Lane B applications launch

Week 20, not week 26

Two things are true at once. The artifact is good enough to show, because the RVV table and the softmax or attention result exist. And week 20 leaves six weeks before the plan ends, which is a window to run a search in rather than a single pass at it. Start then, while the upstream patch is in review: the review thread is part of the story and waiting for the merge buys nothing.

The packet

Three links, one number each. The repository with the measured table. The post with the argument. The pull request with its status stated. Lead with the artifact rather than the CV: “linalg.matmul to RVV, <speedup> over scalar at 512 cubed, <link>”. A reader who opens one link should see the claim on the first screen.

A referral is a request to a specific person

The people who can refer you are the ones who have already read your code: the reviewer who asked for a change, the maintainer whose issue you fixed, the engineer who answered a question on the list. The message names the patch, gives one number, and asks one question. No attachment, no life story, no request to pick their brain.

Track it

One table: date, person or channel, state, next action. Ten targeted approaches with follow-up dates beat a hundred submissions with no record. When a rejection arrives, ask what would have changed the answer, and send that question while the thread is warm.

Things that break

Waiting for the merge, which spends the six weeks the search needed on a status change. A five-hundred-word referral request that asks the reader to summarise you. The same paragraph sent everywhere, which anyone comparing notes will see. And a link to the repository with no number, so the reader has to find the claim themselves.