<?xml version="1.0" encoding="utf-8"?>
<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Yoshimitsu Systems · Engineering Notes</title>
  <subtitle>Sourced, reproducible technical notes, each with code you can run.</subtitle>
  <link href="https://www.yoshimitsu-systems.com/notes/feed.xml" rel="self" type="application/atom+xml" />
  <link href="https://www.yoshimitsu-systems.com/notes/" rel="alternate" type="text/html" />
  <id>https://www.yoshimitsu-systems.com/notes/</id>
  <updated>2026-10-19T00:00:00Z</updated>
  <author><name>Yoshimitsu Systems Engineering</name></author>

  <entry>
    <title>Tomorrow's air: a dilated CNN against the forecasts it has to beat</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0006-dilated-cnn-day-ahead-co2.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0006-dilated-cnn-day-ahead-co2.html</id>
    <published>2026-10-19T00:00:00Z</published>
    <updated>2026-10-19T00:00:00Z</updated>
    <author><name>Morimoto Kenji</name></author>
    <category term="YS·FLOOR" />
    <summary>Can a WaveNet-style dilated causal CNN forecast tomorrow's office CO2 from last week's readings? Tested against seven no-training baselines, linear models and a GRU on 13 real offices, with holidays held out. It works, mostly on empty days, a GRU does better, and the calendar matters more than the network. With code.</summary>
  </entry>

  <entry>
    <title>What a 40-year MTBF actually promises, checked against real failure data</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0005-what-a-40-year-mtbf-promises.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0005-what-a-40-year-mtbf-promises.html</id>
    <published>2026-10-06T00:00:00Z</published>
    <updated>2026-10-06T00:00:00Z</updated>
    <author><name>Declan Lebowski</name></author>
    <category term="YS·CORE" />
    <summary>MTBF is a rate, not a lifetime. What a 40-year MTBF promises for a fleet, how vendors arrive at the number, and what 13 years of Backblaze data on half a million hard drives shows about the constant-rate assumption behind it. With code.</summary>
  </entry>

  <entry>
    <title>Spares for a one-way trip</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0004-spares-for-a-one-way-trip.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0004-spares-for-a-one-way-trip.html</id>
    <published>2026-10-05T00:00:00Z</published>
    <updated>2026-10-05T00:00:00Z</updated>
    <author><name>Declan Lebowski</name></author>
    <category term="YS·CORE" />
    <summary>How many spare parts 225 compute cores need for a 40-year trip with no resupply: Poisson sparing, spare modules versus spare cores, k-of-n tolerance, and why an uncertain failure rate costs more than bad luck. With a calculator and code.</summary>
  </entry>

  <entry>
    <title>What the benchmark threw away: classifying variable stars without their periods</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0003-light-curves-without-periods.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0003-light-curves-without-periods.html</id>
    <published>2026-08-05T00:00:00Z</published>
    <updated>2026-10-04T00:00:00Z</updated>
    <author><name>Takeda Aoi</name></author>
    <category term="YS·NAV" />
    <summary>Pretrained image models on pictures of light curves against MiniROCKET and HIVE-COTE 2 on StarLightCurves, a fair head-to-head, and an OGLE-IV rebuild of the dataset that keeps each star's period to measure what the period is worth.</summary>
  </entry>

  <entry>
    <title>Navigating by parallax: how New Horizons found itself among the stars</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0002-navigating-by-parallax.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0002-navigating-by-parallax.html</id>
    <published>2026-06-05T00:00:00Z</published>
    <updated>2026-10-03T00:00:00Z</updated>
    <author><name>Takeda Aoi</name></author>
    <category term="YS·NAV" />
    <summary>Two stars, one camera, 47 au from home: the first interstellar navigation fix, reproduced from Gaia DR3, with the geometry, the error budget, the best reference stars, and code you can run.</summary>
  </entry>

  <entry>
    <title>Homa vs TCP: the transport that sends the shortest message first</title>
    <link href="https://www.yoshimitsu-systems.com/notes/ys-en-0001-homa-tcp-replacement.html" rel="alternate" type="text/html" />
    <id>https://www.yoshimitsu-systems.com/notes/ys-en-0001-homa-tcp-replacement.html</id>
    <published>2026-04-05T00:00:00Z</published>
    <updated>2026-10-03T00:00:00Z</updated>
    <author><name>Declan Lebowski</name></author>
    <category term="YS·CORE" />
    <summary>Why scheduling the shortest message first beats fair sharing, what Homa measured against TCP on real hardware, what is still unsolved, and where Linux upstreaming stands in 2026. With a simulation you can run.</summary>
  </entry>
</feed>
