Problem
The myelin-accelerator "candidate next" shortlist in docs/source-repos.md was written before several PRs landed and before signal could be measured. Three of its five entries are now wrong.
Measured with the pipeline's own is_bot_user filter:
| PR |
Kept signals |
Status in old table |
Reality |
| #26 |
50 |
absent |
Packed ternary GEMV/GEMM kernels + CUDA 13.3.1 CI (+2653/−111, 36 files). Best candidate in the repo. |
| #18 |
53 |
listed "best overall" |
Holds up |
| #22 |
22 |
"watch … when merged" |
Merged |
| #7 |
25 |
listed |
Holds up |
| #6 |
13 |
listed |
Holds up |
| #4 |
0 |
listed |
No signal survives the filter |
| #2 |
3 |
listed |
3 signals across 648 mostly-target/ files |
docs/source-repos.md has been corrected in #17; this issue tracks the actual extraction.
Proposed shortlist
#26, #18, #22, #7, #6 — five PRs, satisfying the "at least 5 high-signal PRs" rule for a new source repo. Drop #4 and #2.
Tasks
Context
myelin-accelerator is the CUDA backend that grok-ozempic#25 delegates to, so this extends an already-extracted ownership boundary rather than opening an unrelated repo.
Follows #11.
Problem
The myelin-accelerator "candidate next" shortlist in
docs/source-repos.mdwas written before several PRs landed and before signal could be measured. Three of its five entries are now wrong.Measured with the pipeline's own
is_bot_userfilter:target/filesdocs/source-repos.mdhas been corrected in #17; this issue tracks the actual extraction.Proposed shortlist
#26, #18, #22, #7, #6— five PRs, satisfying the "at least 5 high-signal PRs" rule for a new source repo. Drop#4and#2.Tasks
datasets/cards/myelin-accelerator-v0.json(language Rust; domains gpu-compute, cuda, ml-infra)--strict-policydocs/source-repos.mdpromotion from "candidate next" to a full sectionContext
myelin-accelerator is the CUDA backend that grok-ozempic#25 delegates to, so this extends an already-extracted ownership boundary rather than opening an unrelated repo.
Follows #11.