The can × should map
Horizontal axis: how easily an agent builds it — our 0–10 estimate, seeded from canivibecodeit’s verdict. Vertical axis: our regret score. The interesting corner is the top right.
The four bands are the score bands, which is where our verdicts come from — so a dot’s colour and its band agree unless we overrode the maths, and the spread counts those. Build ease is a 0–10 editorial scale carried to one decimal, so a dot sits on its own value rather than in a column of everything that rounded the same way. Where two entries land on the same spot anyway, one is nudged a few pixels aside so neither hides the other.
Most conversations about vibe coding collapse these into one axis and argue about whether AI is “good enough yet”. It is the wrong argument. Difficulty and consequence are independent, and the expensive mistakes live where they diverge — where the thing is trivially easy to produce and quietly expensive to own.