Today I posted on the Subnet 24 120B model. This afternoon the story flipped. Independent analysis says the weights are not what was claimed. An analyst from Gradients ran the check and Const amplified it. Roughly 96 to 98% of the model reportedly matches Ant Group’s Ling mini base 2.0, with the credits stripped. The novel architecture claims, the Loop Transformer, the Engram layers, reportedly do not hold up in the actual weights. Alpha did what alpha does when trust breaks. Down 70 to 75% intraday on thin liquidity. The team disputes it. A rep has publicly bet they can prove the claims in days or hand the subnet over and leave Bittensor. Here is the real lesson. In decentralized AI the weights are public. Anyone can open the file and check the claim against it. That is the entire point of building in the open. The receipts sit in the checkpoint. I report both sides as this develops. But the market voted before the debate finished. Verify the weights before you verify the narrative. Save this one. The people who read the docs always position before the people who read the price.