ChatGPT's audit of a proposal to sponsors for a world-class project with €200,000
User Investigate according to the superradar scheme: https://www.tron.ru/2026/09/invest-ai-project-global.html
AI
Superradar Verdict
The idea in the article is a distributed network for observing animal behavior + AI, which is supposed to find pre-seismic anomalies hours or days before an earthquake. The pain is huge, the potential economic effect is colossal, and modern cameras, microphones, and computer vision models have indeed drastically reduced the cost of testing the hypothesis. But as an investment project in the formulation of the article, I would not invest €200k right now in exchange for a stake.
The reason is not that the idea is inherently impossible. The reason is that the article makes a leap from "there are separate interesting observations" to "a solution exists," and this specific transition has not yet been scientifically proven. Moreover, some of the evidence in the article is presented incorrectly. (tron.ru)
At the same time, a much stronger startup can be assembled from the hypothesis: not "we have learned to predict earthquakes," but BioSentinel / Pre-Seismic Signal Network — an experimental additional data layer on top of existing seismology, which initially works exclusively in shadow mode and must prospectively prove that its signal is statistically better than random guessing and existing baseline models. This is exactly the kind of project I would consider.
My assessment:
| Parameter | Original version | Rebuilt version |
|---|---|---|
| Pain strength | 10/10 | 10/10 |
| Solvency | 9/10, if it works | 7/10 already at the research stage |
| Technical assemblability | 8/10 | 9/10 |
| Scientific validity | 2/10 | the task of the project |
| Verification speed | 3/10 | 6/10 with a distributed network |
| Competitive gap | 7/10 | 8/10 |
| Liability risk | 10/10 | 3/10 in shadow mode |
| Investment attractiveness now | 3–4/10 | 6.5–7/10 |
1. What is actually being proposed
The author wants to get €200k for a six-month experiment in Kazakhstan. The main hypothesis: animals feel an unknown or poorly studied pre-seismic physical signal significantly earlier than humans and instruments; modern AI systems allegedly allow automatic detection of mass anomalous animal behavior. If such observations are collected spatially, it is possible to calculate the location, time, and magnitude of a future event. (tron.ru)
The most important thought here is actually good: an animal is considered not as a "mystical oracle," but as a cheap biological sensor. And a network of animals: thousands of weak noisy sensors → spatiotemporal correlation → aggregated signal. And this is already a perfectly normal research formulation.
2. Scientific due diligence: where the article is strong and where it breaks
Haicheng, 1975
The Haicheng earthquake is indeed a famous episode in the history of forecasting. However, historical reconstruction shows that the situation was much more complex than "animals warned the city." The main precursor was a noticeable sequence of foreshocks; geodetic, hydrological, and other anomalies were also used, including reports about animals. Moreover, researchers concluded that there was no formal official short-term forecast in the usual sense — local evacuation decisions were made in a rapidly developing situation. (researchgate.net)
Therefore, Haicheng cannot be used as proof that "the animal method once worked accurately."
Toads before L'Aquila
This is a real peer-reviewed result. During observations of toads, researchers saw a sharp change in activity about five days before the L'Aquila earthquake, 74 km from the epicenter. But the authors themselves explicitly wrote that it is unknown what exact physical stimulus the animals were reacting to. (zslpublications.onlinelibrary.wiley.com)
That is, this is an interesting observation, not a validated predictor.
A stronger experiment with cows, dogs, and sheep
In 2020, a group led by Martin Wikelski from Max Planck/University of Konstanz did continuously monitor cows, dogs, and sheep with accelerometers near the Italian seismic zone. The authors detected some preceding anomalies and suggested the possibility of short-term forecasting. (doi.org)
But here is exactly what an investor must take into account. An independent group from GFZ re-analyzed this data using standard earthquake forecast verification methods and got the result: the animal signal was statistically indistinguishable from random. (doi.org)
Wikelski himself admitted a very significant thing in his response: it is impossible to make earthquake forecasts based on their dataset; only one group of animals was observed in one region and one sequence of events; there were many false anomalies from feeding, weather, and human activity. (onlinelibrary.wiley.com)
This does not kill the hypothesis. But it moves it from the stage: "a solution exists" to the stage: "there is a hypothesis that can now finally be properly tested."
