td class="n w">PASS &mdash; 100%</td></tr>
+ <tr><td><code class="inline">POST /_ask</code> + <code class="inline">xerj_plan</code> (<a href="https://github.com/xerj-org/xerj/pull/1076" rel="external">#1076</a>) &middot; macro-F1 &ge;0.9, 0 invalid DSL, p50 &le;300 ms</td>
+ <td class="n w">PASS</td></tr>
+ <tr><td>tier-2 local decide head via candle (<a href="https://github.com/xerj-org/xerj/pull/1072" rel="external">#1072</a>, <a href="https://github.com/xerj-org/xerj/pull/1073" rel="external">#1073</a>) &middot; SMS &ge;0.95 acc, ECE &le;0.05, AG News &ge;0.85</td>
+ <td class="n l">FAIL (quality) / PASS (latency)</td></tr>
+ <tr><td><code class="inline">max_tokens</code> on every MCP search tool (<a href="https://github.com/xerj-org/xerj/pull/1067" rel="external">#1067</a>) &middot; response-size budget knob</td>
+ <td class="n t">no measured gate in this suite</td></tr>
+ <tr><td>stemming analyzer for profiler-marked prose (<a href="https://github.com/xerj-org/xerj/pull/1070" rel="external">#1070</a>) &middot; BM25 &ge;+0.01; zero-hit &le;10</td>
+ <td class="n">SPLIT &mdash; +0.0160/+0.0179 PASS; 15 zero-hit FAIL</td></tr>
+ <tr><td>self-judged hits judge stage (<a href="https://github.com/xerj-org/xerj/pull/1077" rel="external">#1077</a>) &middot; beat hybrid beyond spread, or do not ship</td>
+ <td class="n l">FAIL (quality) / PASS (cost)</td></tr>
+ <tr><td>decision-cache flywheel (<a href="https://github.com/xerj-org/xerj/pull/1075" rel="external">#1075</a>) &middot; &ge;80% of later traffic from history at &ge;0.8, acc &ge;0.97</td>
+ <td class="n l">FAIL (share) / PASS (accuracy)</td></tr>
+ <tr><td>real <code class="inline">_watcher</code> evaluate-or-501 + <code class="inline">autoindex --label</code> (<a href="https://github.com/xerj-org/xerj/pull/1082" rel="external">#1082</a>) &middot; wrong-and-confident &le;5% at p&ge;0.8</td>
+ <td class="n t">gate not yet measured</td></tr>
+ <tr><td>calibration layer, <code class="inline">p_cal</code> (<a href="https://github.com/xerj-org/xerj/pull/1080" rel="external">#1080</a>) &middot; held-out ECE &le;0.10</td>
+ <td class="n w">PASS &mdash; 0.0088</td></tr>
+ <tr><td>open xerj-decide-v1 model (<a href="https://github.com/xerj-org/xerj/pull/1084" rel="external">#1084</a>) &middot; deterministic exports, published eval card</td>
+ <td class="n">reported &mdash; card carries the weak numbers</td></tr>
+ </tbody>
+ </table>
+ </div>
+
+ <ul class="deep">
+ <li><strong>Two rows say "not measured" on purpose.</strong> <code class="inline">max_tokens</code> is a response-size budget with no quality claim attached, and the <code class="inline">_watcher</code> detection gate (wrong-and-confident rate at p&ge;0.8) has not been run &mdash; the watcher's mechanics are test-pinned, its detection-quality number is not, and this post does not invent one.</li>
+ <li><strong>One binary, one day.</strong> Every number in this post was measured on 2026-09-30 on release builds of main @ <code class="inline">a179e1ef3</code>, version string <code class="inline">v1.0.0-rc.78</code> &mdash; the rc.79/rc.80 tags are cut after the gates, not before them. The two exceptions are stated where they occur (the tier-2 feature build, the paid provider arm).</li>
+ </ul>
+ <p class="src">The plan: <a href="https://github.com/xerj-org/xerj/discussions/1054" rel="external">discussion #1054</a> ·
+ tracker <a href="https://github.com/xerj-org/xerj/issues/1065" rel="external">#1065</a> ·
