A weather app that shows its working.
Real Temp shows the temperature your body actually experiences on the street: sun, humidity, wind, and surroundings included, with the math visible. Most weather apps give you a number labelled "feels like" and no way to interrogate it. This one is built so the number can be argued with.
- Role
- Solo: product, design, build
- Model
- Steadman apparent temperature, bounded deltas
- Data
- Open-Meteo
- Decisions logged
- 15
A weather app that shows its working.
Real Temp shows the temperature your body actually experiences on the street: sun, humidity, wind, and surroundings included, with the math visible. Most weather apps give you a number labelled "feels like" and no way to interrogate it. This one is built so the number can be argued with.
Every weather app has a number nobody can check.
A "feels like" reading is a single opaque figure from an undisclosed model. Stand in direct sun on a still day and it is wrong in one direction; stand in a shaded wind tunnel and it is wrong in the other. The product bet is that showing the calculation is more useful than hiding it, because a number you can inspect is a number you can learn to trust or discount.
Solo: product, design, build
Solo: product, design, build
A street instrument, not a weather postcard.
The visual direction was set before the build: instrument-like rather than illustrative, with the reading and its components given equal weight. If the argument is that the math should be visible, the interface cannot treat the math as a detail view.
What remains untested.
The street check had not happened when these artifacts were written, which is the whole validation for a product about what conditions actually feel like. The shade map is specified as building footprints rather than the LiDAR pipeline it wants, so shade accuracy is approximate by design. Acclimatisation and bio-calibration are heuristics with bounded effects, not validated physiology. Three artifacts, all planning: nothing here shows a shipped app.
Every weather app has a number nobody can check.
A "feels like" reading is a single opaque figure from an undisclosed model. Stand in direct sun on a still day and it is wrong in one direction; stand in a shaded wind tunnel and it is wrong in the other. The product bet is that showing the calculation is more useful than hiding it, because a number you can inspect is a number you can learn to trust or discount.
- 01Sun exposure and shade change the felt temperature more than the forecast admits.
- 02A model that cannot be inspected cannot be corrected when it is wrong for you.
- 03Personal acclimatisation is real and no consumer app accounts for it.
A published model with bounded adjustments, not a proprietary blend
Steadman apparent temperature, then deltas with hard limits. The base is a published apparent-temperature model rather than something invented for the product, so the starting number is checkable against the literature. Sun, wind, and surroundings are applied as bounded deltas on top: each adjustment has a cap, so no single input can run away with the result. Rejected: an unbounded blended model, which produces confident numbers with no way to sanity-check them.
The name had to change before anything shipped
"RealFeel" is a trademark. The original PRD used a name already held by a major weather company. It was dropped and the codename Real Temp adopted instead. A small decision, logged because the alternative is discovering it after launch, and because the PRD it supersedes exists as a v1 PDF rather than being rewritten in place.
Platform amended: web first, then wrapped for iOS
The street check needed a phone. The first decision was a client-only web app, chosen so there was no server to run and no data to hold. It was amended days later to a wrapped iOS build, because the product only means anything when you are standing outside in the conditions it describes, and a browser tab is the wrong vehicle for that. The v1 build was then pulled forward ahead of the street check so there would be something to check with.
A street instrument, not a weather postcard.
The visual direction was set before the build: instrument-like rather than illustrative, with the reading and its components given equal weight. If the argument is that the math should be visible, the interface cannot treat the math as a detail view.
Reading, then reasoning, then adjustment.
The flow report traces how a reading is reached and how a user gets from the headline number to the components behind it, then to the personal calibration that shifts it. Acclimatisation and bio-calibration are both bounded heuristics rather than claims of precision, which the report states explicitly.
What remains untested.
The street check had not happened when these artifacts were written, which is the whole validation for a product about what conditions actually feel like. The shade map is specified as building footprints rather than the LiDAR pipeline it wants, so shade accuracy is approximate by design. Acclimatisation and bio-calibration are heuristics with bounded effects, not validated physiology. Three artifacts, all planning: nothing here shows a shipped app.
Scoped down to two epics, on purpose.
The MVP line was drawn at two epics and held. Everything past it, including the shade pipeline the product would most like to have, is written down as later rather than smuggled into v1.
Planning and early build. No shipped app, no street validation yet.