#219: What I was doing while the posts went quiet
Moving from arguing a thesis to testing it, across two research programmes.
For the last few months the posts here slowed to one or two a month. That was not drift, and it was not for lack of things to say. I had more to say than ever: a backlog of hypotheses and first-principle observations about how the Earth system moves heat and water, several of which I was fairly sure were right.
That was the problem. “Fairly sure” is not a standard I should be holding myself to in public. A first-principle argument, however clean, is an opinion until something outside my own head has had a fair chance to prove it wrong. I had been writing opinions and presenting them with the confidence of observations. So I stopped adding to the pile and went to find out which of them survived contact with data.
That is what changed the kind of work I was making. For a long time the unit of work was an essay. Take an observation, propose a mechanism, think in the realm of possibilities, argue it through, publish on Substack. The new unit is a paper with a DOI, code that anyone can run, and a result that could come back against me. That work is slower and a lot quieter, and it is most of what I have been doing.
The shift was a deliberate trade. An essay can be read by a friend and revised by the end of the week. A paper with a DOI has to survive a different audience: a stranger with a laptop, the same datasets, and a reason to want me wrong. Every analysis now ships with a code repository, every chart has a CSV behind it, every model parameter is logged and versioned. The intellectual ambition has not changed. What changed is the threshold for posting in public. I stopped trying to convince readers of an argument and started trying to assemble a body of work that would still stand if I disappeared from the conversation tomorrow.
The thread that runs through all of it is one I keep returning to as a systems engineer. The climate ledger has rows for some things and no row for others, and the gaps are not random. The standard accounting (the IPCC’s effective radiative forcing inventory and the CMIP6 models built on it) records well-mixed greenhouse gases, aerosols, and land-use albedo. A forcing that is too small in space, too short in season, or carried by a phase change rather than a gas concentration falls through the floor of that framework. I have spent the quiet months testing two such gaps, in two programmes, that point in opposite directions. One is a human forcing the ledger does not record. The other is a natural cooling it does not fully credit.
The reason to push on both at once is structural, not thematic. If the framework can miss one thing on the warming side and a different thing on the cooling side, the framework is what needs work, not the individual ledger row. A pattern of misses tells you more than a single one does. So the two programmes share methodology, not topic. Within-region controls. Difference-in-differences against an untreated counterpart subject to the same regional climate. Pre-registered decision gates that have to pass before the analysis is allowed to scale. Robustness checks designed to find me out before a reviewer does.