+ issues <a href="https://github.com/xerj-org/xerj/issues/1055" rel="external">#1055</a>&ndash;<a href="https://github.com/xerj-org/xerj/issues/1064" rel="external">#1064</a> ·
+ window PRs #1066&ndash;#1090 · gate PRs
+ <a href="https://github.com/xerj-org/xerj/pull/1079" rel="external">#1079</a>,
+ <a href="https://github.com/xerj-org/xerj/pull/1085" rel="external">#1085</a>,
+ <a href="https://github.com/xerj-org/xerj/pull/1086" rel="external">#1086</a>,
+ <a href="https://github.com/xerj-org/xerj/pull/1087" rel="external">#1087</a>,
+ <a href="https://github.com/xerj-org/xerj/pull/1088" rel="external">#1088</a>,
+ <a href="https://github.com/xerj-org/xerj/pull/1090" rel="external">#1090</a>.</p>
+</section>
+
+<!-- ============================ 02 · THE PASSES ============================ -->
+<section class="scene-section" id="passes">
+ <div class="kicker"><span class="accent">02</span><span class="dash">·</span><span>THE PASSES · THREE GATES, CLEAN</span></div>
+ <h2 class="scene" style="font-size:var(--fs-56);">THREE PASSED.<br><span class="accent">WITH THEIR NUMBERS.</span></h2>
+
+ <div class="wire">
+ <div class="wp">
+ <div class="wh"><span class="wn">PASS 1</span><h3><code class="inline">POST /_ask</code> + <code class="inline">xerj_plan</code>: prompt in, validated DSL out</h3>
+ <span class="wm">#1056 · PRs #1076, #1079</span></div>
+ <p>The gate: &ge;200 (prompt, gold result-set) pairs over public tabular datasets, result-set F1 &ge;0.9, zero invalid DSL, and a latency budget. Measured: <strong>230 pairs</strong> (72 USGS earthquakes / 78 exoplanets / 80 gapminder), macro-F1 <strong>0.9975</strong>, <strong>0 invalid DSL</strong> &mdash; every returned plan passes <code class="inline">parse_request</code> before it is returned &mdash; and byte-identical output across <strong>3 runs</strong>. Wall-clock p50 <strong>1.129 ms</strong>, p99 2.852 ms, n=690 over loopback HTTP. The bar allowed 300 ms; the planner clears it by 265&times;.</p>
+ <p><strong>The one arm not run:</strong> the token gate &mdash; output tokens per solved structured-query task &le;50% of agent-written DSL at equal solve rate &mdash; is <em>designed, not measured</em>. It needs a live agent harness with real model calls, and the cost of that run was not spent this window. The design sits in <a href="https://github.com/xerj-org/xerj/tree/main/benchmarks/ask-plan" rel="external">benchmarks/ask-plan</a> and the claim is not made here.</p>
+ </div>
+ <div class="wp">
+ <div class="wh"><span class="wn">PASS 2</span><h3><code class="inline">xerj_map</code>: the field map an agent was guessing at</h3>
+ <span class="wm">#1055 · PRs #1074, #1086</span></div>
+ <p>Thirty MCP-shaped DSL queries against a reference-coding corpus, issued twice: once guessing field names from the prompt alone, once informed by the map. Guessed: <strong>20 of 30</strong> came back as unknown-field 400s, and only 5 returned hits. Map-informed: <strong>0 of 30</strong> unknown-field 400s, 29 of 30 with hits. A <strong>100% reduction against the &ge;80% bar</strong>.</p>
+ <p><strong>The discovery is worth more than the pass.</strong> Five of the guessed queries did not error at all &mdash; a wrong field name inside <code class="inline">term</code>, <code class="inline">match</code>, <code class="inline">range</code> or <code class="inline">exists</code> is a <em>silent 0-hit HTTP 200</em>. The gate counted 400s; the silent rows are the worse failure, and they are now <a href="https://github.com/xerj-org/xerj/issues/1093" rel="external">issue #1093</a>.</p>
