// project
Field notes from an engineer who writes: one war story, four channels, one branded card.
The brief
The Loopdown is the writing side of the same discipline the rest of this site argues for: a lesson is pulled from a real production incident, written once, and adapted, never re-derived from scratch, for every place it will be read. 17 lessons across 8 series sit alongside a 10-piece back catalogue from before the code, all versioned in one repo with the same public/private split a codebase gets: the engine and what's published are tracked, drafts and personal notes are gitignored.
What shipped
no public push in the last 20 events
// case study
The problem
A lesson pulled from real production work is only worth writing once if it can reach LinkedIn, dev.to, Hashnode and Medium without being rewritten by hand four times, in four different voices, from scratch.
The decision
Build the adaptation as the product: one lesson goes in, four channel-shaped posts come out, each with its own generated branded SVG card, checked against a voice profile derived from the existing archive by a lint step rather than trusted on read.
The result
12 of the 17 lessons are out across all four channels so far, with the rest drafted and channel-adapted (status: ready). The engine produces four-channel output every time; the publishing backlog is the honest, unfinished part.
12 lesson published across all four channels so far
Analyst view: The Loopdown draws on 4 stack entries and links to 1 outside source.
17
lessons · 8 series
4
channels per lesson · dev.to, LinkedIn, Medium, Hashnode
10
archive pieces · from before the code
12
lesson published across all four channels so far
// design notes
A lesson is written once and adapted to LinkedIn, dev.to, Hashnode and Medium, each with its own generated branded SVG card. The adaptation is the product, not the writing. A dev.to post reads like a dev.to post and a LinkedIn post reads like a LinkedIn post, from the same source material.
Every series traces back to a specific piece of shipped work, not a generic topic: "Sensors Who Lie" is field notes from Doori's location engine, "The Coroutine Court" from the −85% crash-reduction work, "The Night Shift" from the 50%→95% GPS accuracy work, "Ghosts in the Recomposition" from the ~87% Compose migration, and "One Brain, Two Bodies" from PaymentsLab-KMP's expect/actual split across targets. The writing has somewhere real to point back to.
"Sensors Who Lie" runs longest at 5 episodes; "Chain Of Custody" runs 4; the rest are shorter. By pillar, data-integrity carries the most lessons (9 of 17): Kalman-filtered sensors and Room migrations produce more "the data lied to you and here's the invariant that catches it" moments than any other pillar the archive covers.
The generated drafts are checked against a voice profile derived from the existing archive rather than trusted on read. A lint step, not a style guide nobody reads, so a draft that sounds like a language model wrote it fails before it reaches a channel.
5 of the 17 lessons are drafted and channel-adapted (status: ready). Actually out across all four channels so far: 12, "Every filter needs a documented exception", from the sensors-who-lie series. The engine is built to produce four-channel output per lesson; the publishing backlog is the honest, unfinished part.
A consolidated back catalogue from Books Before Bros, the original blog: campus lore and short fiction predating the engineering work, 6 pieces of short fiction, 2 essays and 2 humor pieces, kept in the same repo as the lessons rather than left to rot on an old WordPress install.
The whole archive is framed as an engineer stuck in a time loop, filing field notes on the same lying systems each pass: the sensor that reports a position it cannot back up, the coroutine that outlives the screen that launched it. The conceit gives every lesson the same voice without flattening what each one is actually about.
Each recurring character personifies one failure mode rather than one lesson, so continuity builds across entries instead of resetting every post: The Concussed Witness (GPS, or any sensor that reports with total confidence and zero reliability) leads the bestiary at 5 tracked appearances, alongside The Archivist (provenance and the audit trail), Doze the Jailer (Android's background-execution limits) and The Messenger (CancellationException, forever mistaken for an assassin). A living bestiary indexes who's appeared and who's still waiting in the wings, the same continuity discipline a codebase gets from a changelog.
// under the hood
Follow the connections
Notes From The Loop