Eleven stones, one glance
Our first palette looked fine and measured ΔE00 = 3.2 between tiers. Measuring the rendered colour instead of the design swatch is what fixed it.
13 October 2026 · Colour
Blog
How we build things, what we measured, and the versions that did not survive testing.
Engineering notes
Working notes from building our games: the measurements, the constants, and the designs we threw away. Written for anyone else making small web games.
Eight numbers and a stack that will not jitter
Gravity, restitution, friction, damping: the eight constants that decide whether a physics game feels like stone or like rubber.
13 October 2026 · Physics
Thousands of games, zero frames
How we pulled the rules out of the browser so the game can play itself headlessly, and the mechanic that test killed.
13 October 2026 · Testing
Sound from arithmetic
A full set of sound effects with no audio files: oscillators, an envelope, and a pentatonic scale chosen for a specific reason.
13 October 2026 · Audio
The same board for everyone, with no server
A daily challenge built from a seeded generator, with the date as the seed. Including the timezone flaw we decided to keep.
13 October 2026 · Architecture
The drop cooldown was shorter than the combo window
A constant picked by feel and a constant derived from measurement shared one file. A player pressed both at once and the combo counter ran to 34.
14 October 2026 · Engineering
Why we cap devicePixelRatio at 2
A one line cap in the render setup that took a week of arguing with myself to keep. Drawing more pixels than the CSS box can show is what blurs the thin lines.
16 October 2026 · Rendering
One input pipeline, three ways to hold a stone
Mouse, touch, and keyboard all drive the same aim line. Unifying them was less code than I feared and one scroll bug I should have seen coming.
16 October 2026 · Input
We cache every stone as a sprite, then throw it away
Each pebble is pre-rendered once into a cached canvas and blitted every frame. The cache is wiped on resize, and that turned out to be the right call.
16 October 2026 · Rendering
The score ladder was a triangle
The eleven merge values we typed by hand are triangular numbers. What that shape does to a run, and the 7,076 points hidden behind one clean sunstone.
18 October 2026 · Scoring
How we decide what to build
Picking a genre is the highest-leverage decision we make, so we try to do it on evidence instead of instinct. Below are the rules, and the numbers behind them.
How we score game genres before writing any code
A five-dimension model for choosing what to build, and the three hard rules that override the score entirely.
13 October 2026 · Strategy
The number that tells you a window is closing
1,648 products became 2,468 in thirty days. Absolute supply tells you you are late; the rate of change tells you early.
13 October 2026 · Strategy · Measurement
We are the worst judges of our own difficulty score
Four of our five genre scores describe the market. One describes us, and checking it against the others produced two results we did not want.
14 October 2026 · Strategy
When language locks a genre
We score language dependency on its own, apart from build difficulty. A high score means the genre's appeal is tied to references that do not cross a border.
16 October 2026 · Strategy
Our genre model has five scores and behaves like three
We audited the model we use to pick genres. Difficulty and compliance risk correlate at +0.809, the window column is mostly the crowding score, and the composite barely moves.
17 October 2026 · Strategy · Measurement
Player guides
5 tips to score higher in Pebble Merge
Keep the board low and wide, merge from the bottom up, use the walls as shelves — five habits that raise your score.
13 October 2026 · Strategy
Design and studio
Why we build cozy games
The case for calm games, and why making something relaxing is harder than making something exciting. Not easier.
13 October 2026 · Design
Why merge games are so relaxing
Predictable feedback, self-chosen goals, and no penalty for pausing: what the merge genre gets right about calm, and what we borrowed.
6 October 2026 · Design
The art of Pebble Merge: rough stone to cut gem
A measurable color ladder (CIEDE2000), a polish curve from matte rock to faceted gem, and why progress matters more than pretty.
29 September 2026 · Design · Art
How we designed the cascade bonus
We simulated games of real physics to check whether a combo mechanic rewards skill or luck. The answer changed what we shipped.
13 October 2026 · Design · Measurement
Browser games vs mobile apps
What a small studio gives up by not shipping to an app store, and what it gets in return.
13 October 2026 · Business
Why losing waits a second
Two timers stand between a bad spawn and game over. They exist because an instant loss feels like the board hates you, and a patient one feels like your fault.
16 October 2026 · Design
We deleted the achievement badges bad play would have earned
A hundred simulated rounds showed the longest merge chains belong to the worst player, so every badge idea built around chain length went in the bin.
17 October 2026 · Design · Measurement
The accessible thing we never built
Keyboard play and screen-reader labels are in. The one player setting we never read is the one that matters most for a game that sells itself as calm.
18 October 2026 · Design · Accessibility