Launch readiness: HN + LinkedIn checklist

[korbonits]

Tracking what has to be true before posting Sheaf to Hacker News and LinkedIn — not a feature task, just a gate so we don't burn either launch on a half-ready project.

Why a gate, not a deadline

Each platform launches fresh once. Posting "Show HN: Sheaf" with a broken quickstart or no Docker image kills the project's first impression for that audience permanently — recovery is hard. Better to hold until the bar is cleared.

LinkedIn's bar is lower than HN's; sequencing should be HN first, then LinkedIn (the HN traffic + stars become the LinkedIn social proof).

Hacker News — "Show HN" readiness

Items below are blocking. The HN audience clones the repo on the spot.

• Docker image at ghcr.io/korbonits/sheaf:<version> with a one-liner docker run that serves Chronos2. Tracked under v0.8 — Adoption Track #14 (v0.8 Adoption Track).
• Quickstart verified on a clean machine in <5 minutes by someone other than the author. Tonight's pre-fix chronos dtype bug (examples/quickstart_batch.py: chronos-bolt fails inside Ray Data uv runtime env (dtype kwarg) #15) is exactly the class of failure that kills HN launches; need a third-party run-through.
• At least one benchmark vs ≥2 alternatives (raw Ray Serve, BentoML, or TorchServe). Even if Sheaf loses on some axes — published with a reproducer. Tracked under v0.8 — Adoption Track #14.
• Client SDK shipped so the demo can show pip install sheaf-client; client.predict(...) end-to-end. Currently examples interleave server config and inference code; that reads as immature on HN.
• Docs site at a stable URL (mkdocs-material → GitHub Pages, per v0.8 — Adoption Track #14). README + CLAUDE.md + blog post is below the bar.
• At least one real-world use case — author's or a collaborator's. "I'm running this in production for X" beats "look at this thing."
• Stretch: ≥50 GitHub stars before posting so the repo doesn't look empty when ~200 HN visitors click in 5 minutes.

LinkedIn readiness

Lower bar; audience skims rather than clones. Hold until HN has landed so the traffic spike becomes the social proof.

• Existing blog post (korbonits.com/blog/2026-04-14-sheaf-vllm-for-non-text-foundation-models) — already shipped.
• 30-second screencast showing pip install → request → response. Loom or asciinema.
• PyPI downloads at a non-trivial number (sheaf-serve install count visibly >0 on the README badge).
• HN traffic spike + star count in the post, e.g. "Sheaf hit #N on HN this week — here's what we learned."

Cross-references

• v0.8 — Adoption Track #14 — v0.8 Adoption Track (Docker, Helm, mkdocs, benchmarks, CLI DX) — the bulk of the HN-ready work
• Client SDK — separate v0.8 track (no issue yet; should file one)
• Kubernetes-native deployment path #11 — Kubernetes deployment path (Helm chart is part of v0.8 — Adoption Track #14; this issue covers the broader k8s story)
• Blog post — korbonits.github.io/src/content/blog/2026-04-14-sheaf-vllm-for-non-text-foundation-models.md

Definition of done

This issue closes when both checklists above are fully checked and the corresponding launches happen — at which point the comment thread becomes the post-mortem.

[korbonits]

Pre-HN organic-growth playbook (the channels + tactics for hitting the ≥50 stars line item) tracked separately as #19, so this issue stays focused on the launch gate and #19 stays focused on the ongoing tactical work.