The Numbers & The Flies

SIX REAL FLIES

Same starting brain. Six real, different outcomes.

Every fly here shares the same real MaleCNS-derived circuit and the same training recipe - the only thing that differs is what its own real 2,500-attempt search actually found.

PicklesMOST CONSISTENT

MeatballHIGHEST & LOWEST

JohnnyBEST ON MEDIUM

FrankRISES WITH HARD

ZiggyMOST AVERAGE

BeansOVERALL LAST PLACE

HALL OF FAME

The best individual
runs.

Real scores, pulled straight from the actual benchmark logs. Nothing here is invented.

The top 10 real runs across all 6 flies and all 3 difficulties (2,500 real attempts each, 45,000 total). Same 6 flies as everywhere else on this site - see Meet the Flies for who they are.

Loading the hall of fame…

EXPERIMENT HISTORY

How the fly
got better.

Same 6 flies, same shared starting checkpoint, run against three different difficulty settings - 2,500 real attempts per fly, per difficulty. All three are complete, real, recorded results.

Loading experiments…

REAL DISCOVERIES

The interesting part isn't the leaderboard.

Every number below is from an actual logged run - including the ones that make the story messier, not cleaner.

A real cliff, not a curve

Letting a wider slice of the winning fly's brain keep learning produced a real +50% jump on Hard. Opening it even wider didn't taper off gently - it fell straight back to +0%. Same trick, tried on Medium and Easy: nothing.

One outlier beat 5,999 other attempts

The single best real score in one seed's search was hit exactly once across more than 6,000 real exploratory attempts. A genuine needle in a haystack - most attempts landed nowhere close.

Hard is harder on every real axis

Smaller real gaps, a faster real current, and snappier real steering all stack in the same direction on Hard - measured in the actual physics, not assumed. No hidden advantage cancels it out.

The same fly scored the best and the worst

Meatball posted the single highest score of the entire 45,000-attempt benchmark - and, on a different difficulty, a flat real zero. Extreme highs and extreme lows, same brain.

Does the "wider brain" trick generalize?	Start	Result	Change
Hard (Pickles)	9,620	14,420	+50%
Medium (Pickles)	6,880	6,880	+0%
Easy (Meatball)	13,930	13,930	+0%

Real, honest conclusion: the trick is Hard-specific, not a general "more brain access is better" rule - it didn't move Medium or Easy at all. Full write-up, including a real incident where the original checkpoint got overwritten and had to be recovered →

The numbers, and the six flies behind them.