<?xml version="1.0" encoding="utf-8"?> 
<feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en">
 <title type="text">Will Lachance&#39;s Log: Posts tagged &#39;recurse&#39;</title>
 <link rel="self" href="https://wrla.ch/feeds/recurse.atom.xml" />
 <link href="https://wrla.ch/tags/recurse.html" />
 <id>urn:https-wrla-ch:-tags-recurse-html</id>
 <updated>2026-10-10T21:03:33Z</updated>

 <entry>
  <title type="text">IDRs as records of intent</title>
  <link rel="alternate" href="https://wrla.ch/log/2026/10/idrs-as-records-of-intent/?utm_source=recurse&amp;utm_medium=Atom" />
  <id>urn:https-wrla-ch:-log-2026-10-idrs-as-records-of-intent</id>
  <published>2026-10-10T21:03:33Z</published>
  <updated>2026-10-10T21:03:33Z</updated>
  <author>
   <name>Will Lachance</name></author>
  <content type="html">
&lt;p&gt;&lt;img src=&#34;/log/2026/10/idrs-as-records-of-intent/./beaver_pond.jpg&#34; alt=&#34;Beaver Pond&#34; /&gt;&lt;/p&gt;
&lt;p&gt;In &lt;a href=&#34;/log/2026/08/writing-the-docs-2026-edition/&#34;&gt;Writing the Docs: 2026 Edition&lt;/a&gt;, I talked a little bit about &lt;a href=&#34;https://github.com/wlach/idr-tools&#34;&gt;IDRs&lt;/a&gt; (Implementation Decision Records, a pun on &amp;quot;Architecture Decision Records&amp;quot; which predate adoption of LLMs).&lt;/p&gt;
&lt;p&gt;As a refresher, I created them to serve a couple purposes:&lt;/p&gt;
&lt;ol&gt;
&lt;li&gt;To get better output out of LLMs&lt;/li&gt;
&lt;li&gt;To provide a durable record of a decision made during the implementation of a feature&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;I&#39;ve realized that left to their own devices given a vague prompt, an LLM will often just try to fill one in with a textual description of what it planned to implement anyway.&lt;sup class=&#34;footnote-ref&#34;&gt;&lt;a href=&#34;#fn-idrs-as-records-of-intent-1&#34; id=&#34;fnref-idrs-as-records-of-intent-1&#34;&gt;[1]&lt;/a&gt;&lt;/sup&gt;  This makes them pretty lossy as a record of intent (it&#39;s basically not much better than a diff, and arguably worse) and I&#39;m somewhat skeptical it improves their output much.
At best, they might allow you to steer them towards a better outcome before too much code is written.&lt;/p&gt;
&lt;p&gt;A great example is this one, where I asked the agent to update &lt;a href=&#34;https://github.com/wlach/repo-parser&#34;&gt;repo-parser&lt;/a&gt; to use DuckDB for the intermediate representation:&lt;/p&gt;
&lt;div class=&#34;brush: md&#34;&gt;&lt;div class=&#34;colorful&#34;&gt;&lt;pre&gt;&lt;span&gt;&lt;/span&gt;&lt;span class=&#34;gh&#34;&gt;# DuckDB serialization&lt;/span&gt;

Owner: Will Lachance &amp;lt;wlach@protonmail.com&amp;gt;

&lt;span class=&#34;gu&#34;&gt;## Overview&lt;/span&gt;

&lt;span class=&#34;gu&#34;&gt;### Problem Statement&lt;/span&gt;

repo-parser extracts metadata and structure from repositories into an in-memory &lt;span class=&#34;sb&#34;&gt;`Resource`&lt;/span&gt; tree, but this ephemeral representation must be re-created for each consumer. We need a persistent, queryable serialization format that allows multiple tools to consume the extracted metadata without re-scanning the repository.

&lt;span class=&#34;gu&#34;&gt;### Context&lt;/span&gt;

&amp;lt;very long winded, almost inscrutable LLM context&amp;gt;

...
&lt;/pre&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;p&gt;&lt;a href=&#34;https://github.com/wlach/repo-parser/blob/main/idrs/202601030409-duckdb-serialization-and-interactive-ui-architecture.md&#34;&gt;link&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The problem statement isn&#39;t horrible, but it&#39;s also rather vague.
I don&#39;t think it really gave the agent much of a goal aside from &amp;quot;use duckdb in some way&amp;quot; to replace some other mechanism.
I think the result was ok-ish, but I wonder a bit if the IDR had much value in that process.
It certainly doesn&#39;t help explain what the duckdb output of repo-parser is actually good for (a BM25 search index of a repository&#39;s contents, possibly TUI search tools).
And it&#39;s definitely not something I want to re-read later, I cringe looking at it now.&lt;/p&gt;
&lt;p&gt;What I&#39;m increasingly realizing is that for an IDR to be really effective it needs to keep returning to the &lt;strong&gt;intent&lt;/strong&gt; and decisions taken to support it.
Without this, it&#39;s just a long-winded way of restating the diff in english.&lt;/p&gt;
&lt;p&gt;I&#39;m much happier with this one:&lt;/p&gt;
&lt;div class=&#34;brush: md&#34;&gt;&lt;div class=&#34;colorful&#34;&gt;&lt;pre&gt;&lt;span&gt;&lt;/span&gt;&lt;span class=&#34;gh&#34;&gt;# Publish to PyPI from GitHub Actions&lt;/span&gt;

