Small personal notes around research, daily order, role scripts, AI, failed runs, and the habit of checking things before trusting them.
This page is more alive than the research pages on purpose. It keeps the sharable parts: the texture, the jokes, the working mood, and the self-audit trail. Sensitive details are left out.
experiment diary
A failed run is annoying for about five minutes. After that, it becomes evidence. It tells me which assumption was hiding in the code, which parameter I forgot to control, or which figure was too pretty to be trusted.
So I keep more of them now. Not because I enjoy chaos, but because deleted chaos cannot be audited.
life admin
I work better when things are clean, repeatable, and under control. That includes code folders, result tables, figure names, the kitchen counter, and the tiny row in a notebook where I write what changed today.
This is not aesthetic discipline. It is how I keep my brain from opening twenty invisible tabs at once.
tool note
AI gives me leverage. It can draft, refactor, search, compare, and move faster than I can when the task is already well-shaped.
But leverage without inspection becomes another source of error. I do not want generated work to become a black box with nicer formatting.
idea filter
Some ideas are exciting at midnight and suspicious by breakfast. I try to keep only the ones that can pass formulation, implementation, experiments, and the boring reality of time.
Excitement is allowed. It just does not get to be the only reviewer.
self-audit
A role can be performed correctly and still feel wrong. That sentence explains more of life than I expected.
These days I try not to ask only whether I worked hard enough. I also ask whether the role was ever designed for me to be present inside it.
systems metaphor
I often understand myself through systems: inherited constraints, wrong initial conditions, patches, workarounds, and partial repair.
Life does not give me a clean rollback button, so I have learned to document changes before pretending the system was always stable.
working method
I do not like dramatic rewrites unless the evidence demands one. In research, and honestly in life, I trust small corrections that can be explained.
Fix one layer. Check it. Then touch the next one. Very unglamorous, very useful.
sharing note
A finished result asks people to admire it. A working note invites people to see how it was made, where it broke, and what I learned while fixing it.
That is closer to how I actually like to share myself: not as a trophy shelf, but as a desk with traces on it.
site mood
I do not want this site to become loud just to prove there is a person behind it. A margin note, a failed run, a cooking habit, a suspicious chart - that is enough texture.
Alive does not have to mean exposed. It can mean specific.
survival method
Some systems cannot be escaped immediately: academic schedules, codebases, expectations, deadlines, inherited habits. My first move is usually not rebellion. It is inspection.
Once a system is inspectable, it becomes less mythical. It has inputs, outputs, gaps, and places where a small correction can enter.
chart suspicion
I have trust issues with smooth convergence curves. They look calm, but sometimes they are hiding mixed records, infeasible points, or a final value that does not match the table.
A beautiful figure can be useful. It just needs witnesses.
working notebook
I used to think a public site had to wait until I had enough polished achievements. Now I think that was the wrong standard for this stage.
The work is not finished, but it is not empty. It has traces, doubts, fixes, reading notes, and a person still trying to make the system hold together.