← Work Hive Story
Current Work · 2026 – Present
Hive Story
Beekeeping apps are forms and charts, and a keeper closes them because nothing there answers the question they came with. Hive Story composes a colony's sensor readings, inspection logs, and treatment history into continuous prose: what state the hive is in, what to do next, and why. I set the concept and the craft standards, and shipped it in Flutter.
The Problem
- Beekeeping apps are forms and charts. A keeper opens one, sees a table of mite counts and a graph of temperatures, and closes it. The data is all there. The understanding is not.
- In 2025, U.S. beekeepers lost an estimated 55.6% of managed honey bee colonies, the highest annual loss since national tracking began. Colony loss is a slow-failure problem: mites compounding, a queen fading, stores running out. A dashboard hides a slow failure. A story shows it.
- Everything in a hive happens at once. Mites, disease, forage, weather, and the state of the queen all move together. It takes years to read that well, whether you keep two colonies or two thousand.
- Beekeepers do not agree on what good keeping is. Some treat and push for yield. Some refuse chemicals. Some run the hive as close to wild as they can. Advice that assumes one of those is the correct religion does not get used.
- The answers are already in print: more than 100,000 publications on the honey bee, plus university extension guidance. No one, and certainly no beginner, can read that in a veil and apply it.
Strategy
OSBeehives proved you can hear a colony. It did not tell a keeper what to do next, and I have spent years knowing why that failed. Hive Story starts on the other side of it: establish the actual state of each hive, then say the next action for that colony, that philosophy, and that skill level. Knowledge that used to take ten years should be available in year one.
Three commitments set the constraints. Every claim traces to a verified study or a legal product label. Logging, planning, and questions have to survive gloves and sun. And it stays free or cheap for hobbyists, funded by commercial service tiers.
The Assumption
The colony's data needs a better dashboard. Show the readings clearly enough and the keeper will know what to do.
What Was Actually True
Data is not understanding. Keepers think in colonies, seasons, and jobs to be done, so the answer is a story, not a chart: plain sentences composed from that hive's real state, with the recommendation and the receipt attached.
How It Works
What a Keeper Lands On
The colony's own page, written out. A plain-language read on where the hive is now, the next actions under it, and one tap down to the guide, the paper, or the actual product label. Depth on tap is how year-ten judgment shows up in year one, without pretending the complexity went away.
Making the Record Survive Gloves and Sun
Speak or type at the hive. The system transcribes the note, pulls out the structured facts, and asks the follow-up that practice actually needs. A manual form is there for people who want it. The design job was never adding voice. It was getting a usable record out of someone holding a smoker in one hand.
Every Recommendation Gets a Date
A recommendation saves to a day on the calendar, not to a hope that someone remembers it after the veil comes off. Treatment windows, inspections, and feeding sit on one seasonal view per colony.
Every Number Traces to a Label or a Study
Nearly 300 practices, from an oxalic dribble to an alcohol-wash mite count, written from university extension and the same verified corpus. Every guide carries the legal limits from its EPA registration and this hive's own figures on the field card, so nobody is converting anything in the yard. The catalog is still growing.
The Record Starts Before Anyone Types
The Sentry sits in the brood box, and outside APIs bring the weather and the rest of the world. Together they establish the colony's state without cracking the lid, so the story has something true to say on a day nobody opened a hive. I designed the board and the enclosure as well, which is why the surface can assume a baseline instead of an empty page.
Decision 01
Prose, Not Cards
The reflexive pattern for a data timeline is a stack of cards, and I banned it. A story read in fragments is not a story: cards chop a continuous argument into unrelated claims and force the reader to reassemble the colony in their head, which is the exact work the product exists to do for them. Hive Story renders as typeset reading instead. That puts the whole load on typography, so the surface runs on a dedicated reading-token system: a measured type scale, hanging numerals so data sits inside prose without breaking the line, hand-tuned gutters, and three spacing gaps, at the section, the cluster, and the flow, that carry the structure instead of boxes and rules. When something reads wrong here, the fix is almost always spacing, not decoration.
Decision 02
Composed From State, and It Refuses to Bluff
Every line is composed from typed colony state: brood status, mite pressure, stores, active treatments with their real windows. The composer is class-aware, so a formic acid treatment, which reaches mites under the brood cappings, produces a different narrative arc than an oxalic dribble, which cannot. The prose changes because the biology changes. Accuracy is a design property here, not a backend concern: legal limits come from FDA and EPA labels bound word for word, research is screened for quality and checked against the source by two independent systems before it can touch advice, and where the system does not know, the story says so instead of smoothing over it. One confident wrong sentence costs a keeper the colony, and costs the product every keeper who hears about it.
I Don't Hand Off
I shipped Hive Story's rendering pipeline myself in Flutter: the block renderer, the typographic spacing logic, and the theming. Then I built the verification around it, because hand-tuned values are the first thing a refactor destroys. Screenshot harnesses put every block in front of my eyes on demand, widget tests pin the gutters and the gaps so they survive the next change, and writing, fact-checking, and layout each pass an enforced gate before content reaches the surface. Those gates are also how I direct a production system of AI agents against my own standards, which turned out to be a design discipline in itself: it forced me to encode judgment I had only ever held as taste into rules precise enough that generated work either meets the bar or visibly fails it. That pipeline is the only reason a catalog heading for nearly 300 practices is possible for one person.
Where It Stands
Hive Story is the working heart of the app and the surface a keeper lands on. The guide pipeline runs, the composer runs, and the Sentry exists as a fabricated r1 board. What is not done: the boards are not yet through a season in production hives, so there are no field results to claim, and the assistant a keeper can question at the hive is next rather than shipped. I have been on this full time since February 2026, and it goes back to nights and weekends once I am employed. The shape of the work is the part that travels: an ambiguous human problem taken to the form that answers it, craft standards defined down to the gutter widths and then enforced in code, and ownership from the sensor in the brood box to the sentence on the screen.
What I’d Do Differently
At OSBeehives I had a working detector and assumed the hard part was behind me. It was not. The constraints were battery, range, and trust, and I did not treat any of them as design problems until it was too late to matter. This time the architecture starts there. The hard part has never once been the part I could see.