<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Triage Index</title>
  <subtitle>Releases, findings and corrections</subtitle>
  <link rel="self" type="application/atom+xml" href="https://triageindex.org/atom.xml"/>
  <link rel="alternate" type="text/html" href="https://triageindex.org/"/>
  <id>https://triageindex.org/</id>
  <updated>2026-11-10T03:53:25Z</updated>
  <author><name>Triage Index</name><email>contact@triageindex.org</email></author>
  <entry>
    <id>tag:triageindex.org,2026-11-10:release:2026-11-measured2</id>
    <title>Release 2026-11-measured2: Second measured run</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/releases/2026-11-measured2/"/>
    <updated>2026-11-10T03:53:25Z</updated>
    <summary>Fifteen further dilemmas of the intended battery, none of them asked in the earlier releases, in English and Chinese, three wordings each, for fourteen registered configurations. All 7,784 planned calls were made, and each record lists its own coverage. It is shown beside the earlier releases and never pooled with them.</summary>
  </entry>
  <entry>
    <id>tag:triageindex.org,2026-10-29:correction:b6d57d00e0fb</id>
    <title>Correction, 2026-10-29: Founding white paper (documents/triage-index-founding-white-paper.pdf), chapter The subject is the system as tested</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/corrections/"/>
    <updated>2026-10-29T00:00:00Z</updated>
    <summary>The paper said every system in the first wave was asked at a fixed sampling temperature. Five configurations in release 2026-10-measured1 were sent no temperature: Claude Fable 5.1, Claude Opus 5.5, GPT-5.6 Luna, GPT-5.6 Sol and GPT-6 Astra. The section now says that the sampling settings of each configuration, including the temperature where one is sent, are recorded. Each record shows the temperature as sent or as not sent; the published PDF stays as issued. A provider route for those five models did not accept a temperature setting, so none was sent; the records were always correct and the paper sentence was not.</summary>
  </entry>
  <entry>
    <id>tag:triageindex.org,2026-10-29:correction:4dbef840748f</id>
    <title>Correction, 2026-10-29: Releases 2026-10-measured1 and 2026-10-measured1c, the five deviation files (deviations/DEV-*.md)</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/corrections/"/>
    <updated>2026-10-29T00:00:00Z</updated>
    <summary>The deviation files show unfilled template fields, in double curly brackets, where the number of configurations, planned calls, response limits, cut answers and validation thresholds should appear. The files stay online unchanged. The same deviations, with every value filled in, are in each release&#x27;s report (report.md), which is the text to read. The step that wrote the deviation files did not fill its fields; the report filled the same fields correctly.</summary>
  </entry>
  <entry>
    <id>tag:triageindex.org,2026-10-29:correction:2af40ea822af</id>
    <title>Correction, 2026-10-29: Release 2026-10-measured1, report.md, opening paragraph</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/corrections/"/>
    <updated>2026-10-29T00:00:00Z</updated>
    <summary>The opening paragraph counts all calls, including open-ended ones, while the per-model tables count structured calls only, and the paragraph did not say so: &quot;Muse Spark 1.3, 181 empty answers&quot; and &quot;Solar Pro 4, 5 errors (HTTP 404)&quot;. Read those two counts as: Muse Spark 1.3, 181 empty answers (166 on the scale, 15 open-ended); Solar Pro 4, 5 errors (HTTP 404; 3 on the scale, 2 open-ended). No number was wrong, and the report stays online unchanged. The two counts use different denominators and the label did not name them.</summary>
  </entry>
  <entry>
    <id>tag:triageindex.org,2026-10-26:release:2026-10-measured1c</id>
    <title>Release 2026-10-measured1c: Follow-up run with larger response limits</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/releases/2026-10-measured1c/"/>
    <updated>2026-10-26T17:27:05Z</updated>
    <summary>A separately registered follow-up of release 2026-10-measured1. The configurations whose calls there were refused by the price cap or ran out of their response limit were asked the same dilemmas again, each under a new configuration identity with a larger response limit.</summary>
  </entry>
  <entry>
    <id>tag:triageindex.org,2026-10-26:release:2026-10-measured1</id>
    <title>Release 2026-10-measured1: First measured run</title>
    <link rel="alternate" type="text/html" href="https://triageindex.org/releases/2026-10-measured1/"/>
    <updated>2026-10-26T17:27:05Z</updated>
    <summary>Three dilemmas planned in English and Chinese for every registered configuration. Collection stopped early under the run&#x27;s registered stop rule for consecutive failed calls, so not every configuration was asked every planned question and language: each record lists its own coverage. Every registered configuration keeps its record, and calls that were never sent are listed as missing cells with their reason.</summary>
  </entry>
</feed>
