REVIEW · 2022-09-09

Taiji

One verb; the bet is on observation

Steam store ↗

First Impressions

Skim the Steam reviews for Taiji and the first thing you notice is the consistency of the comparison: in English and Japanese alike, the most frequent proper noun isn't the developer's name or a genre — it's The Witness. "2D Witness," "a chance to relive The Witness for the first time" — reviewers receive this game through the memory of that 2016 island before treating it as a new work.

The numbers: Steam's overall rating is Very Positive — about 92% of 1,102 Steam-purchase reviews are positive (across all 1,407 reviews, 1,280 positive vs 127 negative; 2026-06-10 snapshot). Released Sep 9, 2022; developed and published by Matthew VanDevander alone, at $24.99. Tags pair "Puzzle" and "Logic" with "Difficult," "Nonlinear," and "Open World."

The rule fits one sentence: click tiles to toggle them black or white, then commit. That's the whole input. What the top reviews flag is the gap between that simplicity and the total absence of any explanation of how to solve. The positive side calls the gap beautiful; the negative side calls it unfriendly. The score splits over a single design choice.

Putting the Mechanics into Words

The reviews reconstruct a single verb: toggle a tile black or white. The grid is the same everywhere; no pixel hunting, no brute-forcing options. The recurring positive line — "it's always clear what to interact with, never clear what to do" — is, in our vocabulary, thorough verb subtraction: cut the verbs to one and bet everything else on knowledge.

That knowledge arrives as a different "grammar" per area — lines and shapes, dots, diamonds, binary (really just addition, several reviewers reveal), drawn paths. In the late "Gallery," grammars share a board, and the longest top-helpful reviews name these convergence panels the highlight: solvable "one cell at a time, like sudoku." Combinatorial explosion, staged without losing a unique solution.

Tellingly, the recommending reviews often confess they hadn't really understood the rules mid-game. Vague "it feels right" progress (one long review calls it the "Monte Carlo" player) hits a wall until you go back and put the rule into words. The game deliberately separates being able to solve from understanding, and demands the latter.

Teaching

Taiji strips out text almost entirely; even new symbols are introduced as small solvable panels you experiment on to derive the rule yourself. Positive reviews prize this silent teaching — "the intro-hypothesis-test rhythm is satisfying," "the dopamine of the click lands" — tutorial folded into the levels rather than walled off.

The negative side keeps coming back to one phrase: the game "fails to teach its logic." Being open-world, nearly every area is reachable at once with no stated order; one reviewer runs "around like a headless chicken" because each grammar seems to need another first. What positive calls freedom to discover, negative reads as a missing order. Same structure, opposite face.

Unusually, the developer himself weighs in on the forums: the early "Gallery" plus the nearby black-and-white sculpture panels form a double whammy that makes many players bail before the two-hour refund mark, convinced it never adds up — he'd cut those puzzles in hindsight but won't keep editing a shipped game and lose its original cohesion. The 92% is best read as the verdict of those who got past that first hour.

The Texture of Difficulty

When "too hard" and "just right" split a game, collecting where reviewers got stuck sorts the difficulty into kinds. Here there are roughly three: the difficulty of inferring a symbol's rule; the "clerical" difficulty of large boards with long procedures once the rule is known; and the difficulty of the endgame environmental puzzles, where you can't tell where to input the answer.

The pleasure the positive side describes concentrates almost entirely on the first kind — the snap of finally wording a rule. What the negative side calls unfair is the third — the environmental puzzles. "I'd collected the answers but couldn't find where to enter them, so I checked the forum" appears in recommending reviews, in the press, and in the verdicts the developer cites alike. The seam is local, not total.

What separates the first kind from the third is observation resolution. Difficulty inside the board always yields to wording the rule; difficulty outside it — answers embedded in scenery — demands the resolution to notice a puzzle is even there, and with no hint system the last step is outsourced to the community. The Witness carried the same weakness, so this is a difficulty inherited along with the lineage.

Place in the Lineage

Few games make the predecessor this much a premise of their reviews. The developer is open about The Witness's influence, and reviewers point to beat-for-beat references; one outlet called Taiji "The Witness with the insufferable smugness removed." A first-person island translated into top-down 2D pixel art — the observation-and-inference core kept, only the texture swapped.

Reviews are specific about where it isn't mere imitation: where The Witness made you draw a line start-to-finish, Taiji lets you edit any cell anytime, which (multiple long reviews note) enables larger, sprawling "pen-and-paper" logic boards. Amid a recent glut of observation puzzlers — some, like Islands of Insight, competing on sheer volume — Taiji differentiates by subtraction: one consistent grid, not more content.

Reviews Consulted

This piece was written from the Steam store page and community reviews/forums as of 2026-06-10. No review text is quoted; typical claims are reconstructed.

Steam: Taiji (Very Positive / about 92% of 1,102 Steam-purchase reviews positive; across all 1,407, 1,280 positive vs 127 negative; 2026-06-10 snapshot)

・Read a dozen-plus top-helpful reviews (English and Japanese); negative claims reconstructed from the "Game Fails to Teach You Its Logic" General Discussions thread and low-rated reviews; recent 2024–2025 reviews consulted. Completion times from HowLongToBeat (Main 9h, All Styles 13h, Completionist 15h).

・Press consulted: Thinky Games; the developer's design intent from his own forum posts.

Closing

The Taiji the reviews describe is a near-pure instance of one design: cut the verb to one, bet the rest on observation and inference. The positive "beautiful" and the negative "unfriendly" observe the same silence from opposite sides. It's a question of scope — whom omitting explanation lets through and whom it lets go — and the developer has said he'll keep that scope fixed even post-launch.

Against Steam's 92% I put a design score of 8.5; the gap is small. The 8.5 reflects first-rate grammar design and silent teaching; what keeps it from full marks is two local seams — the early stumble and the endgame environmental puzzles, the very first and third the developer himself concedes. It doesn't reach the whole-game reversal of Baba Is You, but it carries a dozen-plus hours on the pleasure of wording rules alone — a rare thing.

Reactions (no login)

Anonymous • one of each per visitor per day

Read next

FEATURED ESSAY · 2026-06-15

Counterpoint on Lorelei and the Laser Eyes — Reading Through the Negative Reviews

Komugi rated Lorelei and the Laser Eyes 9.0/10. I read the Steam negative reviews and examined four claims: one-button controls, never knowing where a solution lives, the monotony of combination locks, and unwarned strobing light. Where I agree, where I push back.

Related reviews