<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Emir Hüseyin İnci — engineering notes</title>
  <subtitle>Measured, dated, and corrected in public.</subtitle>
  <link rel="self" type="application/atom+xml" href="https://emirhuseyin.tech/feed.xml"/>
  <link rel="alternate" type="text/html" href="https://emirhuseyin.tech/writing/"/>
  <id>https://emirhuseyin.tech/</id>
  <updated>2026-09-16T00:00:00Z</updated>
  <author><name>Emir Hüseyin İnci</name><uri>https://emirhuseyin.tech/</uri></author>
  <entry>
    <title>Why Rust</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/why-rust/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/why-rust/"/>
    <id>https://emirhuseyin.tech/writing/why-rust/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>Not for speed. Determinism you can replay, resource bounds you can state in advance, and one compiled engine behind three surfaces — plus the hundred-fold constant error the type system was happy with.</summary>
  </entry>
  <entry>
    <title>How many days ahead is a weather forecast worth trusting?</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/forecast-trust-horizon/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/forecast-trust-horizon/"/>
    <id>https://emirhuseyin.tech/writing/forecast-trust-horizon/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>7,358,400 archived forecasts scored against a guess that needs no model, across thirty Turkish cities and two years. In eight of them the hourly rain forecast never beats “it will not rain”, at any lead from one to seven days.</summary>
  </entry>
  <entry>
    <title>The best of 1,000 coin flips beat my best trading rule</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/thousand-coins/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/thousand-coins/"/>
    <id>https://emirhuseyin.tech/writing/thousand-coins/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>Eight measurements closing a search for an edge in FX bar data. A thousand strategies that read no prices produced a best of 2,862; the best rule found by reading everything made 944, which puts it at the 90th percentile of noise.</summary>
  </entry>
  <entry>
    <title>The stamp covered the worker, not the engine</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/cache-stamp/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/cache-stamp/"/>
    <id>https://emirhuseyin.tech/writing/cache-stamp/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>A released page reported the previous version. The served bytes were correct and the cache-busting stamp was doing exactly what it had been written to do, which covered one of the three files that have to move together.</summary>
  </entry>
  <entry>
    <title>The uplift number my own product refuses to use</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/uplift-we-do-not-use/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/uplift-we-do-not-use/"/>
    <id>https://emirhuseyin.tech/writing/uplift-we-do-not-use/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>Uplift computed on a dataset with no treatment assignment describes the generator rather than the customer, so Aegis labels it simulation and excludes it from the decision path.</summary>
  </entry>
  <entry>
    <title>A reviewer that cannot say what is right</title>
    <link rel="alternate" type="text/html" hreflang="en" href="https://emirhuseyin.tech/writing/reviewer-that-must-answer/"/>
    <link rel="alternate" type="text/html" hreflang="tr" href="https://emirhuseyin.tech/tr/writing/reviewer-that-must-answer/"/>
    <id>https://emirhuseyin.tech/writing/reviewer-that-must-answer/</id>
    <updated>2026-09-16T00:00:00Z</updated>
    <published>2026-09-16T00:00:00Z</published>
    <author><name>Emir Hüseyin İnci</name></author>
    <summary>Four design decisions about the critic stage of an LLM agent pipeline. Three of them fail closed; the fourth is a heuristic with a guessed threshold, written down as a guess.</summary>
  </entry>
</feed>
