The record against my concepts

A calibration first

The claims being tested here are the exploratory ones stated in the introduction, a method, a set of modalities, and worked examples, not a delivery claim. Some of what follows would be damning against “I built an accessible Tetris”; against “I explored what accessible Tetris means”, several of the same findings are simply the exploration doing its job and returning results. I have tried to keep that distinction visible without using it as a shield. The annotated survey notes, with sources, are in the research notes on GitHub (opens in a new window).

Where the record supports the work

The architecture was right, and independently invented. The FORTH unified-design work is the strongest support because it is parallel invention: two groups, unaware of each other, concluded that accessible dynamic content requires an abstract game adaptively rendered per user. When independent lines of work converge on a structure, the structure is probably sound.

Epistemic action gives the sonar its theory. Kirsh and Maglio’s classic Tetris studies showed that expert players physically rotate pieces rather than mentally rotating them: external action is cheaper than internal simulation, so experts offload cognition onto the world. My musical sonar re-scanning on every rotation is this loop rebuilt in sound; the re-scan is the payoff of the epistemic action, restored in another design space. In 2009 I had the mechanism but not the name. The cognitive-science literature had the name all along.

Tetris really is visuospatial to its core. Emily Holmes’s line of work uses Tetris to disrupt traumatic-memory consolidation because the game saturates visuospatial working memory. That is independent experimental confirmation of this case study’s central premise: a sonic rendering is not translating decoration, it is re-housing the game’s entire cognitive substance.

The mainstream caught up and validated the instincts. Forza Motorsport’s Blind Driving Assists steer a racing game, a harder real-time spatial problem than Tetris, entirely through layered, player-tunable audio cues: commercial-scale proof that the layered-soundscape approach works. Celeste’s assist mode made player-controlled difficulty a celebrated design pattern, which is my mixing desk’s philosophy of information density as the player’s dial. The Last of Us Part II set full completion without sight, not a side-mode, as the industry bar, which is the full-game principle as product requirement. And the Audio Game Hub’s six-figure download counts document the demand the audio-games community has always asserted.

The one direct predecessor validates the problem by dodging it. The only published academic sound-Tetris made the game audible by simplifying it, reportedly to a single piece type. It confirms both that the problem is real and that the obvious escape route exists. The distinction I want to hold is between simplification as the accessibility solution (making the rendering work by gutting the game, a one-piece Tetris presented as Tetris made accessible) and simplification as demonstrator scope, where the techniques are designed against the real game and a stated subset of it is implemented to exhibit them. My 2009 build ran all seven pieces at dilated tempo; the coming web demonstrator will state its subset openly. The unified-design school and the disability community’s own standards both aim adaptation at the whole game. The exploratory point is to learn how far that aim can be carried, not to redefine the target downward and declare victory.

Time dilation became codified practice. The accommodation that looked most radical in 2009, stretching the game clock thirteen-fold for the sonic configuration, is now an industry pattern with a name. The Game Accessibility Guidelines carry a specific guideline to include an option to adjust the game speed (“issues with precise timing can be alleviated very effectively with a choice of game speed”); AbleGamers’ APX catalogue has a challenge pattern called Slow It Down; Celeste ships a 50–100% game-speed setting and Forza Horizon 5 goes to 40%. My exploration arrived at the pattern from first principles, the game must wait for the user, a decade before it was catalogued. The demonstrator should honour that lineage by making speed a visible, first-class percentage setting rather than a buried timeout.

The manual/automated split called the industry’s path. The 2009 work envisioned two adaptation approaches, manual (designer-built templates per capability profile) and automated (the game self-adapting), and expected the automated one to fail more often. The record since is striking. The industry’s accessibility wins have come almost entirely from the manual path of player-controlled assist modes, speed sliders, and remapping, while dynamic difficulty adjustment, a mature literature for engagement, remains conspicuously thin on disability. The assistive-technology abandonment literature suggests why: the strongest predictors of abandonment include lack of user involvement and control, which is what automated adaptation takes away. My old skepticism reads better with age than I had any right to expect. The automated path is still open, as research, with the same caution attached.