3. The most interesting new information of 2026
Here the article unexpectedly hits a truly relevant technological shift. In April 2026, a paper was published in Scientific Reports specifically on AI analysis of animal vocalizations as possible earthquake precursors. The authors report very high classification metrics — up to 98.87% test accuracy for Bi-LSTM. (nature.com)
On the surface, this looks like almost direct confirmation of the idea. But upon careful reading, the main caveat is revealed: the original audio recordings of animals were collected from YouTube and Google, there were few of them, after which the set was artificially expanded using pitch shifting, time stretching, noise, volume changes, and other methods up to hundreds of thousands of samples. (nature.com)
That is, the model answers quite well to a question roughly like: "does this recording look like a set of audio recordings that researchers marked as abnormal/pre-earthquake?" But this is completely different from: "will the system predict future earthquakes in the real world with continuous observation of animals, taking into account millions of cases of stress, illness, thunderstorms, predators, feeding, and human intervention?"
It is characteristic that the authors themselves call the next step the creation of large systematically labeled datasets from real IoT sensors in seismically active areas. (nature.com)
And this is where the real startup is.
4. A very serious red flag in the article: the patent
The article states that in Israel in 2008, a patent WO/2008/053463 was issued for "this same method." (tron.ru)
The PCT publication does exist: WO2008053463, author Alexandr Yagodin. But the patent does not contain animal behavior analysis technology. It proposes a completely different physical hypothesis: sensors should register specific "peaks" of a supposed wave moving towards the future epicenter at about 100 km/h; by the arrival time of these peaks at several sensors, it is supposed to calculate the epicenter, time, and magnitude. (patents.google.com)
Google Patents also shows the family as ceased. (patents.google.com)
Consequently:
- the patent ≠ confirmation of the animal method;
- the patent ≠ scientific validation of viability;
- and the existing publication does not look like an active global IP moat for the project.
For due diligence, this is a significant problem.
5. Another red flag: "scientists ignore it because they will lose their jobs"
This part of the article is actually refuted by the scientific landscape itself. USGS has been researching reports of animal behavior for decades. Max Planck conducted instrumental experiments. GFZ published systematic reviews. In 2026, Scientific Reports published an ML paper specifically about animal vocalizations. (pubs.usgs.gov)
The problem is not a lack of interest. The problem is different: there is no reproducible statistical link with an acceptable false alarm rate. And this is absolutely critical for the market.
Imagine a system that successfully "predicts" 80% of strong earthquakes, but every two weeks requires evacuating Almaty without a subsequent earthquake. The economic utility of such a product will be negative.
Therefore, the main metric here is not the model's accuracy at all. The main metrics are: False alarms / unit time + missed earthquakes + share of time in alarm state + lead time.
6. But the pain is completely real
Existing Earthquake Early Warning (EEW) systems do not predict earthquakes. They fix the already started destruction and use the difference in wave propagation speeds and communication to give other areas a few seconds or tens of seconds. USGS explicitly emphasizes this difference. (usgs.gov)
Hence the real Competitive Gap. On Reddit, there are regularly two opposite complaints: users either do not receive a warning at all / receive it after the shaking starts, or they receive false alarms. At the same time, even four seconds of real warning are perceived by users as extremely useful. (reddit.com)
If someone actually gives not 4–40 seconds, but 4–20 hours, then this will not be a 20% improvement of the existing market. This is a change in the product class.
