Changing the Question: From Regeneration to Breakthrough
The revision
On 30 April 2026 we cut the scope. The project stopped being parametric study of the regeneration process and became parametric study of adsorption breakthrough. Anything in the repo that still says "regeneration", and the entire old study plan, is superseded — I moved it into an archive folder named, unambiguously, archive (DO NOT OPEN).
That folder name is a joke, but the discipline behind it is not. Killed scope that stays mixed in with live scope is how a report ends up arguing with itself in Week 17.
Why
Regeneration is the more commercially interesting half of Direct Air Capture — it is where the energy cost lives, and it is the number a policymaker asks about. It is also the half that needs temperature control I could not deliver at bench scale in the time I had, with the equipment I had.
Breakthrough, by contrast, is measurable on the rig I already built. Five runs exist. The curve shape encodes the mass transfer. And a breakthrough model that is actually validated is worth more than a regeneration model that is merely written down.
I want to be clear that this was not a downgrade dressed up as a pivot. It is a smaller claim that I can defend, replacing a larger claim I could not. Given the choice, I will take the defensible one every time — and I would rather record the reasoning here than quietly reframe it in the final report.
The gates
The part I am proudest of is that the solver is not allowed to produce a headline result until it passes three gates, all by the end of Week 6:
Gate A — linear advection–diffusion. Solve a problem with a known analytical answer. L² error under 1%. If a solver cannot reproduce the case where the maths is already settled, nothing it says about my sorbent means anything.
Gate B — Rankine–Hugoniot chord velocity. Within ±10%. This checks that the mass-transfer zone propagates at the speed the conservation law demands, not at whatever speed the discretisation prefers.
Gate C — literature benchmark. Breakthrough time within ±20% of Stampi-Bombelli (2024) at 400 ppm. Somebody else's experiment, somebody else's rig, my solver.
Gates A and B test the numerics. Gate C tests the physics. Passing A and B while failing C would have told me the code is right and the model is wrong, which is a genuinely useful thing to be able to distinguish.
The stack, such as it is
Python, NumPy, SciPy, Matplotlib. A breakthrough_fit module holding the 24-model fitter, a pipeline that turns the five real runs into per-run output folders, and an assembly step that regenerates every figure and table headless at 300 dpi.
Everything generated is regenerable from one command. That rule exists because I have already once spent an evening trying to work out which version of a figure was in which draft of a document.
What I am carrying forward
Write down the date you changed the question, and write down why. A scope revision recorded honestly is a research finding about your own process. A scope revision hidden is just a story that stops making sense when someone reads it carefully.