Two weather records disagree. Who decides the heat relief?
A relief fund and a farmers' cooperative fix the rule before summer. As the heat builds, the part of the payment both records already support is released at the next weekly check after publication; only the difference is ever disputed, and anyone can recheck every step.
96% of the relief both records agreed on would have been available to the cooperative before the season ended — replaying 2021–2025 at 10 stations in Türkiye.
- Real
- GHCN-Daily station and ERA5 records at 14 stations in Türkiye, Greece, Romania, Bulgaria, 1966–2025; engine cross-checked with xclim
- Executed
- Solidity contract that counts the heatwave itself from daily readings posted on chain, releases money during it and disputes only the providers' difference; standalone verifier
- AI tested
- LightGBM gap filling missed 3 of 9 heatwaves, so it was not adopted. See the test
- Simulated
- Fund, cooperative, checker, DEMO money, chain clock
Each case shows a run recorded from the real contract and provider services, with a bundle you can verify yourself.
Agree the climate statistics before the heat. Release money as the data show it. Let anyone check.
- Before the season1 Compare in the browserFund, cooperative and checker see what each data choice would have paid in past summers, at 14 stations, 1966–2025.
- Before the season2 Lock on chainThe rule, its 122 daily thresholds and every contract term become one fingerprint. Three parties sign it; the fund's signature escrows the deposit.
- During and after the season3 Settle on chainEach week a station provider and an ERA5 provider post the daily readings already published; the contract counts the hot spell against the signed thresholds and releases the share both support. Only the difference is ever disputed.
- Any time later4 Verify on your machineOne standard-library Python file recomputes the climate statistics from the raw data and checks every term and payment on chain.
What changes for the people the money is for: five questions, before and after
Comparison with an illustrative fund that computes one record and approves payments manually. Existing parametric platforms also automate payouts; this table is not a claim about every alternative.
| Question | Example: a manually operated fund | ExtremeProof |
|---|---|---|
| Who decides whether the heat counted? | The paying side, from one record and its own calculation | The contract: named providers post the daily readings from public records named in advance, and it compares each with the threshold signed before the season |
| Can the rule change after the heatwave? | A signature shows who signed, not that no other version exists | One fingerprint per agreement on chain; the contract refuses a different rule once locked |
| What if two records disagree? | Not visible: one source was chosen in advance | Shown summer by summer before signing; afterwards only the difference between them is disputed, the agreed part is paid at once |
| Is the money there when the rule is met? | Paid after the payer approves | Escrowed at lock; released only by the rule, the parties together or the fallback |
| Can the community check the result? | Only if the payer shares its model and data | Anyone can rerun the verifier on the published bundle |
Real-world impact
Relief during the season, as published readings qualify. Replaying 2021–2025 at our 10 Turkish stations under the default rule: in the 19 summers where both records supported a payment, 96% of that money would have been released before the season ended, in 10 of them by the end of July. Weekly checks, assuming data are published 5 days late; pinned final data, revisions not modelled.
In September 2021 Türkiye's state-backed farm insurance pool reported paying 214 million lira to 14,213 producers for drought losses (MEIR); since 2021 the pool also covers heat damage to citrus (AA). The Ministry of Agriculture names Adana, Antalya, Hatay and Mersin as the leading orange provinces (TEPGE 2024): five of our ten stations are there (Adana, Antalya, Finike, İskenderun in Hatay, Mersin).
Red Cross societies already run heatwave protocols that release money when a threshold is crossed: Greece in 2024, Kyrgyzstan in 2025 (Climate Centre, IFRC). Relief funds, insurance pools and cooperatives that pay on a heat index face the same question: whose record, which normal, what if they disagree. ExtremeProof answers it before signing and proves it after.
The rule needs only a station ID and a grid point: GHCN-Daily and ERA5 are free and global. The same code runs unchanged at 4 stations in Greece, Romania and Bulgaria next to the 10 in Türkiye (step 2, regional scan), and any of them can be signed and settled. One agreement is a handful of transactions; checking it is one standard-library Python file. Adaptation, not mitigation: it does not cut emissions, it makes heat relief arrive by a rule people can check.