THE FIVE-HEAD SILO — THE VOTE AND ITS CEILING

heads 4 & 5 added (seeds 10, 11), trunk frozen · ens5 K=0: 0.998 · the redundancy ceiling found: a 0.05% hard core no head gets
tier: lit · pairwise independence ratio 1.17 at K=0, but all-5-wrong = 45× the independent floor · 2026-07-10

3D — THE PENTAHEAD: FIVE A-STACKS IN AN ARC, ONE VOTE

five worldlines hunt the same cue (white marker) from five different birth phases · heads 1/2/3/4/5 converge at the vote node · all stacks idle at 1-3-4-2

2D — 3 VS 5: THE MARGINAL HEAD & THE HARD CORE

top: K-sweep — grey thin singles, dashed teal ens3, bold teal ens5, dotted maj5 · bottom: head-scaling curve at K=0 and the hard-core ledger · ladder extended to 99.8

🖍 TODDLER CORNER

We added two more voting robots. Five friends now! Score: 99.8 out of 100 (was 99).

But we found something sneaky. If you check the robots two at a time, their mistakes look nicely different — barely any overlap. Check all five at once, though, and there’s a tiny pile of puzzles — about 1 in 2000 — that every single robot gets wrong. Not because the robots are bad. Because those puzzles are genuinely smudged — the secret itself is broken.

You can invite a hundred more friends to vote; nobody can out-vote a broken puzzle. Each new friend also helps a little less than the friend before (friend 3 was worth 4 points, friends 4–5 worth 1 point together).

The big lesson: voting fixes different-mistakes. It cannot fix same-mistakes. When everyone fails together, the problem isn’t the team — it’s the puzzle. Fix the puzzle (or accept that 1-in-2000 lives there). lol