+ </div>
+ <div class="wp">
+ <div class="wh"><span class="wn">PASS 3</span><h3>Calibration: <code class="inline">p_cal</code> beside every <code class="inline">p_raw</code></h3>
+ <span class="wm">#1063 · PRs #1080, #1087</span></div>
+ <p>The baseline is the number this blog already published: on FiQA, a hosted judge probability of 0.93 meant relevance 34% of the time &mdash; ECE <strong>0.3109</strong>. The gate: held-out ECE &le;0.10 after fitting. Measured on <strong>19,440 raw (query, doc, p_raw, gold) pairs</strong> rebuilt from 648 judged FiQA queries &mdash; <strong>$0.2125</strong> of provider calls, 5,059,586 input tokens, zero failed calls &mdash; with a query-level split: the fit never sees any pair from a held-out query, which is the deployment shape. Isotonic held-out ECE <strong>0.0088</strong>. Temperature scaling, the comparison arm, managed <strong>0.3163</strong> and fails the same bar; both arms' ECE is published by the node at <code class="inline">/_decide/_calibration</code>.</p>
+ <p><strong>The gate also audited the fitter.</strong> Writing an independent naive PAVA in Python found a knot-index bug in the benchmark's mirror &mdash; fixed there, and the shipped Rust module was verified against the naive implementation to 8.7e-19 at all 98 pooled points. The defect was in the measuring stick, not the product; both are now pinned.</p>
+ </div>
+ </div>
+ <p class="src">ask-plan: <a href="https://github.com/xerj-org/xerj/tree/main/benchmarks/ask-plan" rel="external">benchmarks/ask-plan</a> (gate table with the measured rows) and
+ <a href="https://github.com/xerj-org/xerj/tree/main/benchmarks/ask-plan/results/2026-09-30-latency-gate" rel="external">results/2026-09-30-latency-gate</a> ·
+ map gate: <a href="https://github.com/xerj-org/xerj/pull/1086" rel="external">PR #1086</a> (artifacts in the PR) ·
+ calibration: <a href="https://github.com/xerj-org/xerj/tree/main/benchmarks/decisions-calibration/results/2026-09-30-pairlevel-fiqa" rel="external">results/2026-09-30-pairlevel-fiqa</a> ·
+ pairs and shortlists committed beside them.</p>
+</section>
+
+<!-- ============================ 03 · THE FAILURES ============================ -->
+<section class="scene-section" id="failures">
+ <div class="kicker"><span class="accent">03</span><span class="dash">·</span><span>THE FAILURES · READINGS, NOT EXCUSES</span></div>
+ <h2 class="scene" style="font-size:var(--fs-56);">THREE FAILED, ONE SPLIT.<br><span class="accent">THE NUMBERS STAY IN.</span></h2>
+
+ <h3 class="scene" style="font-size:var(--fs-20); margin-top:var(--sp-8);">Stemming (#1059): a real win and a real miss, in the same gate</h3>
+ <p class="wide">Paired arms on one node, differing only in the create-time analyzer. BM25 nDCG@10: SciFact <strong>0.6572 &rarr; 0.6732 (+0.0160)</strong>, NFCorpus <strong>0.3016 &rarr; 0.3195 (+0.0179)</strong> &mdash; both clear the +0.01 bar. Zero-hit NFCorpus queries: <strong>25 &rarr; 15, against a bar of &le;10</strong>. Stemming repaired exactly the ten plural and inflection queries (bagels, leeks, pineapples, turnips&hellip;). The fifteen that remain &mdash; Fosamax, Zoloft, eggnog, halibut, Yale &mdash; are single words whose <em>stem also occurs nowhere in the corpus</em>. No stemmer can lexically bridge a word that is absent in every inflected form. That half of the gate was unreachable by the feature it gated; only the vector arm can fix those queries, and that is the honest reading: the gate was half right to exist, and the miss is a boundary, not a bug in the stemmer.</p>
