Skill runs criteria on the cantilever: five load tests
Why the cantilever uses five tests, not a score formula
We tried weighted spreadsheets on earlier desks. They produced tidy numbers and muddy writing. Skill runs are short sessions; the reader needs to know what we pressed, what broke and what we refused to invent. Five named tests keep the protocol spoken in the same voice as the reviews.
The tests are Anchor, Reach, Deflection, Load and Release. A build can fail a test and still be published with a warning. A build that fails Release — no reliable way back to the desk — does not ship.
1. Anchor
Anchor asks whether the first verb is obvious in under five seconds. Hold, drag and steer are acceptable anchors. Menus that bury the verb behind three panels fail. Soft tutorials that block input until dismissed also fail unless they can be skipped on the second run.
On this shortlist Girder Span anchors on hold, Gust Path on drag, Crack Chase on steer. Different verbs, same test.
2. Reach
Reach is the length of a clean first run to the fail state. We do not require long runs. We require a fail that reads. A twenty-second death that clearly shows why you died beats a two-minute muddle.
Reach numbers in reviews are ranges from protocol passes, not promises. Your hands will differ.
3. Deflection
Deflection is one risk/reward lever that bends scoring without a second control scheme. Wind on a plank, gold in crosswind, scoops near a crack — these count. A wholly separate mode with new buttons does not count as deflection; it is a second game.
4. Load
Load means repetition under varied hardware. Three desktop passes and one narrow viewport are the minimum. We watch for input conflicts, HUD collisions with the Close control and camera framing that hides the threat.
5. Release
Release is the exit. Every hosted build must offer a fixed Close link on menu, play, pause and over. If the engine covers it, we shift the HUD. If we cannot exit, the review does not publish.
What never enters the criteria
Store ratings, download counts, ad density, IAP ladders and “meta” progression outside a single run. Those belong to store desks. This cantilever hosts skill runs and measures the session in front of us.
How ranking uses the tests
Ranking is a judgement after the five tests, not a sum of points. When two builds pass equally, we prefer the clearer first fail. That is why Girder Span sits above Gust Path and Crack Chase on the current beam, checked 5 November 2026.