FlowSkill launched on Product Hunt on November 3. Eleven days later, here is what the launch actually measured — every number below was read from the source, not estimated.
The numbers
| Channel | Result, 11 days after launch |
|---|---|
| GitHub stars | 13 total, +4 since Oct 30 (9 then); +2 in the trailing 7 days |
| Product Hunt | 1 upvote, rank #554 on the launch-day board, 2 product followers |
| dev.to launch post | 50 views, 0 reactions, 0 comments |
| Bluesky launch thread | 3 posts, 0 likes, 0 reposts |
| Substack | 1 subscriber, 3 views in 7 days |
| X | No launch thread — account paused since Oct 22 |
| Hacker News | No Show HN — account flagged, appeal unanswered |
| No launch post — warmup paused since Oct 19 |
Four people starred the repo after launch: one on Nov 5, one on Nov 6, two on Nov 8. I will not claim the launch caused those stars. The timing is consistent with it; consistency is not causation, and four data points prove nothing either way.
What the numbers actually say
First, distribution was near zero before launch, and it stayed near zero through launch. A launch day does not create distribution. It only measures the distribution you already have. We had almost none, so the launch measured almost none. That is the honest read.
Second, the channels that were supposed to carry the launch were not available. X carried no launch thread because the account has carried a temporary label since October 22. Hacker News carried no Show HN because the account is flagged and the appeal sent October 25 has had no reply in twenty days. Reddit carried no launch post because warmup was paused October 19 after the platform auto-removed two comments from a two-day-old account. Three of the eight channels were out before launch day started. The plan assumed they would be in.
Third, the quiet parts still worked. The repo had zero stars in the trailing 7 days before October 30, then four in the eleven days after launch. The dev.to launch post reached 50 views with no promotion behind it — more than the earlier posts had shown at the same age, when all three were still under 25 views. These are small numbers. They are also the first numbers that moved in the right direction in three weeks.
What I got wrong
The launch plan assumed a launch is an event. It is not. It is a test of the channels you built before it. We built the product, the receipts, the dual gates — and almost no audience. The 48-hour launch sequence in the plan was written as if the channels were ready. They were not, and the go/no-go review on October 31 said so plainly: X, Hacker News, and Reddit were all NO-GO. We launched anyway on the channels that were GO, and the result is the table above.
The other mistake was quieter: the launch assumed attention would find the repo. Nobody was waiting. A gated harness is a tool for people who have already been burned by an agent saying "done." Those people exist, but they do not browse Product Hunt at midnight looking for us. They find tools through a thread, a comment, a post that answers a question they already have. That is slower work, and it is the work that was paused when the accounts got flagged.
What changes next
Nothing about the product changes. The gates stay. The receipts stay. What changes is the order of work: distribution first, launch second. That means one useful comment or post per week on a channel where the audience already is, measured by replies and stars, not by posts published. The next dev.to post will report the same table again in two weeks, with the same sources, so the trend is visible instead of claimed.
A reader left the sharpest comment on the earlier "21 harnesses" post: exit code 0 is not evidence — it only says the runner did not object. The receipt has to be written to disk by the runner, not narrated by the model being judged. He is right, and that distinction is now the standard this project holds itself to: a claim without a receipt written by something other than the claimant is a vibe, not proof. The same standard applies to this article. Every number above has a source listed at the top. If a number moves before publication, the article gets re-checked, not re-asserted.
Honest limitations
- Eleven days is not a trend. The +4 stars could be noise. The next measurement, two weeks out, is the one that matters.
- I cannot attribute the stars. None of the four new stargazers said where they came from. Timing suggests the launch; suggestion is not evidence.
- Product Hunt rank moved. The listing showed #505 on Nov 4 and #554 on Nov 7, still #554 when last shown Nov 13. Ranks on that board shift as products are added. The upvote count — 1 — did not move.
- This is a sample of one launch, by one person, with no audience. Nothing here generalizes to your launch. It only records what happened to this one.
The launch was quiet. The measurement was not wasted: it told us exactly how much distribution we had, which was almost none, and which channels were actually available, which was five of eight. The next launch starts from those two facts, not from the plan.
I write about this weekly in Proof over vibes, the FlowSkill newsletter — verification debt, gate design, and what the live trials find once they run. If "done" should mean something you can check: https://flowskillharnes.substack.com
Numbers in this article: GitHub 13 stars / +4 since Oct 30 (9 then), +2 trailing 7 days, stargazers chico 2026-11-05T10:30:22Z + lyonsv 2026-11-06T22:56:07Z + Vietanh099 2026-11-08T04:21:27Z + nhatvuong-tech 2026-11-08T08:00:53Z (GitHub API, 14/11); PH 1 upvote / rank #554 (last shown 13/11; #505 on 04/11) / 2 product followers / 1 comment (browser, 14/11); dev.to launch post 50 views, tổng <500, 0 reactions, 1 comment, 0 followers (dashboard, 14/11); Bluesky 1 follower / 8 posts, thread launch 0/0 (public API, 14/11); Substack 1 sub, 7d views 3, 14d views 16 (dashboard, 14/11); X label từ 22/10, 1 follower, queue 22 bài (watcher-state, 14/11 10:12); HN comment #3 dead:true/flagged. Re-checked 14/11 trước khi đăng.
Top comments (0)