Owner: Will Lachance &amp;lt;wlach@protonmail.com&amp;gt;

&lt;span class=&#34;gu&#34;&gt;## Overview&lt;/span&gt;

&lt;span class=&#34;gu&#34;&gt;### Problem Statement&lt;/span&gt;

repo-parser packages can only be published to PyPI manually, which is time consuming, error prone and less secure than doing so via GitHub actions.

&lt;span class=&#34;gu&#34;&gt;### Context (as needed)&lt;/span&gt;

&lt;span class=&#34;gu&#34;&gt;### Goals&lt;/span&gt;

&lt;span class=&#34;k&#34;&gt;-&lt;/span&gt;&lt;span class=&#34;w&#34;&gt; &lt;/span&gt;Make it easier to publish new releases
&lt;span class=&#34;k&#34;&gt;-&lt;/span&gt;&lt;span class=&#34;w&#34;&gt; &lt;/span&gt;Improve security

&lt;span class=&#34;gu&#34;&gt;### Non-Goals&lt;/span&gt;

&lt;span class=&#34;k&#34;&gt;-&lt;/span&gt;&lt;span class=&#34;w&#34;&gt; &lt;/span&gt;Automate version bumping (will still entail a seperate commit to bump version)

&lt;span class=&#34;gu&#34;&gt;### Proposed Solution&lt;/span&gt;

Use the idiomatic best practice way of doing this, using the official packaging guide:

https://packaging.python.org/en/latest/guides/publishing-package-distribution-releases-using-github-actions-ci-cd-workflows/

In general this means adding a new GitHub workflow which ties uploads to PyPI to new releases.

&lt;span class=&#34;gu&#34;&gt;## Other reading (as needed)&lt;/span&gt;

&lt;span class=&#34;k&#34;&gt;-&lt;/span&gt;&lt;span class=&#34;w&#34;&gt; &lt;/span&gt;[&lt;span class=&#34;nt&#34;&gt;Official Guide&lt;/span&gt;](&lt;span class=&#34;na&#34;&gt;https://packaging.python.org/en/latest/guides/publishing-package-distribution-releases-using-github-actions-ci-cd-workflows/&lt;/span&gt;)
&lt;span class=&#34;k&#34;&gt;-&lt;/span&gt;&lt;span class=&#34;w&#34;&gt; &lt;/span&gt;[&lt;span class=&#34;nt&#34;&gt;Stamina&#39;s implementation&lt;/span&gt;](&lt;span class=&#34;na&#34;&gt;https://github.com/hynek/stamina/blob/b25d4bc359ff603496aafbb217ab82c5a43715a6/.github/workflows/pypi-package.yml&lt;/span&gt;) (likely best practice)
&lt;/pre&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;p&gt;&lt;a href=&#34;https://github.com/wlach/repo-parser/blob/main/idrs/202512310404-publish-to-pypi-from-github-actions.md&#34;&gt;link&lt;/a&gt;&lt;/p&gt;
&lt;p&gt;The problem statement describes the &amp;quot;itch&amp;quot; that made me want to perform the intervention.
The goals, non-goals and solution are short, punchy and explain the end result without getting into implementation details. When I go back and look at a repository later, &lt;strong&gt;this&lt;/strong&gt; is the sort of thing I want to see.
Bonus: when using an LLM it provides clearer guard-rails.
Every sentence does the work of steering behaviour towards the actual result I want to see, rather than being a set of instructions that it might misinterpret.
If I need more detail, I&#39;ll just look at the diff.&lt;/p&gt;
&lt;section class=&#34;footnotes&#34;&gt;
&lt;ol class=&#34;footnotes-list&#34;&gt;
&lt;li id=&#34;fn-idrs-as-records-of-intent-1&#34; class=&#34;footnote-item&#34;&gt;&lt;p&gt;They&#39;ll also do this if you ask them to create a commit message without strong guardrails &lt;a href=&#34;#fnref-idrs-as-records-of-intent-1&#34; class=&#34;footnote-backref&#34;&gt;↩︎&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;/section&gt;
</content></entry>

