ProgressReport 13 March 2026
Personal Habits have all been good. The diet and meditation has slipped a bit, I need to get them back on track. Reading has been going well, Ive gotten into a good rhythm with Shakespeare. The other books have been a bit of a slog, which means I should probably move on. There have been a lot of recent social activities: dinner at Manfreds, Fengwei's wedding, we also had Kat and Tony over.
The big news in the past month has been the work on agents. Ive really leaned into bringing AI agents into my every day work flow. I have setup and integrated Claude Code which has been big sucess. I have been using this as the main code developer in 4b and the primary driver of my laptop. I have had it vibe code a Notes website and have started an issue-bot to to fix and push code based on feature requests in git-lab. I have integrated it into my RoamNotes… so fun!
I have also been pushing on Openclaw. This has been more frustrating to use, but it is clearly a glimpse of the future. I have set up a personal agent (Jack Nosila) on my old MacBook. He has gmail, github, x-ect accounts etc, but I have had lots of problems maintaining coherence over several days. I will definetely come back to this once I get Claude Code really dialed in.
The Airbnb is starting to show promise. We have been regually booked have have gotten some big ticket stays in the spring. Hopefully, we can start turning a profit in April.
There has been a flourish of good work on the 4b side. This was largely facilitated by Claude Code. I have finished a full pass of the mixed datasets for Run3. This included getting the FvT fits going which were needed for mixing with ttbar subtraction. I also reran the Synthetic data with the ttbar subtraction. I implemented the pz boosted logic when matching hemispheres, this is a much cleaner approach and improves the quality of the mixed data. I presented all this work in the HH4b working meeting and people seemed impressed. I have several ideas on next steps. The obvious one is to start using the mixed data a "3b" proxy from which to extrapolate the background. This should lead to smaller extrapolation uncertianties and could allow us to use the btagging scores directly in the SvB classifier. I set up this next, "mixed-to-data" or MvD, step entirely with claude in ~2 days. Its close but there are some residual problems to work out.
I have also use claude to do a lot of good code clean up in 4b. This is more or less fun, and is really made possible by all the unit tests we have. Next, I want to push the issue-bot and find gaps in our unit tests. Alejandro will push on getting a reAna like system setup on falcon.
On the HGC side we have been making high-quality modules. The next blocker is the MMTS. There has been good progress, but there is still much to do. It looks like Arvind will be staying for another year and we have made an offer to another tech, Austin. This will give us a full complement. I have leads on two other post-baccs, I think we could probably support them as well. I will be going to Alabama in early April for an EHS review of their site, should be a good chance to figure how they really fit in.