7. Solvent demand is already proven — but for an adjacent outcome
Buyers are already paying not for "seismology," but for a specific economic result. USGS licenses ShakeAlert data to commercial partners who automatically stop trains, close valves, switch equipment to safe mode, and warn people. (shakealert.org)
Early Warning Labs claims its commercial systems protect over 10 million people and serve hospitals, transport, schools, municipal facilities, and enterprises. (earlywarninglabs.com)
Safehub went even further and linked its own sensor network with parametric insurance. The University of California installed 180 sensors as part of a multi-million dollar insurance coverage. (libertymutualre.com)
And in August 2026, Liberty Mutual announced the first real parametric payout triggered by Safehub network data after an earthquake in Peru. (libertymutualre.com)
That is, a solvent market definitely exists. Only the client buys: "reduce my financial/operational damage," not: "give me a beautiful new theory of earthquakes."
8. Competitive landscape
| Player/Class | What it can do | Lead time | Weakness relative to the idea |
|---|---|---|---|
| USGS ShakeAlert | Fixes an already started earthquake | seconds | no warning before rupture |
| Early Warning Labs | Commercial alerts + automation | seconds | same |
| SkyAlert | Private EEW solutions in Mexico | seconds/minutes | event has already started |
| SeismicAI | Localized EEW for enterprises | seconds | same |
| Safehub | shaking + damage analytics + insurance | after/during | does not predict the event |
| State seismic networks | monitoring/risk assessment | years → seconds | no proven short-term prediction |
| Animal-precursor science | experimental research | potentially hours | no proven forecasting skill |
The most important thing: there is no direct strong commercial competitor for an animal pre-seismic predictor precisely because the market has not yet proven the possibility of the product. This is a classic zone: Competition Gap is high, but Scientific Risk is even higher.
9. Kazakhstan is actually a very good testing ground, but not for the reason in the article
Almaty is indeed interesting. The World Bank attributes almost the entire urban agglomeration to a zone of high seismic exposure. (thedocs.worldbank.org)
And the main thing: right now there is a major infrastructure update there. According to official information, an additional early warning network is being built in Almaty; out of 70 automated stations around Almaty, 65 had been built at the time of publication, plus city stations are being created and a three-year project for digital study of tectonic faults. (gov.kz)
This gives an excellent transition trigger: the state is already modernizing the monitoring stack → you can offer an experimental biological-sensing sidecar. That is, not to sell: "Your seismologists are wrong, replace them with animals." But: "Let's add an independent experimental sensor layer. It does not affect Mass Alert in any way. In a year, we will show whether it brings additional predictive information or not." This is a fundamentally different GTM.
10. Low-Cost Entry Vectors
In our extended superradar methodology, it is important to look not only for a product wedge, but also for workflow integration, transition moment, distribution, and business model arbitrage. Here I see five vectors:
| Vector | Construction | Score |
|---|---|---|
| Process integration | Biosentinel score as an additional channel next to the state seismic network | 9/10 |
| Shadow mode | no public alarms; the system predicts "into the void," then compares with facts | 10/10 |
| Distribution arbitrage | use existing farm cameras, livestock AI, pet cameras instead of your own hardware network | 8/10 |
| Transition trigger | connect during the modernization of seismic networks in Almaty/other cities | 8/10 |
| Business model | paid research/data pilot → API/data licensing after proof | 8/10 |
And the option: "an app warns the population about an earthquake based on dog behavior" I currently rate 1/10. It immediately creates unsolvable problems of liability, trust, and false alarms.
11. The strongest wedge
I would formulate it like this: BioSentinel is an independent experimental layer of early pre-seismic signals that analyzes the collective behavior of thousands of animals and in shadow mode checks whether additional information appears hours before an event that is not in regular seismic data.
- Do not promise prediction.
- Do not promise evacuation.
- Do not issue consumer alerts.
The first-stage product is the Evidence Engine.
Evidence Engine Pipeline:
cameras / microphones / wearable sensors
↓
normal model of each animal/group
↓
anomalies
↓
spatial correlation between independent points
↓
control of weather / time of day / feeding / noise
↓
BioSentinel score
↓
SHADOW MODE
↓
automatic verification with seismic ground truth
This is already much more interesting.
