Skip to content

Evaluation

A full-stack run is scored exactly as an end-to-end run is: an automated validation pass that catches gross failures cheaply, then a hand-written review by a person who plays the build. On a validator-rated case version the validators decide the score and the functional rating and the reviewer rates the run’s aesthetics with one run-wide tier; on a legacy version the reviewer assigns the score in points and the functional rating. What full-stack adds is that the reviewer also judges the assets the model produced.

This page covers the differences. Read End to End → Evaluation for the shared detail.

The automated pass is identical to end-to-end. The load check builds the produced implementation with the manifest’s required [build] commands, serves the static output, loads it in a headless browser, and records whether it runs at all. Checks (opt-in reference comparison), proofs (submitted evidence, recorded rather than graded), and instrumentation all work as they do for an end-to-end case.

The produced assets carry no validation of their own. They are build inputs, exercised by being loaded and played inside the running program and judged as part of that program by the reviewer. Action-log regeneration and cheat detection belong to an asset-generation run, where the asset is the scored output.

The review carries the same things an end-to-end review does. On a validator-rated run that is a writeup, one run-wide aesthetic rating (legendary, amazing, good, okay, or slop) covering how the whole build looks, sounds, and feels, and optionally overrides of validator verdicts. On a legacy run it is a writeup, a functional rating (flawless, great, passable, scuffed, or broken) per domain, and a checklist of pass/fail verdicts, one per [[review_item]] or one per sub-item for an item that declares them. The overall functional rating is the worst across the domains.

The reviewer is judging assets the model produced rather than assets the case provided, so the quality of the art, the sprite motion, the particle effects, and the sound is part of the experience being rated. That quality lands in the run-wide aesthetic rating, and a case can also word its review items to cover the asset behavior that must hold: an item for the art direction, for the feel of the effects, for the audio. A build whose code is solid but whose assets are placeholder rectangles or silence is unfinished work and is rated as such.

Scoring is identical to end-to-end. Each [[review_item]] carries a point weight, a pass earns it and a fail earns none, and the run’s score is the earned weight over the total declared weight. The score and the functional rating are shown together on the run, with the aesthetic rating beside them when the run has one, and the case’s leaderboard ranks models by points. Publishing a completed legacy run requires at least one review; a validator-rated run publishes on its validators’ verdicts and gains its aesthetic rating when a person plays it, code and assets alike.