+ <p class="wide">Two adjacent records: <code class="inline">GET _settings</code> does not echo the analysis block and <code class="inline">_analyze</code> shows the standard path even when the stemmer is declared and provably applied by search &mdash; filed as <a href="https://github.com/xerj-org/xerj/issues/1092" rel="external">issue #1092</a> rather than tuned around. Raw per-query outputs are committed.</p>
+
+ <h3 class="scene" style="font-size:var(--fs-20); margin-top:var(--sp-8);">The tier-2 decide head (#1057): misses its bars, ships behind the flag anyway</h3>
+ <div class="vt-wrap">
+ <table class="vt">
+ <thead><tr><th>Decide head, xerj-decide-v1 weights · measured on the wire</th><th>Accuracy</th><th>ECE</th><th>Bar</th><th>Verdict</th></tr></thead>
+ <tbody>
+ <tr><td>SMS held-out, noul · 1,574 items</td><td class="n">0.9193</td><td class="n">0.287</td><td class="t">acc &ge;0.95, ECE &le;0.05</td><td class="n l">FAIL / FAIL</td></tr>
+ <tr><td>AG News test, 4-way choice · 7,600 items, never trained on it</td><td class="n">0.2599</td><td class="n">0.060</td><td class="t">&ge;0.85</td><td class="n l">FAIL (chance = 0.25)</td></tr>
+ <tr><td>Banking77 test, 77-way choice · 3,080 items</td><td class="n">0.1185</td><td class="n">0.0717</td><td class="t">reported, not gated</td><td class="n t">reported</td></tr>
+ <tr><td>Latency, 30 questions, 8 pinned cores</td><td class="n" colspan="2">p50 8.4 ms</td><td class="t">&le;50 ms</td><td class="n w">PASS</td></tr>
+ </tbody>
+ </table>
+ </div>
+ <p class="wide">The reading is in the issue's own rule: <em>below the bar it ships behind a flag with the numbers published</em> &mdash; <code class="inline">--decide-mode local --decide-model-dir</code>, tier 2 off unless asked for. The artifact is <strong>28.5 MiB of F32 weights, not quantized</strong> (candle 0.9 has no quantized safetensors <code class="inline">VarBuilder</code>; the PR documents that as deliberate), well under the 300 MB download budget &mdash; and there is no download at all: the tier loads a local directory and makes no egress. Tier 1, the history vote, stays ahead of tier 2 on every home-ground row: Banking77 <strong>0.8289 vs 0.1185</strong>, SMS <strong>0.9848 vs 0.9193</strong>, AG News <strong>0.9182 vs 0.2599</strong>. That is the design working: where labelled history exists, the vote wins; where it does not, tier 2 exists at all &mdash; and its zero-shot boundary is now a published number instead of a hope.</p>
+
+ <div class="loss">
+ <strong>The gate's second catch is the more useful one.</strong> Measuring tier 2 on the <em>stock</em> release binary found that <code class="inline">--decide-mode local</code> was a no-op there: release binaries were built without the <code class="inline">decide-local</code> cargo feature, the node logged a warning, and the ladder quietly stayed on the history vote. Every tier-2 number above was therefore measured on a feature build of the same commit, with the no-op evidence committed (<code class="inline">stock-binary-noop.txt</code>). The fix is release engineering, not code: the shipped build must carry the feature &mdash; the runtime switch stays opt-in &mdash; and it is queued ahead of the cut. A gate that cannot reach the feature it is gating is measuring the build, and that is exactly what it caught.