12. Where the real technical moat is
AI for recognizing animal behavior is already a commodity. For example, CattleEye in 2026 reports monitoring over 200,000 animals in 23 countries; M&S is deploying a computer vision system across its entire dairy network. (cattleeye.com)
Furbo/Petcube can already recognize barking, meowing, and unusual activity of pets. (petcube.com)
Therefore, the moat is not: "we use AI for animals." The moat is: a multi-year synchronized set of negative + positive observations. That is, you need not only recordings of: "the dog barked strangely before the earthquake." You need billions of hours of: "the dog barked strangely, and the earthquake DID NOT happen." It is precisely the lack of negative examples that has destroyed research in this topic for decades.
13. Why the proposed six-month experiment is dangerous
One region + six months creates a huge risk of getting a beautiful but meaningless correlation. If enough independent events of suitable magnitude do not occur nearby during this period, the experiment will prove nothing. If one series occurs from the main shock and dozens of aftershocks, then ML will easily learn the cluster, but this is also not proof of universal predictive ability. Therefore, the experiment must be multi-geographical and pre-registered. I would demand as an investment condition:
| Requirement | Why |
|---|---|
| 3–5 independent seismic regions | exclude local overfit |
| frozen model before events occur | exclude post hoc fitting |
| fully automatic timestamps | exclude retrospective selection |
| control periods without earthquakes | count false alarms |
| weather, feeding, humans, predators, and noise as confounders | separate normal stress |
| Molchan / precision-recall / time-in-alarm | normal forecasting skill check |
| comparison with seismic baseline | prove exactly incremental value |
| independent statistician/seismologist | buyer trust |
| publication of the protocol before obtaining the result | remove suspicion of cherry-picking |
14. €200k: is it enough?
To prove "we can predict earthquakes" — no. To create a good first prospective evidence experiment — yes, possibly. But I would distribute the money completely differently: not build an expensive proprietary hardware network, but connect to existing flows as much as possible. For example: farms with cameras + a few accelerometers + microphones + environmental sensors + existing seismic APIs. The most expensive part of the project is not the LLM and not computer vision. The most expensive part is a reliable prospective dataset.
15. TAM / SAM / SOM
Here the traditional giant TAM is almost useless. If the technology really reliably predicts M6+ in 12 hours, the TAM is practically the entire global seismic-risk economy: governments, infrastructure, industry, logistics, insurance, real estate. These are tens of billions in potentially prevented losses. But the current commercial TAM of such a system is almost zero, because the buyer has nothing validated to buy. Therefore, for the superradar, I would count it like this.
Stage 1 — Evidence/Data Platform
Assumption: 10–20 government centers, insurers, universities, and infrastructure operators × $25–100k research/pilot contract. SAM ≈ $0.25–2M/year. Competition specifically for research/data budgets is moderate, условно 4/10. SAMconc ≈ $0.15–1.2M. SOM of the first year: $100–300k, if it is possible to get 2–4 institutional pilots.
Stage 2 — additional precursor signal proven
50–100 seismic agencies, insurers, EEW vendors, and critical infrastructure owners × approximately $100–500k data/API contracts. Estimated: SAM ≈ $5–50M ARR. Strong substitutes give Competition ≈ 7/10. According to the formula we use: SAMconc ≈ $1.5–15M ARR. These are my bottom-up scenarios, not published market statistics. But there is one asymmetry: if the system in a well-designed prospective experiment really shows a signal in 3–20 hours with a low false-positive rate, these estimates will become irrelevant in just a few months. A new market will begin.
16. Adversarial re-check
The most dangerous version of explaining the result: animals do not feel the process of a future earthquake; they just feel small foreshocks better than humans. A systematic review of over 700 reports of anomalous behavior found that the distribution of such reports is suspiciously similar to the distribution of foreshocks; the vast majority of the works did not have normal continuous time series at all. (eurekamag.com)
If this is so, a commercial effect may still exist — an animal can be a cheap ultra-sensitive seismic sensor. But then a fatal question arises: why use a cow if you can install a MEMS/seismometer? That is why the experiment must show exactly: the animal contains predictive information absent in simultaneously registered physical sensors. If this is not there — the moat disappears.
