Specification v0.2
You score the four things a photograph cannot know. Everything visible is the machine's problem.
Structural integrity of the skin at the moment of contact. Measured in actual, remembered audio.
1 – 3
Skin has surrendered. Wet paper texture. Silent bite.
4 – 7
A polite crunch that fades immediately. Acceptable in a hotel bar.
8 – 10
Audible across the table. Skin fractures like glazed ceramic and holds its shape under sauce.
How the sauce actually tasted: balance, heat, depth, conviction. A gorgeous lacquer that tastes like a squeeze bottle is not a 9.
1 – 3
Flat, sugary or aggressively salty. Heat with no flavor behind it.
4 – 7
Fine. Generic. You would not mention it afterward.
8 – 10
Layered and specific. You considered drinking the remainder.
Tenderness and quality of the chicken itself. Juiciness, texture, whether it left the bone willingly.
1 – 3
Dry, stringy, rubbery, or unsettlingly spongy.
4 – 7
Cooked correctly. Unremarkable chicken.
8 – 10
Juicy, tender, pulls clean off the bone in one piece. Tastes like actual chicken.
Would you order these again, and would you tell someone else to. Not an average — a verdict.
1 – 3
A mistake. You ate them out of obligation.
4 – 7
Fine wings. You would eat them again if they were in front of you.
8 – 10
You will drive across town for these and bring people with you.
These are read off the photo and recorded as model output. You are never asked to score them, because you can see them just as well as it can.
The model guesses all four numbers from the photo alone. Your numbers are the truth. Both are stored, side by side, forever — the gap between the guess and the truth is the entire point of this exercise.