Meteosure
A deterministic weather-evidence system for one insurance question: was this event recorded at this location on this date?
The weather finding is produced by deterministic rules over traceable data. AI can explain the frozen result, but it cannot change the measurements, grade the evidence, or decide whether a claim is covered.
Pulse handles ingestion and normalization; Arago executes deterministic rules and evidence grading; Mastra generates faithful explanations of frozen findings.
The 9 AM Phone Call — Meteosure Instant Proof
What Can Meteosure Do for an Insurance Agent?
For a General Insurance Agent (Agent Général d’Assurance), managing storm and weather claims is a race against client anxiety, administrative back-and-forth, and expert delays. Meteosure turns weather verification from a multi-week bottleneck into an immediate answer during the first phone call.
Answer Clients Instantly on the First Call
When an insured client calls at 9 AM reporting storm damage, generate a deterministic weather report instantly while still on the line. Provide immediate reassurance backed by official weather station data.
Eliminate Unnecessary Expert Expenses
Avoid paying €450+ domain expert fees for routine claims. Meteosure delivers a frozen, SHA-256 anchored proof certificate with an A–D evidence grade that both the client and insurer can trust.
Accelerate Settlement & Payout Timelines
Instead of waiting 2 to 3 weeks for manual weather certificates, attach the instant proof report directly to the claims file at intake, enabling fast-track approval from underwriters.
Verifying One Fact Takes Weeks
When a major climate event hits, insurers get flooded with claims. The first question for every one of them — did the event actually happen, at this address, on this date — still gets answered by hand, because the data isn’t centralized.
Certificates Are Slow and Narrow
Existing verification providers take up to 5 days (estimation) and charge as much as €65 (estimation) per certificate — and the result is technical: it doesn’t explain why a claim can or can’t be covered.
Experts Cost More Than Routine Claims Justify
For frequency claims, insurers call in a domain expert just to confirm the weather — at an average cost of €450 (estimation), often disproportionate to the claim itself.
The Wait Hurts Everyone
Calling an expert stretches the claims timeline, raises handling costs, and hurts customer satisfaction — for many files, confirming the weather alone can take 2 to 3 weeks (estimation).
Meteosure closes that gap to under 60 seconds — a deterministic algorithm that returns a result and its explanation, whether the answer is yes or no.
Three parts, one responsibility each
Pulse
Downloads, normalizes, and quality-checks raw Météo-France data (hourly readings, SYNOP records, the station directory). Every source file is versioned and hashed; results are indexed into Elasticsearch for fast lookup and PostGIS for the spatial dimension. Asynchronous — Arago only ever reads from it.
Arago
Synchronous, fully deterministic, no AI. Takes an address, a date, and a contractual threshold, and applies a fixed rule set — the same rules for every request — to reach a verdict with an evidence grade. Its output is frozen and hashed before it goes anywhere else.
Mastra
Turns the verdict into plain words a client can understand — built specifically for the hardest case: explaining a denied claim clearly. Non-decisional: it only ever sees Arago's already-frozen, already-hashed output, never raw data, never the power to change a number. An automated check confirms nothing it writes is absent from the source payload.
How Arago Decides
The claim address is geocoded, then every station within 50 km is ranked by distance.
Only the closest 12 candidates move to the next check.
Stations without usable measurements around the claim date are excluded — and documented in the report with the exclusion reason.
Fewer than 3 and the verdict is INDETERMINATE (evidence grade D) — no forced answer.
Storms are localized; averaging would deny real, localized claims.
An isolated value ×1.5 unconfirmed by neighboring stations → INDETERMINATE (evidence grade C), expert review recommended.
Result: THRESHOLD REACHED or THRESHOLD NOT REACHED.
Not a 0–100 score — a score like that blends everything and explains nothing. Each grade is a named, checkable criterion, closest relevant proof first.
Direct gust measurement, under 30 km from the risk, ≥3 stations telling the same story. Strong proof.
Direct measurement, but the nearest probative station is over 30 km out, or stations disagree. Good proof, with the reservation spelled out.
Estimated value (no direct gust on record) or an unconfirmed isolated reading. Expert review recommended.
Insufficient data. The verdict is automatically INDETERMINATE — never forced.
Design Principles
The Strongest Gust Wins, Not an Average
A storm can be highly localized. Requiring agreement across multiple stations would deny real, localized claims whose proof exists at just the nearest one.
Doubt Is Embraced
Too few usable stations, or an isolated measurement no neighbor confirms — Arago answers INDETERMINATE rather than force a verdict. That's what makes the rest of the system credible.
Nothing Is Corrected Silently
Some stations only measure average wind, not gust. Where that’s the case, gust is estimated with a 1.5 coefficient — the documented ratio for open terrain, per meteorological literature dating to the Durst curve and WMO guidance. It’s labeled "internal, v1, conservative" in every report, shown next to the raw value it came from, and it automatically caps the evidence grade at C. Direct measurement always wins when one exists.
Responsibilities and Accountabilities
When a system like this fails, who is responsible — the model provider, the organization, the designer, or the professional who approved the decision? I don’t think there’s a clean answer yet. That's exactly why the human keeps the authority, and why every verdict here is traceable back to its source.
What a Proof Unit Contains
The report provides a deterministic weather finding, an evidence grade, and traceable sources. It does not recommend or decide claim coverage. The final call stays human.
The question that costs days to answer
IARD claims handlers absorbing claim-volume spikes with no extra headcount. Insurance experts spending a disproportionate amount of time cross-referencing weather before they can actually work the file. General agents who need to justify a decision to a client still on the phone.
The Cost Is Already Real
Weather-related events cost the French insurance market 5.2B€ in 2025; storms Nils and Pedro (Feb 2026) alone are estimated at 1.2B€. Claim-volume spikes hit claims handlers hardest, with no extra headcount to absorb them.
Fraud Pressure Raises the Bar
656M€ in IARD fraud was detected in 2024. Verification now has to be defensible, not just fast — that time cost lands on insurance experts, who cross-reference weather before they can work the file at all.
Clients Expect a Faster, Justified Answer
The Cat Nat surcharge was raised to 20% in January 2025 — policyholders paying more expect an immediate, justified answer. General agents are the ones who have to give it, often with the client still on the line.
Market-context figures (2025 climate-event cost, storms Nils/Pedro, IARD fraud, Cat Nat surcharge): sourced in the project's own market study, not independently verified by me.
Frequently Asked Questions
Is Arago's verdict legally binding?
No. The report provides a deterministic weather finding, an evidence grade, and traceable sources. It does not recommend or decide claim coverage. The insurer keeps the final call.
What happens when there isn't enough station data?
Fewer than 3 usable stations and the verdict is automatically INDETERMINATE — never forced. The system does not extrapolate or guess when data is insufficient.
Can Meteosure integrate with our claims management system?
The system is designed to integrate into existing insurance claims workflows. The proof report can be consumed as a standalone document or via API. Integration specifics are discussed during the pilot phase.
How is the gust-estimation coefficient validated?
Meteosure uses traceable, timestamped weather data from recognized meteorological providers. Each data point carries its source identifier and retrieval timestamp for full auditability. The gust-estimation coefficient follows published meteorological conversion tables.
When will meteosure.com be live for the pilot?
Meteosure is currently in prototype stage, preparing for pilot deployments. Launch date and availability details are discussed directly during pilot scoping conversations.
Tech Stack
I design the safeguards before the model: evidence traceable to source, deterministic verdicts where the risk demands it, decision authority that never leaves the professional.
A technical walkthrough of Pulse, Arago, and Mastra — the architecture, the decision sequence, and where the human stays in control.