The Night Market Notebook That Learned to Guess Missing Trades
Under lantern light, I ran the bracelet swap table with a notebook of who traded with whom. A gust flipped pages, and a few lines smeared into ink fog. Each person felt like a dot, each trade a string, and each bracelet style a clue tied to the dot.
The hard part showed up fast. Most people never traded with most others, so the notebook was mostly blank space. If I guessed from blanks alone, I’d be wrong a lot. Old shortcuts picked either the trade strings or the bracelet clues, and acted sure either way.
I tried something new. I gave each person a hidden profile, like a pencil sketch with a sharp center and a soft blur. I built it by passing hints across trades twice, from you to your trade partners, then through their partners, so your sketch held your bracelet and your neighborhood.
To fill a missing line, I compared two sketches. If they leaned the same way, a trade seemed more likely; if they leaned apart, less likely, and I squeezed it into a clean chance from zero to one. I also let each person loop back to themselves once, so their own bracelet didn’t get washed out.
There was a simpler option too. I could drop the blur and keep one fixed sketch per person, then rebuild the notebook from that. It can work well, but when the notebook is thin, the blurry version can admit, “I’m not sure,” instead of faking equal confidence.
When I used both the trade strings and the bracelet clues, the rebuilt notebook did a much better job putting real missing trades above random non-trades in different crowds. Without bracelet clues, it still held up against older trade-only guessing. One snag, the sketches start near “average,” but the link scoring can push them away, so scaling up may need a better starting point.