+ </div>
+
+ <h3 class="scene" style="font-size:var(--fs-20); margin-top:var(--sp-8);">The flywheel (#1061): frozen as shipped, 4&times; under the bar even when forced</h3>
+ <p class="wide">Two arms, both worse than the bar. <strong>As shipped</strong>: 2,000 fill requests fired, and only <strong>12 answers were ever cached</strong> &mdash; write-back fires only when tier 2 answers, and after the first cached document the history vote returns BM25 support for ~99.4% of traffic, so tier 2 stops answering and the cache freezes at 12 documents. That is a defect, filed as <a href="https://github.com/xerj-org/xerj/issues/1094" rel="external">issue #1094</a>, and it makes the gate's premise unreachable through the ladder itself. <strong>The premise forced</strong> &mdash; 2,000 gold answers cached in the write-back's own document shape &mdash; later traffic answered from history at &ge;0.8 confidence: <strong>19.72% against the 80% bar, four times under</strong>, at accuracy <strong>0.9953</strong> against the 0.97 bar, which <em>met</em>. The reading, not the excuse: the confident slice is easily accurate enough, but 2,000 answers at k=10 only clear 0.8 confidence on a fifth of Banking77-shaped traffic. The share bar needs a far larger cache, or the hybrid neighbours, before it is reachable &mdash; and until #1094 is fixed the cache cannot grow at all.</p>
+
+ <h3 class="scene" style="font-size:var(--fs-20); margin-top:var(--sp-8);">The local judge (#1060): loses to its own first stage, everywhere measured</h3>
+ <div class="vt-wrap">
+ <table class="vt">
+ <thead><tr><th>nDCG@10 · 3 runs each, spread 0.0000</th><th>First stage</th><th>+ judge</th><th>&Delta;</th><th>Bar</th><th>Verdict</th></tr></thead>
+ <tbody>
+ <tr><td>SciFact · 300 q</td><td class="n">0.7045 hybrid</td><td class="n l">0.5967</td><td class="n l">&minus;0.1077</td><td class="t">beat hybrid beyond spread</td><td class="n l">FAIL</td></tr>
+ <tr><td>NFCorpus · 323 q</td><td class="n">0.3419 hybrid</td><td class="n l">0.2924</td><td class="n l">&minus;0.0495</td><td class="t">beat hybrid beyond spread</td><td class="n l">FAIL</td></tr>
+ <tr><td>FiQA · 648 q</td><td class="n">0.2382 BM25</td><td class="n l">0.1650</td><td class="n l">&minus;0.0732</td><td class="t">&ge;0.30 (hosted Jev: 0.3638)</td><td class="n l">FAIL</td></tr>
+ <tr><td>Added latency, p50, top-30</td><td class="n" colspan="3">+0.7 ms / +0.5 ms / +0.6 ms</td><td class="t">&le;40 ms</td><td class="n w">PASS</td></tr>
+ </tbody>
+ </table>
+ </div>
+ <p class="wide">Why it loses, plainly: the shipped judge is a <strong>lexical scorer</strong> &mdash; window-local BM25 saturation with page-local IDF, no semantic signal. Both first stages already read those words: BM25 ranked by them, hybrid fused them with MiniLM vectors. Re-reading the same words over the page with page-local statistics destroys information the first stages had &mdash; corpus-level IDF and the vector arm &mdash; so the judged order is a worse order. The effect is biggest exactly where the first stage is strongest. The gate arms carried no <code class="inline">min_p</code>, because a threshold can only remove recall and cannot be part of a passing attempt; an informational arm with <code class="inline">min_p: 0.5</code> kept 0.1 of 30 hits per page on SciFact and emptied 274 of 300 queries. The stage stays what PR #1077 shipped: opt-in, named <code class="inline">lexical</code> in every response, <strong>no quality claim</strong> &mdash; its own loses-does-not-ship rule, honoured. One cell could not be measured at all: hybrid+judge on FiQA, because the hybrid first stage runs <strong>13&ndash;26 s per query</strong> on the 57,638-doc index &mdash; <a href="https://github.com/xerj-org/xerj/issues/1091" rel="external">issue #1091</a>, found by this gate.</p>
+ <p class="src">Stemming: <a href="https://github.com/xerj-org/xerj/tree/main/benchmarks/beir-hybrid/results/2026-09-30-stemming-1059" rel="external">results/2026-09-30-stemming-1059</a> (PR #1085) ·
