Journal

Desktop pivot + unified wiki UI: 88 commits across 5 repos

voice-generatedtech

Signal

Jobs-apply went desktop-only. Oil got a public sync script. Public-lab got a unified wiki UI. 88 commits across 5 repos, with no single star. This is the first day of the Apr 14-18 vault engine v1 shipping arc: public-lab unified, rv sync pipeline hardened, vault-api endpoints extended, and stella gets its scope gate. The scrubbed unifying theme across today’s five threads is simple: every project is moving from “works on my machine” to “works the way I want it to in public.”

Evidence

  • jobs-apply (5 commits): dashboard pulled off the website so the middleware gates web access to /settings only, making desktop the sole product surface. Lever discovery fix (networkidle wait) and OAuth localhost mismatch both landed. The desktop transition was the decision; the five commits are just the plumbing to make it real.
  • oil (20 commits, +16,456/-12,115): sync-public-lab.sh shipped so the oil model dashboard can be regenerated onto public-lab without a manual copy. Kalshi daily pricing wired in. Optimizer structural parameters updated. The auto-refresh launchd is picking up cleanly.
  • public-lab (41 commits, +18,811/-221): unified-vault-wiki merge landed, so the wiki no longer lives in a separate iframe. Skills pages generated from the vault skill registry. Sidecar generation for per-frame enrichment. This was the bulk of the day’s character count.
  • rusty-bloomnet (9 commits): vault-api gained versions, tags, and related endpoints, so the Leptos pages can show temporal history and cross-frame edges without recomputing on the client. Apple Photos path fix so media frames resolve correctly.
  • rusty-dakka (13 commits, +3,450): delight engine wired into app.js, 69 egg effect implementations. The agent avatars and dither v2 thread keeps warm even on days when other projects take center stage.

So What

This is a 5-front day, not a 1-front day. The work was not a single narrative: it was five parallel threads each shipping something small. The desktop pivot on jobs-apply is the load-bearing decision, because jobs-apply as a website was bleeding on every LinkedIn restriction and every OAuth callback mismatch. Moving dashboard off the website and gating web to /settings only cuts the surface area by something like 80% and lets the product focus on the one environment where it actually works (real Chrome via CDP on the user’s machine). Unified wiki UI on public-lab means the next audit can navigate between dimensions instead of siloed pages; before today, visiting the research dimension meant leaving the journal context entirely, which made cross-dimension links feel abstract. The oil public sync script is the smallest change on paper but the most structurally clarifying: it formalizes that oil, like every other project, has a private model and a public artifact, and the public-lab is the destination for the artifact. That pattern is now the template for bloomnet, stella, and the rest.

What’s Next

Tomorrow (Apr 15) lands two new repos in one day: brand-voice distillation (the archive-to-profile compiler) and x-digest (a browser extension for browsing, tagging, and exporting links). Both initial commits. The question is whether 5 parallel threads is a sustainable day or whether the context tax rises sharply past that number. Five was legible. Eight is the worry. Seven is the number I am watching for.

Log

  • Sessions: 0 (bloomnet.db lag, evidence from live git)
  • Top repos by commit volume: public-lab (41), oil (20), rusty-dakka (13)
  • Commits: 88 across 5 repos (+41,243 / -12,460)
  • Notable: desktop transition decision for jobs-apply, unified wiki UI merge
  • Cost: not tracked (post-bloomnet-ingest window)