Guides, Games and Tangents
Game design and colour, storytelling and interactive fiction, productivity tools, and why online shopping is the way it is.
Kalabux: a Steampunk Paint-Factory Puzzle for People Who Love Colour
Kalabux is a steampunk paint-factory puzzle for Android and iPhone: pipe red, yellow and blue through brass machinery, mix real subtractive colours where the streams meet, and keep a small Victorian paint works running. It's free, playable in twenty languages, every advert can be skipped, and the only purchase is an optional ad-skipper.
Point Your Camera at Any Artwork. Get Instant AI Insights.
Artsplainer gives you an art expert in your pocket — point your camera at any artwork, anywhere, and get instant AI analysis in seconds.
AI Story Games That Actually Have an Ending: What to Play and What to Skip
Most AI story games wander forever with no payoff. Here's a plain comparison of what's out there and which ones actually deliver a story with shape.
Why Do Some Stories Feel Impossible to Put Down - and Others Stall After Ten Pages?
The difference between a story you can't put down and one that stalls comes down to a handful of craft decisions - mostly about stakes, forward motion, and whether your choices feel like they matter.
AI Text Adventures Compared: Which Ones Actually Remember What You Did?
Most AI text adventures forget your choices within minutes. Here's an honest look at what's out there and why coherent memory matters more than any other feature.
Why Didn't Our Prompt Improvement Replicate? Re-run the Baseline When the Model Changes
In August two narrator features, switched on together, appeared to give our AI-narrated game a large gain; re-run this month as a 2x2 ablation after the provider updated both the narrator and the judge, the gain didn't replicate and both features against neither came out within noise. Taken apart, the sensory palette showed small gains and the story critic added nothing measurable on top of it, so we kept the palette and switched the critic off. The lesson: a baseline is only valid for the model it was measured on.
What Makes a Good AI Text Adventure? We Measured It, and the Answers Weren't the Ones We Expected
Across 90 judged passages the two dominant failures of an AI-narrated text adventure were flat endings (41%) and scenes where nothing changed (29%), and both traced to one missing instruction. Fixing it measurably improved a real playthrough. A second model whispering direction to the first looked like the biggest win of all, until a re-run a month later, on the provider's updated models, failed to replicate it. A shorter prompt wrote better prose and broke something more important.
Three Primary Colours: Why Did We Ever Think That Was Enough?
Red, yellow and blue as the three primaries is a useful simplification - but it's also a bit of a fib. Here's what the old model gets right, what it quietly glosses over, and why it stuck around anyway.
Why Did Colours Get Cheaper? The Strange History of Paint You Could Actually Afford
Paint was once expensive, rationed by chemistry, and colour itself was a status symbol. Here's how we went from that to a billion-shade colour picker in your pocket.
Why Is Brown So Hard to Mix - and Why Does It Matter?
Brown isn't a colour you can just grab. It's what happens when red, yellow and blue meet - and getting the right brown means understanding how all three interact.