+ decide + flywheel: <a href="https://github.com/xerj-org/xerj/pull/1088" rel="external">PR #1088</a> —
+ <code class="inline">benchmarks/decisions-as-retrieval/results/2026-09-30-rc80gates/</code>, incl. <code class="inline">run-2026-09-30.md</code> and <code class="inline">stock-binary-noop.txt</code> ·
+ judge: <a href="https://github.com/xerj-org/xerj/pull/1090" rel="external">PR #1090</a> —
+ <code class="inline">benchmarks/beir-hybrid/results/2026-09-30-judge-gate/</code>, incl. the aborted FiQA hybrid log behind #1091.</p>
+</section>
+
+<!-- ============================ 04 · WHAT THE FAILURES BOUGHT ============================ -->
+<section class="scene-section" id="next">
+ <div class="kicker"><span class="accent">04</span><span class="dash">·</span><span>THE NEXT WINDOW · WHAT rc.81 DOES ABOUT EACH LOSS</span></div>
+ <h2 class="scene" style="font-size:var(--fs-56);">EVERY LOSS HAS<br><span class="accent">A FILED ISSUE.</span></h2>
+ <p class="wide">A failed gate that files nothing is a buried gate. These filed four, plus two direction changes the numbers forced:</p>
+
+ <div class="vt-wrap">
+ <table class="vt">
+ <thead><tr><th>The loss</th><th>The filed work</th></tr></thead>
+ <tbody>
+ <tr><td>silent 0-hit on unknown fields (found by the map gate)</td><td><a href="https://github.com/xerj-org/xerj/issues/1093" rel="external">#1093</a> &mdash; <code class="inline">QueryError::UnknownField</code> is constructed nowhere today; wrong fields must 4xx, not 200-empty</td></tr>
+ <tr><td>flywheel freeze (12 of 2,000 cached)</td><td><a href="https://github.com/xerj-org/xerj/issues/1094" rel="external">#1094</a> &mdash; fix the write-back rule so the cache can grow; the 19.72% share measurement is then re-run on a live cache</td></tr>
+ <tr><td>hybrid 13&ndash;26 s/query on FiQA (blocked the judge's FiQA hybrid cell)</td><td><a href="https://github.com/xerj-org/xerj/issues/1091" rel="external">#1091</a> &mdash; first-stage cost on the 57,638-doc index; the hosted-comparison row needs it</td></tr>
+ <tr><td>analyzer invisible in settings/_analyze</td><td><a href="https://github.com/xerj-org/xerj/issues/1092" rel="external">#1092</a> &mdash; echo the declared analysis block</td></tr>
+ <tr><td>lexical judge loses everywhere</td><td>the model arm: a judge that ships must be the <code class="inline">decide-local</code> cross-encoder head, beating these same hybrid numbers under this same harness &mdash; the bar does not move</td></tr>
+ <tr><td>the 15 unstemmable zero-hit queries</td><td>the vector arm &mdash; the only repair for words absent from the corpus in every inflected form</td></tr>
+ </tbody>
+ </table>
+ </div>
+
+ <p class="wide">Two more states, stated so nobody has to guess. The open model: <strong>xerj-decide-v1 is 7,419,907 parameters</strong>, its exports are byte-identical across runs, its eval card carries the weak numbers above in it, and the Hugging Face publication is staged behind an operator credential &mdash; <strong>no URL is claimed until the page exists</strong>. And the decide-head build fix: the shipped binary must carry the <code class="inline">decide-local</code> feature so the opt-in flag is not a no-op; that change rides ahead of the cut.</p>
+
+ <div class="loss">
+ <strong>What we claim, and what we do not.</strong> Claimed: every number on this page, verbatim from the gate runs linked beside it, on release builds of main @ <code class="inline">a179e1ef3</code> measured 2026-09-30 &mdash; the tier-2 numbers on the feature build of that commit, the calibration pairs on one paid provider run whose determinism bound is stated in its provenance file. Not claimed: any quality for the judge stage, any quality for tier 2 beyond the published numbers, any token-saving result for <code class="inline">xerj_plan</code> (designed, not run), any detection-quality figure for <code class="inline">_watcher</code> (gate not yet measured), and any published URL for the model. The gates were written before the work; three said yes, three said no, one split; the repo carries all eight verdicts and the four issues the noes filed.