Where the record challenges it

Temporal parity: my own numbers are the sharpest challenge. My PhD build preserved every rule and ran the sonic configuration at 4,000 ms per row against the visual 300 ms, a thirteen-fold time dilation. The one academic predecessor simplified the rules; I simplified time. Read against a delivery claim that would be a failure; read against the exploratory claim it is the finding, a first measurement of where the modality boundary actually lies, which is the kind of result choosing a too-hard problem is supposed to produce. But the finding still challenges any successor. Continuing to succeed as speed increases may simply be unreachable in audio at visual tempos, and that possibility has to be inherited honestly, not promised away. Is a game whose identity includes time pressure still “the full game” when time-dilated by an order of magnitude? I do not have a clean answer. The most defensible position, and the one Celeste’s assist mode models, is that parity of rules is the goal to hold to, while parity of tempo is a per-player setting on a continuous axis, speed being itself a capability dimension. The corollary must be stated: cross-modal score comparison is then meaningless, and the sonic game is the same game played at a different point in the speed-difficulty space. An accommodation, not an equivalence.

Audio may invert the expert’s cheapest move. The epistemic-action result supports the sonar and cuts against the wider design. Physical rotation is a net win for sighted experts because feedback is instant and free. In my design, rotation triggers announcements that cost seconds: braided, serialized, possibly queued behind other updates. If rotating-to-see becomes expensive, audio players are pushed back toward the mental simulation that sighted experts demonstrably avoid; the design would punish the very strategy that makes experts expert. The consequence is a hard requirement the 2009 design never stated. The rotation-to-fit feedback loop must be the fastest path in the soundscape, a sub-second burst, always interruptible by the next rotation, whatever the general interruptibility policy says. Latency budgets, not just modality characters, belong in the capability model.

Modern AI interpretation weakens the transliteration argument, partly. My critique of assistive technology rested on Barbosa’s dictum that interpretation is bounded by the expressiveness of the interpreter’s domain models, and in 2009 that bound was crippling. In 2026, agentic overlays are making existing inaccessible games playable by having large models interpret the screen, and large models carry enormous implicit domain knowledge, including of Tetris. An LLM watching the board genuinely can narrate the game. What it cannot do is re-house it. Interpretation of the visual rendering inherits the visual information architecture, the flat rectangle and the observational stance; it can describe the board but not make it a place around the player. The first-person shift remains beyond transliteration’s reach, because it is not a description of the screen at all. But my field’s 2009-era claim that AT “cannot catch” metaphor-borne content is now only half true, and this case study’s argument must be restated more modestly. Interpretation is no longer impossible; it is second-best. Demonstrating how much better native abstract-model rendering is has become an empirical obligation rather than an assertion.

Generic HRTF will under-deliver for some players. My psychoacoustic rules assume the textbook figures, but the browser’s HRTF is one generic head, with no individualization, and real listeners vary. Elevation discrimination and front/back resolution will straggle the lab numbers for a meaningful fraction of players; my five elevation zones may still be optimistic for some ears. This does not break the design, because redundant coding was already a rule, but it promotes redundancy and per-user cue weighting from good practice to necessity, and it makes the wobble’s effectiveness something to measure, not assume.

My own palette has an internal collision. The PhD musical sonar encoded fit quality as pitch: higher is better. The new terrain scan encodes column height as pitch: higher is taller, and taller is danger. Walker’s auditory-display work established that mapping polarity is an empirical matter that designers guess wrong, and here I have two of my own metaphors assigning opposite valences to the same channel. If both survive into the build unresolved, the player meets pitch meaning “good” in one phrase and “threat” in the next. Before implementation, one of three resolutions must be chosen: separate registers/timbres for the two metaphors, retiring pitch-fit in favour of the terrain scan’s consonance-as-quality encoding, or retiring the old sonar entirely. This is the kind of defect a design review against the literature is for. I did not see it until the survey forced the two metaphors into one table.

