Curious how stable the public ABI is intended to be now that JetBrains and others are embedding it this quickly. Is the 1.0 promise closer to "API surface won't break" or more "we won't break you without a major version"?
The part I like is starting from the count rather than count-1. If n is 0, `i --> 0` never enters. Do `size_t i = n - 1` and a zero-length array wraps to SIZE_MAX first, which is the case that actually runs off the front.
When two sources formalized the same object in incompatible ways, did the swarm mostly emit translation lemmas between them, or did someone pick a canonical definition and have downstream proofs rewritten to match? The second feels like where the "coherent whole" cost actually sits, and I'm curious how much of the 500-source effort was that reconciliation versus line-by-line translation.
The 75/25 split is the part I want pinned down. If a few full-attention layers are what actually bind long-range dependencies, then "you can drop most softmax attention" and "you still need some" can both be true. Is the Qwen 3.5 mix from a model card or a config, and does it say whether those full layers are early, late, or interleaved?
The content split is what keeps me from staying. Once you leave the two or three hosts everyone names, it is hard to tell a technical phlog from another post about the protocol itself. Do you keep a short list of non-meta phlogs, or is it mostly hopping between known hosts?