+ </div>
+
+ <p class="wide">If you want to check a number rather than read about it: each gate directory above carries its raw per-query outputs, its manifest and its harness, and the four gate PRs (#1086, #1088, #1090, and #1085's merged artifacts) are diffable line by line. The engine itself is <a href="/docs/install">one binary</a> with a <a href="/docs/quickstart">quickstart</a>, the <a href="/playground">playground</a> needs no install, and <a href="/benchmarks/">the benchmarks hub</a> keeps the standing boards. The nearest prior chapter is <a href="/blog/does-xerj-beat-jev">the Jev verdict</a> &mdash; the 0.3109 ECE this window's calibration gate finally answered.</p>
+ <p class="src">Everything on this page: the gate runs cited section by section above ·
+ program <a href="https://github.com/xerj-org/xerj/discussions/1054" rel="external">#1054</a> ·
+ gate PRs <a href="https://github.com/xerj-org/xerj/pull/1079" rel="external">#1079</a> <a href="https://github.com/xerj-org/xerj/pull/1085" rel="external">#1085</a> <a href="https://github.com/xerj-org/xerj/pull/1086" rel="external">#1086</a> <a href="https://github.com/xerj-org/xerj/pull/1087" rel="external">#1087</a> <a href="https://github.com/xerj-org/xerj/pull/1088" rel="external">#1088</a> <a href="https://github.com/xerj-org/xerj/pull/1090" rel="external">#1090</a> ·
+ issues <a href="https://github.com/xerj-org/xerj/issues/1091" rel="external">#1091</a>&ndash;<a href="https://github.com/xerj-org/xerj/issues/1094" rel="external">#1094</a> ·
+ related: <a href="/blog/does-xerj-beat-jev">the Jev verdict</a> · <a href="/benchmarks/">the benchmarks hub</a>.</p>
+</section>
+
+</main>
+
+<!-- seo:footer-begin — generated by scripts/seo/fix_links.py · do not hand-edit -->
+<nav class="footer-map" aria-label="Site map">
+ <div class="fm-col">
+ <span class="fm-h">PRODUCT</span>
+ <a href="/product">PRODUCT</a>
+ <a href="/architecture/">ARCHITECTURE</a>
+ <a href="/benchmarks/">BENCHMARKS</a>
+ <a href="/pricing/">PRICING</a>
+ <a href="/security/">SECURITY</a>
+ <a href="/playground">DASHBOARDS</a>
+ <a href="/demo/">REAL-DATA DEMO</a>
+ <a href="/aise-demo">PRODUCT TOUR</a>
+ </div>
+ <div class="fm-col">
+ <span class="fm-h">SOLUTIONS</span>
+ <a href="/solutions/">SOLUTIONS</a>
+ <a href="/use-cases">USE CASES</a>
+ <a href="/case-studies">CASE STUDIES</a>
+ <a href="/industries/">INDUSTRIES</a>
+ <a href="/industries/finserv">FINANCIAL SERVICES</a>
+ <a href="/industries/public-sector">PUBLIC SECTOR</a>
+ <a href="/industries/healthcare">HEALTHCARE</a>
+ <a href="/industries/retail">RETAIL</a>
+ <a href="/use-cases/code-security-audit">AI CODE-SECURITY AUDIT</a>
+ <a href="/use-cases/semantic-analytics">SEMANTIC ANALYTICS</a>
+ </div>
+ <div class="fm-col">
+ <span class="fm-h">DOCS</span>
+ <a href="/docs/">DOCUMENTATION</a>
+ <a href="/docs/quickstart">QUICKSTART</a>
+ <a href="/docs/install">INSTALL</a>
+ <a href="/docs/migration-from-es">MIGRATE FROM ELASTICSEARCH</a>
+ <a href="/docs/recipes/">RECIPES</a>
+ <a href="/docs/agents/">DOCS FOR AI</a>
+ <a href="/for-agents">FOR AI AGENTS</a>
+ <a href="/agent-search/">AGENTIC SEARCH DEMO</a>
+ <a href="/answers/">ANSWERS</a>
+ <a href="/compare/">COMPARISONS</a>
+ </div>
+ <div class="fm-col">
+ <span class="fm-h">RESOURCES</span>
+ <a href="/resources/">SALES KIT</a>
+ <a href="/blog/">BLOG</a>
