3/ More providers means more ways to reach a good answer and more room for the swarm to optimize. The thing being produced is deep research that is ch
2/ The search backends are interchangeable, so the quality lives in how they are used. No single provider wins every question, so miners tune which on
1/ Each provider does ordinary web search and page retrieval. That is the raw material a research run pulls from. The research itself is what happens
SN67 miners can now choose from five search providers and mix them per task: DeSearch, Parallel, Firecrawl, Exa, and Tavily. More ways to gather sourc
6/ Champion emission stays separate on top, so winning still pays most. Everything beneath it is now proportional and weighted toward original work. I
5/ Everything left after champion emission is then split in proportion to those weights. Nothing is carved out first, so at the same performance tier,
4/ What counts as novel is judged against a wider field. Your submission is compared against the champions and every entry from the last two completed
3/ The middle rung is new. Before, anything short of a brand-new architecture fell in with the near-copies. Now a substantial rework of an existing pi
2/ The fix is a grid with two axes. How far your solution got: top 50% counts 1x, top 10% 2x, the main stage 5x. How different it is: a near-copy 1x,
1/ The old split paid fixed amounts by rank first, then divided the remainder among novel work. That made the payoff hard to predict: you could not te