 <entry>
  <title type="text">Building a homelab</title>
  <link rel="alternate" href="https://wrla.ch/log/2026/09/building-a-homelab/?utm_source=recurse&amp;utm_medium=Atom" />
  <id>urn:https-wrla-ch:-log-2026-09-building-a-homelab</id>
  <published>2026-09-23T13:44:30Z</published>
  <updated>2026-09-23T13:44:30Z</updated>
  <author>
   <name>Will Lachance</name></author>
  <content type="html">
&lt;p&gt;Over the past half a year (starting around this post:
&lt;a href=&#34;/log/2026/03/music-updates/&#34;&gt;Music updates&lt;/a&gt;) I&#39;ve been slowly building out
a homelab.&lt;/p&gt;
&lt;p&gt;Despite being adjacent to communities where this is common (&lt;a href=&#34;https://www.recurse.com/&#34;&gt;Recurse&lt;/a&gt;, etc.),
this wasn&#39;t something I thought I would do a few years ago: too much effort
for the expected return.  A couple of things changed since then:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;My &lt;a href=&#34;https://marimo.io/blog/case-study-voltus&#34;&gt;present work at Voltus&lt;/a&gt; requires me to at least think of infrastructure
issues a fair bit more: the engineering team is still small enough that
I need to get into the weeds of how things are configured, deployed and
monitored-- not just how they work at the data and application layer.&lt;/li&gt;
&lt;li&gt;The software stack around it has gotten better: we&#39;re long past the days where
this meant dangerously exposing a raw Linux system on a router. Also there
are many more actually useful things you can &lt;em&gt;run&lt;/em&gt;.&lt;/li&gt;
&lt;li&gt;LLMs make this type of thing vastly easier to build independently with only
generalist-level knowledge.&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;The exact architecture is still a work in progress, but the basic components
are in place:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;Using &lt;a href=&#34;https://tailscale.com/&#34;&gt;Tailscale&lt;/a&gt; for the network boundary. This lets me run a
bunch of network services using my local machines without
exposing them directly to the internet. I can run services with minimal (or even no)
built-in authentication since access is controlled by the tailnet.&lt;/li&gt;
&lt;li&gt;Using an old w520 for the main &amp;quot;persistent&amp;quot; system server running the network services.
8 GB, Intel Sandy Bridge. It&#39;s underpowered by today&#39;s standards but more than enough to run the
types of things I might want to do (music servers, read-it-later, lightweight
notebooks, ai agent harnesses).&lt;/li&gt;
&lt;li&gt;Storing persistent configuration as a git repository. Mostly a bunch of
docker service definitions and systemd units, along with README.md files describing
how to set it up and run it when I inevitably forget. I&#39;m thinking about moving some
of this to &lt;a href=&#34;https://nixos.org/&#34;&gt;nix&lt;/a&gt; someday, but haven&#39;t found a compelling reason to do so other than
&amp;quot;nix is cool, I want to learn more about it&amp;quot; (maybe that&#39;s enough).&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;Here is the server in all its glory:&lt;/p&gt;
&lt;p&gt;&lt;img src=&#34;https://wrla.ch/log/2026/03/music-updates/w520.jpg&#34; alt=&#34;ThinkPad w520 running homelab&#34; /&gt;&lt;/p&gt;
&lt;p&gt;The setup so far is:&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;a href=&#34;https://jellyfin.org/&#34;&gt;jellyfin&lt;/a&gt; for music listening&lt;/li&gt;
&lt;li&gt;&lt;a href=&#34;https://readeck.org/en/&#34;&gt;readeck&lt;/a&gt; for saving useful posts to disk (after Mozilla discontinued Pocket, I determined
that I should never trust a third party service to do this again)&lt;/li&gt;
&lt;li&gt;&lt;a href=&#34;https://github.com/marimo-team/marimohub&#34;&gt;marimohub&lt;/a&gt; for exploring analytical ideas / questions in notebook form (it&#39;s
a little under-documented right now, but it supports tailscale authentication
headers out of the box, I might write more about this later)&lt;/li&gt;
&lt;li&gt;&lt;a href=&#34;https://pi-web.dev/&#34;&gt;pi-web&lt;/a&gt; for interactive coding (honestly mostly used for working on this system recursively, but maybe more one day)&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;I&#39;m pretty happy with the setup as a learning experience. I&#39;ve been using
various LLMs to build it, but try to understand their end products. Feels like
the appropriate use of technology: the configuration is mostly boilerplate,
the interesting questions for me are higher level: what does each additional piece of functionality allow me to do, what are the
risks, etc. I could of course &amp;quot;do things the hard way&amp;quot; and reproduce tailscale
configuration and systemd files from scratch but it just doesn&#39;t seem quite
worth it. At present, I don&#39;t have the ambition of becoming an infrastructure engineer.&lt;/p&gt;
&lt;p&gt;I don&#39;t really have illusions that this system is incredibly useful or life
changing in itself-- it just replaces a few cloud services I would
otherwise pay for (or be sad when they go away, as happened with Pocket). Most
of the value is in the experience of building it.&lt;/p&gt;
</content></entry>
</feed>