Lab journal Reverse chronological

Working log

Lab journal

What I have actually been doing, in the order I did it. Less polished than the papers; that is the point.

  1. 29 Apr 2026

    Entry № 007

    Reread Chiang et al. and re-scoped the labelling task

    Spent the morning rereading the Chiang paper end-to-end. The labelling task is smaller than I had been treating it: 100 high-confidence pairs is enough to test the architecture; the full 200 is for calibration. Splitting the work into two batches; first batch starts this week.

    Earnings Call Analyser
  2. 28 Apr 2026

    Entry № 006

    OpenClaw briefing rewritten to surface the avoidance log

    The morning briefing was telling me what I planned to do, but not what I was avoiding. Added a section that pulls the Avoidance Log from the Tasks DB and ranks by days-avoided. Day one feedback: blunt, useful, mildly annoying. Keeping it.

    Personal AI layer OpenClaw
  3. 26 Apr 2026

    Entry № 005

    Personal website redesigned (Vol. 04)

    Old draft was a generic dark-mode template; this is the first version that feels like it actually says something. Switched to a working-journal frame, broadened the project set, added Notes / Reading / Curriculum surfaces.

    Personal Website
  4. 22 Apr 2026

    Entry № 004

    Lineup pipeline refactored into an installable package

    The pipeline finally lives as a proper Python package (src/lineup_pipeline/) with explicit entry points. Two latent bugs surfaced and were fixed in the process; both were the kind that only show up when imports run in a different order.

    Lineup
  5. 15 Apr 2026

    Entry № 003

    OpenClaw 2026.4.15 cut

    Cut as a release. Telegram bot answering correctly; LaunchAgent autostart confirmed across reboots. The gog skill (Google services) is the one that needed the most ironing; everything else was straightforward.

    Personal AI layer OpenClaw
  6. 08 Apr 2026

    Entry № 002

    NutriPlan: preference embeddings updated for the second time

    Second pass at the preference-update step. Cleaner gradient on what the system thinks I will or will not like. Moved the update from per-meal-feedback to per-day-batch; cheaper, less noisy.

    NutriPlan
  7. 30 Mar 2026

    Entry № 001

    Started SMB AI Discovery as a structured offer

    First serious attempt at packaging the discovery method I ran inside AGI for SMB engagements. Three-axis scoring rubric drafted; first conversations next week.

    SMB AI Discovery

End of log · 7 most-recent entries · Older entries available on request.