2026¶

August 3, 2026
3 min read

Monthly Refresh 2026.07

keyword

flink adoption (the missing app)
the clustering-library port, no breakthrough
KB restructure (json over html)

journal

The month opened already on fire: a prod incident from late June followed me into July — a clustering node had lost its state, and even working through it with claude-code, it took a few rounds before we landed on a recovery approach.
Flink got greenlit as the answer to our OLAP mess, and by 07.17 I had the first real deployment running. A real milestone — except the one piece that would make it usable, a deployable app, was still missing a week later, on 07.24.
In parallel I kept working on a side project: porting a clustering library from zig to rust. Agents carried some of the pattern over, but "no breakthrough" is a phrase that showed up in my log two weeks running.
Opened a new project, flos, as a learning sandbox, and then just... left it there. Drafted, unjudged.
Miku Symphony 2026 at Suntory Hall on 07.05 — a great, very orthodox performance, though I still couldn't quite connect the parts of the music together.
07.09 turned into a messy debate about a paperwork gap that might affect my PR application. It surfaced once and never came back up.
The stay-home days mostly amounted to food and games — hot pot that wrecked my stomach for a few days after, a "cheap beef vs. shrimp" lesson learned the hard way, and clearing out the last of the seasonal events in ZZZ and Genshin before their version updates.
Sang for about six hours straight at Akabane on the 07.20 holiday, then had a perfectly reasonable dinner at Saizeriya.
Drafted the H2 2026 OKR, anchored in 黄勇的 OKR 实战笔记 — read once, understood a lot less than I'd hoped.
Also finished two books this month: a reread of 哈佛幸福课, and 10x程序员工作法. ~30 books logged so far this year already beats all of 2025 — turns out volume was never the gap, connecting what I've read is.
Ran exactly once, 3km, in the first week — then nothing. Four Sundays in a row of the same "not much" answer.
Restructured the whole knowledge base to run on JSON instead of scraping HTML — a real infra upgrade, not busywork.

I'm good at opening things this month and bad at closing them — flink, the clustering-library port, flos, the O3-KR3 self-assessment (already flagged overdue back in May) all stayed exactly as open as they started, some of them named the same way a week apart. Real life had the same shape: four weeks of not running, and a reading pile that's already bigger than all of 2025 but still isn't turning into anything. The two books and the KB restructure are the clean wins; everything else is a carry-over into August.

resolution

This month I finally drafted the H2 2026 OKR (Aug–Dec), so starting now, resolutions get measured against it instead of floating free:

Body (O1) — running goes from "basically zero" to an actual habit: 3 runs/week toward 500km by year end. Four weeks of "not much" was the real finding this month, not a rounding error.
Learning (O4) — stop logging more books and write synthesis notes connecting the ones already read, starting with the wealth-psychology cluster.
Work/infra (O3) — ship the missing runnable app on the flink deployment, take the clustering-library port to an actual checkpoint against the original (or shelve it honestly), give flos a real go/no-go, and finally do the O3-KR3 self-assessment that's been named since May.

https://www.gingerbill.org/article/2026/07/10/good-tools-are-invisible/
- landed the same week my astro + starlight experiment failed for being too rigid to hook into
https://antirez.com/news/169
- you can't just say "implement XYZ" — the honest read on why my clustering-library port kept stalling
https://bun.com/blog/bun-in-rust
- pre-work, trial-run, verification, dispatch — a name for the loop I'm already running on agent-built infra work
https://akitaonrails.com/en/2026/04/20/clean-code-for-ai-agents/
- KISS yet with context, a good note to close the month on

June 28, 2026
2 min read

Monthly Refresh 2026.06

keyword

the driver licence test (second try)
prod-release
typhoon and earthquake

journal

I failed the driver's licence test on Jun 1st, which felt like a head-shot — I had thought it would be quite easy.
I also tried to implement my own project. Due to my lack of experience, I had to rework a lot: I didn't have a grounded idea (the design), so the implementation (the steps) didn't fully follow it...
Several typhoons and frequent earthquakes — no big deal, but still, life wasn't always at peace.
I stayed up quite late these recent weeks: too many things, too little time.
The prod release itself was smooth, but issues still surfaced afterwards, because the test env and the prod env (running 24/7) differ.
- And I didn't have enough context or capability to write tools and investigate the issues efficiently.
I studied a few columns but didn't finish them all, and it's quite hard to say what I learned from them, even though I set OKRs before spending any time.
I switched to a new platform for reading books, thinking I might focus on one book at a time — and yes, it worked as expected. But, hmm, I still didn't finish many books this month. Well, I finally practiced the skill of quickly checking a book (cover, TOC, chapter start/end) to assess it, and that was really effective for dropping the unnecessary ones.
k3s + tailscale + traefik turned out to be a better solution than my previous k3s + tailscale-subrouter + svc: no more latency and everything runs smoother. I deployed even more services to my k3s — now 50+ pods running 24/7.