4 of 8 · ~4 min read

What I'm actually trying to do

People sometimes ask: are you trying to build a platform? raise money? publish a paper? get a job? all of it? The honest answer is yes — but not at the same time. Here's the order, and here's where I am right now.

The most important thing to take from this page: I am pursuing every one of these goals, but only one of them at a time. Right now I'm at Stage 1. The most useful thing anyone can do for me is something on the Stage 1 list — Stage 4 things aren't the bottleneck yet, and pushing on them prematurely would be a distraction.

The arc, in five stages

Each stage is a real destination with a specific completion signal. The stages are sequential because each one produces information the next stage needs. You can't credibly pilot before someone has evaluated. You can't scale before a pilot has worked. You can't pick a long-term form before scaling has revealed who the real users are.

Stage 1 · Right now
Reach — get one Paraguayan stakeholder evaluating the product

The technical kill metric (+25.7% over GFS) is passed. The relational kill metric — at least one Paraguayan institution seriously looking at this — is not. That's the gap I'm closing right now.

Done when

At least one of {cooperative, MAG, DMH, Itaipú, US Embassy contact} has had a structured 30-minute conversation about the project and given concrete feedback. The first one is the moment the project stops being a research artifact and starts being something someone is paying attention to.

What helps at this stage

What doesn't fit yet

Stage 2 · Next
Pilot — one cooperative using the forecast for a real decision

A single real user transforms the project from "research artifact" to "operational tool." Without a pilot, every other goal is theater.

Done when

At least one cooperative agronomist has used the 10-day probabilistic forecast to inform a real planting, spraying, or harvest decision for one season — and has shared whether it changed the outcome. The pilot doesn't need to be successful in the first season. It needs to produce honest feedback on what they actually did with it and whether it was worth their time.

What helps at this stage

What doesn't fit yet

Stage 3 · Later this year (if Stage 2 lands)
Validate — methodology paper, denser truth data, research partner

A pilot result has to be defensible to people outside the room it happened in. Stage B gauge data (DMH archive, Itaipú network, INMET expansion) and a peer-reviewable methodology give the work credibility that lasts beyond one cooperative.

Done when

A methodology paper is on arXiv and in submission to a relevant journal (Weather and Forecasting, Monthly Weather Review, or an AI-for-Earth-systems venue). Gauge-validation results are documented publicly. A research institution — UNA in Paraguay, a US university partner, or both — has signed on formally as collaborator.

What helps at this stage

What doesn't fit yet

Stage 4 · Next 1–2 years
Scale — operational service across multiple cooperatives, government adoption

Once the methodology and one pilot are credible, the question becomes whether this can be the actual forecast product Paraguayan agriculture relies on day-to-day.

Done when

Three or more cooperatives are operational users. MAG (Paraguay's ag ministry) references the product publicly. The partner network has expanded into Argentina or Brazil for cross-border validation. The forecast is being delivered through the interface the users actually want — most likely a WhatsApp / SMS / web channel, not a Jupyter notebook.

What helps at this stage

What doesn't fit yet

Stage 5 · Open question
Form — what does this become long-term?

Three valid directions, deliberately not committed to in advance. The right shape depends on what Stages 1–4 reveal about who the real users are and what they'll pay — in money or in institutional attention — for the product.

Three possible forms

FormWhat it isWhat it requires
Research program University-hosted, grant-funded. Regional climate-forecasting capability for the Southern Cone. Academic partner, recurring grants, an academic publication track. Slow, durable, low-equity.
Startup Paid product for cooperatives and ag-tech licensees. Hosted forecast service. Co-founder, seed capital, willing-to-pay users. Fast, commercial, high-risk.
Public utility Open-source toolkit + government and/or multilateral partnership. Forecasting infrastructure for PY (potentially cross-border). Government adoption, sustained mission-funding (foundations, multilaterals). Slowest, hardest to fund, most durable.

Why this is deliberately open

Picking the form before Stages 1–4 reveal who the real users are is premature. If the answer is "cooperatives will pay for a hosted forecast service," that's a startup. If it's "MAG wants to operationalize the methodology nationally," that's a public utility. If it's "the methodology is generalizable but no one institution will host it operationally," that's a research program. All three are legitimate. The project's job at this stage is to learn which one is true, not to pre-commit.

What helps at this stage

What doesn't fit yet

What this means for someone reading this

If you've been wondering "what is Kevin actually trying to do" — the short answer: I'm at Stage 1, and the most useful thing anyone can do for me right now is something on the Stage 1 list.

If something on a later stage's list is easy for you, please hold onto it. I'll come back when the time is right. Premature introductions are not generally helpful — investors who pass at Stage 1 because the asset isn't there yet often won't reset their judgment when it is. Same with grant officers, journal editors, and government counterparties. Timing in those relationships matters more than people typically expect.

Two things that are universally helpful regardless of stage: (1) honest critique, and (2) telling the project's name to people whose work overlaps. The rest is stage-sensitive.

Next: the people — who's involved and how the team is organized.