6 of 8 · ~2 min read

Cost & time

The real numbers. What's been spent so far, what the next 90 days will cost, and what would justify spending more.

What's been spent

Out of my own pocket
$60
on Modal cloud GPUs over 2 months
Modal free-tier credits used
$60
$30/month × 2 months — total compute spent: $120
Software / tools
$0
all open-source: Earth2Studio, PyTorch, scientific Python stack
Data
$0
all public sources: ERA5, CHIRPS, GFS, GHCN, INMET
Personal time
~2 months
part-time, evenings and weekends

Cheap. Solo project, self-funded. The only line item that's not nominal is my time — and that was the plan.

What the next 90 days cost

ItemCostWhat it buys
Bilingual analyst (Workana contract) $500–1,500 30–60 hours of paid outreach work over 3 months
Modal compute ~$0–50 Light. Forward-poller is already running. No big re-runs scheduled.
Translation / formatting / printing the one-pager ~$50 Spanish-language stakeholder material
My time ~5 hrs/week Friday sync + letter reviews + per-cooperative views in Track 2

Total cash for the 90-day push: ~$600–1,600.

What success on Track 1 unlocks (cost-wise)

If by day 90 the relational pipeline is working — at least three of four institutional doors open — the math changes. A few things that are deferred today become worth doing:

If a stakeholder asks for...It justifies...Cost
"Is this stable across more dates?" Validation on 120 dates $5–8
"Can you handle heavy events better?" twCRPS-objective refit at heavy thresholds $0 (CPU)
"What about sub-25-km resolution?" Statistical downscaling spike $0–50
"Insurance product for cooperatives" Probability-of-exceedance API $0 (engineering)
"Kilometer-scale precipitation" Phase 5 — CorrDiff diffusion model $600–20,000

Each is a known unlock with a known cost. None of them are pursued on speculation. Each is gated on a real stakeholder conversation that justifies the spend.

What failure on Track 1 means

If by day 90 zero or one institutional door has opened, the structure was wrong. The honest read is that the project's framing — independent US researcher with a Paraguayan ex-DMH collaborator — isn't enough to get Paraguayan institutions to pay attention. In that case the next move is to either:

All three are valid outcomes. Track 1 is the experiment that tells me which one is true.

The thing that doesn't appear in the budget

Patience. Government file movement in Paraguay takes weeks. A DMH waiver might be 4–8 weeks from submission to decision. An Itaipú technical-collaboration MOU might be 6–12 weeks. The cooperatives are faster — agronomists return calls in days, not weeks — but the bigger institutions move on government time.

That's why 90 days is the engagement window. It's the minimum duration where the slowest thing has time to either move or obviously not move.

Next: how to help — specific things that would actually be useful, if you have the inclination.