Decision-replay twin on a public multi-echelon dataset
Replay a recurring supply-chain decision, compare the model to what people actually did, and mine the gap for constraints.
From my notebook
[ COMMUNITY ]OPEN COLLABORATION
The Open Notebook & Idea Board
I run small, finished experiments in operations research and agentic systems in public. This space is where I track ideas, share problems I am evaluating, and invite challenge on real-world constraints.
[ FOR NOW ]Nothing submitted here is ever rendered automatically on the live site. The idea board below is edited by hand from my notebook. Submissions go to a private inbox where I review every operational constraint and dataset individually.
[ DISCUSSION FORUM ]
Argue in the open
Have an alternative formulation, a flaw in my failure section, or benchmark data? Open a thread on GitHub Discussions where technical critiques stay public.
[ ESSAYS & NOTES ]
Get new spins by email
Roughly twice a month: one new finished experiment, raw benchmark logs, and what broke. Delivered via Substack without spam.
What I am thinking about, building, and parking
Considering(2)
Experiments I am scoping or evaluating.
Twelve things planners actually say about their optimizer
Complaints, workarounds and wishes from public forums, in practitioners' own words.
From my notebook
Building(1)
Experiments currently on the workbench.
Battery-dispatch agent that argues with its own optimizer
Spin 05: an agent proposes, the MILP disposes, and the disagreements become constraints.
From my notebook
Shipped(0)
Finished experiments published as spins.
No community-suggested spins shipped yet.
Parked(0)
Ideas paused, blocked, or set aside.
Nothing currently parked.
Bring a constraint, dataset, or problem
If you have a messy industrial dataset, an optimizer edge case, or an operational challenge where human judgment and algorithms disagree, send it here.
What happens to what you send
- Never rendered automatically: Your submission lands in my private Netlify inbox. Nothing touches the public board or git repository until I write and verify it.
- Anonymity guarantee: If you select "Anonymous", your name, link, and email fields are cleared prior to submission.
- Public data only: I will never build experiments using closed proprietary data. If an idea requires non-public access, it will be parked or adjusted to run on public surrogates.
- Attribution: If an experiment proceeds, submitters who chose public credit will have their name and profile link permanently noted in the spin write-up.