+ <a href="/brand">BRAND</a>
+ <a href="/brandbook/">BRAND BOOK</a>
+ <a href="/llms.txt">LLMS.TXT</a>
+ <a href="https://github.com/xerj-org/xerj" rel="external">GITHUB</a>
+ </div>
+</nav>
+<!-- seo:footer-end -->
+<footer class="footer">
+ <span>XERJ.AI · V1.0.0-RC.78 · 2026</span>
+ <span>SEARCH FOR THE AGENT ERA</span>
+ <span><a href="/product">PRODUCT</a> · <a href="/docs/">DOCS</a> · <a href="/brand">BRAND</a> · <a href="/playground">DASHBOARDS</a> &middot; <a href="/llms.txt">LLMS.TXT</a></span>
+</footer>
+
+<script>
+ (function () {
+ const root = document.documentElement;
+ for (const btn of document.querySelectorAll('[data-theme]')) {
+ btn.addEventListener('click', () => {
+ const t = btn.getAttribute('data-theme');
+ root.setAttribute('data-theme', t);
+ for (const b of document.querySelectorAll('[data-theme]')) {
+ b.classList.toggle('active', b.getAttribute('data-theme') === t);
+ }
+ });
+ }
+ })();
+</script>
+
+</body>
+</html>
diff --git a/landing/sitemap.xml b/landing/sitemap.xml
index 6f1b002af..f936805bf 100644
--- a/landing/sitemap.xml
+++ b/landing/sitemap.xml
@@ -334,7 +334,7 @@
</url>
<url>
<loc>https://xerj.org/blog/</loc>
- <lastmod>2026-09-29</lastmod>
+ <lastmod>2026-09-30</lastmod>
</url>
<url>
<loc>https://xerj.org/blog/cutting-the-index-28-percent</loc>
@@ -352,6 +352,10 @@
<loc>https://xerj.org/blog/one-vulnerability-three-names</loc>
<lastmod>2026-09-29</lastmod>
</url>
+ <url>
+ <loc>https://xerj.org/blog/the-rc80-gates</loc>
+ <lastmod>2026-09-30</lastmod>
+ </url>
<url>
<loc>https://xerj.org/brand</loc>
<lastmod>2026-09-27</lastmod>
diff --git a/scripts/seo/pagedata.py b/scripts/seo/pagedata.py
index 825dd1c60..0c331c81d 100755
--- a/scripts/seo/pagedata.py
+++ b/scripts/seo/pagedata.py
@@ -136,11 +136,21 @@ PAGES: dict[str, dict[str, str]] = {
description="The same Rust vulnerability arrives as RUSTSEC, GHSA and CVE. We collapsed 4,107 advisories to 1,950 identity-resolved records, packed them, and signed the pack — because no vulnerability database we surveyed does.",
published="2026-09-27", updated="2026-09-27",
),
+ # The rc.80 gates write-up, 2026-09-30 — the blog half of the packaging
+ # rule for the ten-item program (#1055-#1064): every gate verdict, wins
+ # and losses, each number traced to its gate run. The window was planned
+ # as rc.79 + rc.80 + v1.1.0 and measured as one suite; the post says so.
+ "blog/the-rc80-gates.html": dict(
+ label="The rc.80 gates", kind="article",
+ title="The rc.80 gates: three passed outright, three failed, one split",
+ description="Ten features, eight measured gates: /_ask macro-F1 0.9975, xerj_map 20/30 to 0/30, calibration ECE 0.0088 pass; the judge, the flywheel and the decide head fail their bars, numbers published.",
+ published="2026-09-30", updated="2026-09-30",
+ ),
"blog/index.html": dict(
label="Blog", kind="collection",
title="XERJ.ai — Blog: the engineering log",
- description="Measured results with the losses left in: the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.",
- updated="2026-09-27",
+ description="Measured results with the losses left in: the rc.80 gates post, the Jev decisions study, the JEV rerank verdict, the storage formats that cut the index 28%, and the signed rust-vulns corpus pack.",
+ updated="2026-09-30",
),
"demo/index.html": dict(
label="Real-data demo", kind="software",