How a teardown gets made
Everything below is a rule we actually follow rather than an aspiration. Where we have broken one, we say so, and where a rule has changed we date the change. If a teardown appears to contradict something on this sheet, the sheet is wrong and we would like to hear about it.
Minimum hours before anything is published
03.1 · Time on task- Rule set
- March 2025, revised once
- Breaches
- One, logged and explained
- Median actual
- 41 hours across seven teardowns
A system cannot be described until it has been placed under load, and different genres take different lengths of time to reveal where they bend. These are floors rather than targets: the figure below is the point at which we are permitted to start writing, not the point at which we stop playing.
| Genre | Floor | Why that figure |
|---|---|---|
| Tactics and strategy | 40 hours | Growth curves need three quarters of a campaign before a mismatch becomes visible. |
| Simulation and builders | 50 hours | Technical ceilings only appear at scale, and scale takes real time to reach. |
| Roguelike and roguelite | 60 hours | Meta-progression is the system under test, and it is designed to reveal itself slowly. |
| Stealth and action | 35 hours | Enough to finish twice, which is when the second-order systems become legible. |
| Adventure and narrative | Completion, plus 4 hours | These end when they end. The extra four hours are for revisiting earlier scenes. |
The one breach. Our adventure teardown was written after twenty-two hours, which met the completion rule but included only two hours of revisiting rather than four. We published anyway because the writer was travelling. It is recorded here because a rule you quietly bend twice is not a rule.
What we test on
03.2 · Machines
Fig. B — The four machines, drawn to scale. The laptop is the oldest and does the most useful work.
Every teardown is played on at least two of these, and the stability bar is measured on the weakest machine that the game claims to support. We keep the laptop deliberately old because a game that only behaves on current hardware is a game most readers cannot use.
| Ref | Machine | Specification | Role in testing |
|---|---|---|---|
| M1 | Desktop, built 2024 | 8-core CPU, 32 GB, current mid-range GPU, NVMe storage | Primary playthroughs and any test that needs headroom. |
| M2 | Desktop, built 2021 | 6-core CPU, 16 GB, previous-generation GPU, SATA SSD | The realistic case. Most stability figures come from here. |
| M3 | Laptop, 2019 | 4-core CPU, 16 GB, integrated graphics, SATA SSD | Floor testing. If a game claims low requirements, we check them here. |
| M4 | Handheld PC, 2023 | Handheld form factor, 16 GB, 7-inch display, controller-only input | Small-screen legibility and controller-only navigation. |
We buy all four ourselves and replace them on a rolling five-year cycle. Nothing on this bench was supplied by a manufacturer, and no machine here has been provided in exchange for coverage.
What each bar measures
03.3 · ScalesFour bars, each measured separately, never averaged. Two of us score independently and the figures are reconciled in conversation; where we cannot agree within eight points, the disagreement is written into the teardown rather than split down the middle.
Counts: whether the central activity is legible, whether it offers real choices, and whether those choices have consequences you can trace. A high figure means the thing you do most often is well built.
Ignores: presentation, story, length, price. A game with a superb loop and nothing else can score above ninety here and still not be worth your evening, which is why this bar is never reported alone.
Typical range: most games we cover land between seventy and ninety. Below sixty is rare and usually means the loop is fighting itself rather than being simply dull.
Counts: whether the game is as interesting at hour forty as at hour four. We look for difficulty curves that flatten, economies that outrun their opposition, and content that repeats without variation.
Ignores: total length. A twenty-hour game that stays sharp throughout scores higher than an eighty-hour game that stops developing at hour thirty. Endurance is about holding shape, not about duration.
Typical range: this is our most variable bar and the one that most often separates two otherwise similar games. Anything below fifty means we can name the hour at which the game stopped growing.
Counts: six things, weighted equally. Remappable controls including modifier keys. Subtitle size and background contrast. Whether colour alone is used to convey required information. Whether timed inputs can be extended or disabled. Whether the interface can be operated entirely from a controller. Whether text scales without clipping.
Ignores: difficulty. An easy mode is not an accessibility provision, and we have stopped accepting it as one. Two of the six of us have specific needs here — one uses a modified controller, one relies on subtitles — and both check every game before this bar is set.
Typical range: wider than we would like, from the high twenties to the low nineties. This is the bar that has attracted the most reader disagreement, and the flight sim teardown is where that argument was had in public.
Counts: crashes, save corruption, frame pacing that changes the difficulty of an input, and load times that grow with progress. Measured on the weakest machine the game claims to support, not on our fastest.
Ignores: whether a patch is promised. We report the state of the object in front of us on the date of publication, and if a later patch changes the position we amend the teardown and log it.
Typical range: usually the highest of our four bars, which is why a low figure here is worth noticing. The transit builder's forty-one is the lowest we have recorded and it was the reason that teardown missed a sound-structure mark.
Rules we hold ourselves to
03.4 · Conduct- Review copies
- Accepted, always disclosed
- Paid coverage
- Never, in any form
- Embargoes
- Honoured, but not if conditional on tone
- We buy most of what we cover. Five of the seven teardowns on this site were written about games we paid for. Where a copy was supplied, the teardown says so in its first paragraph, not in a footnote.
- No embargo that specifies a tone. We have declined two review copies whose terms required a summary to be approved before publication. Both games were later covered after retail release, without the copy.
- No affiliate links and no advertising. There is nothing on this site to click that earns us anything, which is also why we publish three teardowns a month rather than twelve.
- Amendments are published, not silently applied. If a patch changes a bar or a reader corrects a fact, the change appears in the amendment log with a date and a description. We have made six.
- Nobody reviews a game made by someone they know. This has come up once and the teardown was reassigned. The connection was mentioned on the front sheet at the time.
The point of writing the procedure down is not to look rigorous. It is so that when we get something wrong, the reader can see which part of the process failed rather than having to guess at our motives.