Novel interfaces face an abandonment base rate, and adaptations carry social meaning. The assistive-technology literature reports 30–40% of devices abandoned, with the predictors being lack of user voice in selection, inadequate training, poor performance, and unmet changing needs. A seven-metaphor novel soundscape could trip every one of them. And the player-motivation research (Beeston, Power, Cairns, Barlet) found that what disabled players value most is feeling enabled, on a level footing, playing the same game as everyone else. That finding pulls in two directions for this work. It supports full-rules adaptation, and it warns that an audio Tetris experienced as the special version would miss the very thing players say they want, however good its rendering. The Sekiro discourse showed how much stigma can attach to accommodation, and Celeste’s assist-mode wording change showed the remedy is partly editorial: the language around an adaptation carries as much dignity or stigma as the mechanism. The social meaning of the sonic view is a design surface I had treated as neutral. It is not.

Nothing here is evaluated. Still. Every metaphor’s pass/fail is one designer’s introspection. For exploratory work that is a legitimate place to stop, since hypothesis generation is what exploration is for, and the field’s evaluation standards existed in 2009 (I cited them in my own reading notes) because the thesis planned a later stage it never reached. The trap to avoid now is inheritance. The demonstrator must treat “works surprisingly well” as a hypothesis it helps test, never as established fact it builds upon.

What I simply hadn’t considered

Co-design, not just evaluation. The pattern behind the recent successes is hard to miss. Forza’s blind-driving assists were built through years of collaboration with blind consultants; The Vale was developed with CNIB consultation. My plans, 2009 and 2026 alike, treated blind players as evaluators at the end of the pipeline, not designers at the start of it. Given that I work at CNIB, surrounded by the very expertise those studios sought out, this is the most embarrassing gap in the plan and the easiest to fix. The metaphor prototypes of the build’s first phase should be co-design sessions, not demos.

Motor and cognitive access. My analysis has been vision-first throughout. The standard survey framing, that a player must receive stimuli, determine responses, and provide input, exposes the other two barriers. Input: Tetris’s timing demands (auto-repeat tuning, lock delay, key dwell) are motor barriers, the one-switch community has adapted falling-block games for single-switch scanning, and my five adjustable timeouts help but no switch-input mode was ever designed. Cognition: my layer model controls information density, but rule relaxations in the Celeste sense (larger lock windows, slower progressions, undo) are a different accessibility axis that the capability model claims in principle and specifies nowhere.

The braille-graphical design space. A community project has driven a refreshable braille terminal as a 20 × 4 tactile raster for an inverted Tetris. Braille displays as low-resolution graphical devices constitute a fourth design space, tactile-graphical, distinct from the vibrotactile haptics I deferred, and my domain chart can host it as naturally as a Braille View service domain. I had never enumerated it.

Deaf players of my game. Once the functional audio is the soundtrack, game-relevant state (the danger key-shift, the urgency tempo, the lock countdown) lives in sound. A deaf or hard-of-hearing player of my accessible Tetris gets a worse game than classic Tetris unless the visual view mirrors that state visually. My own 2009 note on emotive loss in deaf adaptation predicted this failure mode, and seventeen years later I still designed it in. The visual view must render the tension system, not as an accessibility afterthought but because the architecture’s symmetry principle demands it.

The working-memory question. If visual Tetris consumes visuospatial working memory, what does audio Tetris consume? Spatialized audio may still engage spatial cognition; musical encoding may shift load to auditory-verbal systems; the truth is unknown and testable with the paradigms Holmes’s group uses. This is a genuine open research question that the accessible build makes askable for the first time. It could be the project’s most novel scientific contribution, and it was not on my list.

The name. Tetris Holding v. Xio Interactive (2012) settled that game mechanics are unprotectable ideas while their audiovisual expression is protected. That favours this project twice over: the mechanics are free to implement, and my entire purpose is a different expression. But Tetris® remains a trademark vigorously enforced. Scholarship sits comfortably under nominative use; a playable public release less so. The playable build needs its own name, with “after Tetris” in the description. A decision to make before anything ships, not after.