The Fishing Net That Taught Knots to Share Clues
In a harbor shed, a torn fishing net lay flat on a wooden table. A few knots wore bright tape, but most were bare and hard to place. The knots weren’t random. The way each knot touched its neighbors felt like it could tell you where it belonged.
That same headache shows up in big webs of connected things, like pages that link to pages or notes that point to other notes. You might know a little description for each one, but only a few have confirmed labels. Older fixes either treated each thing alone or pushed guesses through links with rigid rules.
A newer move stays local. Each knot makes a fresh little note by blending two voices: its own note and the notes of the knots tied right next to it. Then you repeat that move a small number of times, so hints can travel a couple steps across the net without trying to “solve” the whole net at once.
Two small details keep it from getting messy. Every knot also listens to itself, like tying a tiny loop so it never forgets its own signal. And the neighbor mixing is balanced so a knot with lots of strands doesn’t drown out quieter ones. Takeaway: careful nearby mixing beats a big, sweeping guess.
The taped knots are the only ones that can say, for sure, “I belong here.” Using just those, the net can tune a few simple knobs that shape how notes get rewritten each pass. After only a couple passes, the untaped knots start landing in the right sections, and each pass just follows the strands that exist.
When people used this idea on real networks of linked documents and big knowledge maps with very few known labels, it outperformed several older approaches and ran faster under the same limits. Back at the table, the surprise was plain. You don’t need tape on every knot, and you don’t need to shake the whole net. You just need a few steady, local passes that keep each knot anchored to itself.