# Gameplay review: the mechanics, and nothing else A third reviewer, brought in for one job: judge the **game**, not the package. I do not care here about the art, the eight languages, the share card, the audio, or the achievement icons, other reviews cover those and they cover them well. I care about the decision the player makes, over and over, and whether it is a good one: is it a real choice with skill in it, does mastering it feel like mastering something, does it get harder in ways that demand more of the player, and do the pieces (the loop, the doubt prompt, the touch layer, the endings, the collection) form a system that talks to itself. I read the resolution code, then I played to lose and played to win and watched what my own decisions were actually made of. **Verdict: 6.5/10 as a game.** The core is a clean, honest, genuinely unusual skill test with no exploit in it, and it is also almost the entire game, run at one difficulty of *kind* from start to finish, with a second verb bolted alongside it that never once feeds back in. What is here is well-made. There is just not very much of it that is *mechanical*, and the parts that look like systems (the doubt rating, the false exit, the collection) turn out, under the hood, not to change the decision you are making. It plays like an excellent prototype of one mechanic that has not yet been built into a game *of* mechanics. ## What is mechanically right - **The truth is clean and unexploitable.** `correct = (choice == "back") == has_anomaly`, resolved server-side, with the answer stripped from every payload. There is no page-inspection cheat, no tell in the DOM, no way to game it except by actually remembering. For a browser game about honesty with yourself, the integrity of the core is exactly right. - **It is fair in the way that matters.** If the changed property would not have been described this loop, the build swaps it into the shown set (`hallway.py`), so you are never asked to catch a change you were structurally incapable of seeing. That single guard is the difference between a memory test and a coin flip, and the team got it right. - **The skill it tests is real and uncommon.** Because the prose rotates through equivalent sentences and only four to six of eight properties appear each loop, you cannot diff text, you must hold the *meaning* of a place and compare against it. That is a legitimate cognitive skill and I have not played much that tests it this cleanly. Memory load is real and well-tuned. - **The red herring is the best-designed mechanic in the game.** A sensory detail that reads as evidence and is deliberately uncorrelated with the truth (the poster is freshly printed; it means nothing) attacks the exact failure mode of the core skill, mistaking plausibility for change. That is a mechanic that makes the central decision *harder in kind*. There should be ten more like it. ## Where it falls down as a game ### 1. The one real decision has no in-game inputs Everything you need to make the call lives in your biological memory and nowhere on the screen. That is the concept, and I respect it, but mechanically it means there is no *reasoning* in the moment, only *recall*. You cannot deduce, combine partial clues, hedge, or reconstruct; you either remember or you don't, and then you press. A great decision mechanic gives the player levers to think with. This one gives them a yes/no and asks their hippocampus to do the rest. The ceiling on skill expression is therefore your memory, flat, with no technique to develop on top of it. ### 2. The "how certain are you?" prompt is not a mechanic It looks like risk/reward, it is not. Confidence never enters the correctness check; rating yourself *guessing* just buys one more look, and rating yourself *certain* just commits. There is no bet, no stake, no cost to overconfidence and no reward for calibrated doubt. This is the single biggest missed opportunity in the design: a game *about* doubt has a doubt button that does not do anything to your outcome. Make certainty a wager, more progress for a confident correct call, a steeper fall for a confident wrong one, and you would turn a pacing speed-bump into the game's second real decision. ### 3. Difficulty changes the odds, never the task From level 1 to level 8 exactly one thing moves: `anomaly_chance`, ~0.18 up to a 0.55 cap. The *decision* is identical at every depth; only the base rate shifts. Worse, that shift is meta-gameable, a player who knows later loops are likelier to have changed can lean on the base rate instead of their memory, which subtly *rewards knowing the ramp over playing the game*. Difficulty should add kinds of problem: two simultaneous changes, a change that reverts to baseline a loop later, a decoy property that legally drifts every loop so it is always suspicious and never wrong, a "name the property that changed" confirmation. The engine already picks a property and a value; several of these are days of work, not rewrites. ### 4. The second verb never touches the first Touching the room (Act 2) is explicitly walled off from the go-on/turn-back resolution, on purpose. As a systems critic that wall is the problem: two verbs that never interact are not a system, they are two features sharing a screen. The collection now pays you to touch, so you touch, get lore, and make the *same untouched decision*. Imagine instead one arc where a touch legally reveals one property's history ("this was blue last loop") at a cost, now the two verbs are in tension and every touch is a real trade. As shipped, the touch is content delivery, not play. ### 5. The false exit is a blind gamble dressed as a secret The alternate endings are reached by taking a rare tempting exit, but the player has no information about whether it is real, so taking it is not a *decision*, it is pressing an unmarked button and seeing what happens. Discovering the ending once is a nice surprise; there is no skill in it, nothing to read, nothing to weigh, and no reason to ever take it twice except to tick the badge. A secret that rewards guessing rather than noticing is at odds with a game whose whole skill is noticing. ### 6. The fold is a random event, not a risk you manage The fold (a touch that resets you) fires on an RNG roll weighted by depth, not on anything the player reads or controls. You cannot see it coming, cannot play around it, cannot trade for it. Mechanically it is a tax on the exact behaviour the collection is paying you to do, the game hands you a reason to touch and a random punishment for touching, and neither is legible to the other. Pick one: make touching safe and purely rewarding, or make the fold a readable risk the player chooses to run. ### 7. The collection measures participation, not mastery Most badges are attendance: escape once, escape all three, play eight times, find the fragments. Only "flawless" (a clean run) rewards *playing well*. There is no accuracy score, no close-call meter, no measure of how good your reads were, nothing that lets a strong player see, or a weak player chase, actual mastery. A game with a skill this clean should surface a skill signal. Right now winning and barely-winning look identical on the board. ### 8. There is no mastery arc Once you have internalised "remember the meaning, watch for the one change, don't trust plausibility," you have learned the whole game; deeper play is the same skill under more repetitions, not a laddder of new techniques, tells, or reads. Expert play differs from novice play only in memory quality and calm. There is no strategic layer to grow into, which is why a session ends when your attention does, not when you have hit a wall you want to beat. ## If I were fixing the game, in order 1. **Make certainty a wager.** Tie the confidence rating to the stakes, bigger step for a confident correct call, harder fall for a confident wrong one. This turns the doubt theme into a doubt *mechanic* and gives the loop its second real decision for almost no new content. 2. **Escalate in kind.** Ship two or three new anomaly *types* (double change, revert, legal decoy, confirm-the-property) and roll them in by depth so level 8 is a different problem, not a rarer one. 3. **Let one verb cost the other.** In at least one arc, make a touch trade real information for a real price, so Act 2 finally feeds Act 1. 4. **Reward the read, not the guess.** Give the false exit and the fold a readable tell, so taking or avoiding them is a decision with skill in it. 5. **Score mastery, not attendance.** Track and surface read accuracy / close calls; give the strong player something to perfect and the board something to measure beyond "showed up." ## One line > Underneath everything is one clean, honest, unusually good decision, and the > game has built a museum around it without ever giving it a second decision to > talk to. Fix the doubt button so doubt costs something, make the difficulty > change the question instead of the odds, and let the two verbs finally touch. > The mechanic deserves to be a game.