Update for the week ending on Friday, Aug 8, 2025

mtlynch.io

• Started writing my July 2025 retro
• Updated an old post to announce shutdown of What Got Done

Refactoring English

• Continued working on “Get to the Point” chapter
• Added an acknowledgments chapter
• Started taking notes about writing for your audience
• Emailed five readers to see if they have questions about the book.

Hit the Front Page of Hacker News

• Recovered recordings I made last year
  • I realized they were on my Windows drive and not in my backups.
  • Fortunately, they were still sitting on the drive and easy to recover.
  • They’re now backed up along with my normal backups.

HN Observer

HN Observer is a tool I’m working on to predict the outcome of submissions to Hacker News based on past historical data.

• Started the switch to DuckDB.
  • I’ve been hearing people talk about time-series databases for the past few years and never thought to ask why.
  • On a call last week, someone mentioned using a time-series database to examine data at different levels of time granularity, and I was like, “Oh! That’s probably a thing time-series database do well.”
  • I’d been struggling with switching views on HN Observer because polling HN at 1-min intervals creates a lot of data.
  • In particular, viewing aggregate scores of everything on the front page was super slow with SQLite and took about a minute to calculate.
  • I asked an LLM for suggestions of time-series databases that were like SQLite, and it suggested DuckDB.
  • The LLM basically one-shotted rewriting the app (Go) from SQLite to DuckDB.
  • The harder part was migrating several months of historical data (10 GB worth) from SQLite to DuckDB. The first attempt was going to take about 12 hours to migrate.
  • I realized that most of my SQLite data was actually extraneous.
    • For example, I’m recording every observation of HN, but if I query a story, and it has 20 upvotes, then a minute later, it still has 20 upvotes, then a minute later, still, etc. I can infer all the middle states from knowing the start and end states, as upvotes increase monotonically.
  • I had the LLM rewrite to not migrate extraneous datapoints, and it got the migration down to about one minute.
  • I still need to do a quick review of the new code and update my NixOS module to use DuckDB instead of SQLite.

Gleam Chat Log Parser

A toy project to teach myself Gleam. I’m trying to write parsers for my old AIM chat logs.

• The parser can now parse one real plaintext AIM log!
• Connected the CLI to the actual plaintext parser
  • Previously, the CLI just printed filenames, and I was only running the parser through unit tests.
• Added support for parsing lines with Windows-style line endings
  • Weird discovery: in Erlang, the sequence \r\n is considered a single character.
• Added support for unterminated sessions
  • Some files don’t have the session close footer. I suspect this is due to unexpected program shutdowns.

What Got Done

• Realized the site has been down for two weeks and I didn’t notice
• The issue was that Litestream was failing to recover from a snapshot
• I initially tried to just update to litestream 0.3.13, but the problem persisted
• I eventually had to roll back to an earlier snapshot, then push it as the latest snapshot

Misc

• Had lunch with a friend visiting from Japan.
• Took a road trip to visit a friend for their birthday.
• Downloaded a course I’d purchased.
  • It’s by default streaming-only, but I used an LLM to write a scraper that uses yt-dlp to scrape the streaming URLs and download it locally.
• Got new fire extinguishers.
  • Realized mine were 12-13 years old.
• Cold emailed an indie blogger I like.
• Listed my old managed switch for sale
  • But then I realized I listed it under the wrong product name, so I canceled and relisted.
• Reinstalled LeechBlockNG.
  • I stopped using it because it was causing hangs in Firefox, but now my browser habits are degrading, so I’m giving it another shot.