Before writing the real game I read about 4,000 reviews of 77 puzzle and level-editor games on here. The pattern in the ones that ship a hint-giving companion: players do not hate difficulty, they hate being handed the answer in a form that turns the puzzle into reading. The companion that explains is the companion people get angry at.
So the creature in It Tries gets verbs and no words. It walks, jumps, falls, bumps into a wall, and looks back at you. That is the entire vocabulary. It never produces text, an arrow, a number, or a path overlay.
What that constraint buys:
- Failure is the hint. Stay stuck long enough and it sets off on its own and fails in front of you. It never says where the wall is; you watched it fail, so you already know. Then it walks back and waits. The return is the invitation.
- Memory is the reward. Place the first block of a route it has already seen and it finishes that route by itself. No fanfare, no text. You taught it something, and you watch it use it.
- Zero-text completion. With no words in the feedback loop, the whole game can be finished without reading a single string. That is the accessibility feature I care most about, and it fell out of the design constraint instead of being bolted on afterwards.
What it costs, honestly: the first level has to teach everything by itself, and the palette has to carry meaning by brightness and shape because colour alone is not allowed to be the signal (no red/green pairing). And the canvas still has no screen-reader description of the grid, so a screen reader gets an empty canvas today. That is the first item on the v1.1 list.
The game (browser, no download, $0 pay-what-you-want): yangjy15033.itch.io/it-tries
Launch devlog with the full notes, including how the par numbers were proved by exhaustive search: It Tries is live
Disclosure: the code was written with an AI coding assistant. Design, levels, and par values are human-led and machine-verified. No AI-generated art, audio, or text.