17. What I would propose to the investor instead of the deal from the article
Not "€200k for a stake to launch a global forecasting system." I would structure this as milestone financing.
- Stage 0 — €20–30k, 6–8 weeks. Independent scientific audit, pre-registered protocol, access to existing data, reproduction of published animal studies, assessment of the false-positive problem.
- Stage 1 — €50–70k. Shadow pilot at 10–20 existing farms/observation points + integration with official seismology.
- Stage 2 — the remaining capital only after achieving a pre-set statistical metric. For example: prospective result is significantly better than random baseline and better than seismic-only baseline, with a pre-limited time-in-alarm.
And only after that discuss operational prediction.
Superradar Conclusion
I would reject the original investment thesis, but I would not reject the technological hypothesis itself. The most valuable thing here is not the statement: "animals can predict earthquakes." The most valuable thing is the coincidence of four factors right now:
- an old, still unresolved scientific hypothesis;
- very cheap continuous cameras/microphones/edge AI have appeared;
- animal-behavior AI already exists in production-grade;
- government and insurance clients are already paying for seismic data and risk reduction.
Therefore, I would give the idea a separate Superradar status: HIGH-IMPACT / HIGH-SCIENTIFIC-RISK. And the most interesting startup here is not an "AI earthquake predictor," but: a global network of biological sensors that collects prospective evidence and tries to discover a new class of pre-seismic signal. First science/data infrastructure, then risk API, and only after proof — prediction. In case of a negative result, you get a scientific dataset and animal-anomaly infrastructure. In case of a positive result — a completely new class of catastrophic-risk technology potentially emerges. It is this asymmetry that makes the hypothesis worthy not of a standard venture investment, but of a strictly milestone-controlled moonshot verification.
User
Conduct a separate analysis only according to the commercialization radar.
AI
If we look at TRON strictly through the "Technology Commercialization Radar," and not through the startup radar and not through scientific due diligence, the picture turns out noticeably colder.
Key conclusion: The market around seismic data, automated response, IoT sensors, and parametric insurance is already commercializing. AI observation of animals is also already a production technology. But the link "animal behavior → reliable earthquake forecast for hours/days" is still pre-commercial. That is, the intersection of two already commercialized technological waves is interesting here, but the main missing link has not yet been proven.
1. Technology decomposition
I would divide the idea of the article into four layers:
| Technological layer | Stage in 2026 | Commercialization |
|---|---|---|
| AI analysis of animal behavior | Production | already underway |
| Distributed IoT/seismic networks | Production / Scale | already underway |
| Seismic data → automated actions/insurance | Production / Scale | growing rapidly |
| Animal behavior → pre-seismic signal for hours/days | Research / pre-pilot | has not yet begun |
It is the last layer that is actually TRON. This is an important distinction: component readiness ≠ final technology readiness.
2. Signal No. 1: capital movement — 7/10
Money is already moving in the adjacent market. Safehub raised $9M Series A back in 2021 from investors including JLL Spark and insurer HDI/Hannover; now the company shows partners like Liberty Mutual Re, Munich Re, Hannover Re, AXA XL, Chubb, Aon. (safehub.io)
This is a very good signal for the radar: institutional capital already believes not just in seismology, but in the commercial value of additional high-frequency earthquake data. Separately, another component of TRON — computer vision for animals — has also moved from startups to industrial corporations. CattleEye belongs to GEA, which opened a separate R&D team in 2026; the system monitors over 200k cows on 140+ farms in 23 countries. (cattleeye.com)
Interpretation: Capital finances both "halves": AI animal sensing and seismic sensing/risk data. But there is practically no capital directed at connecting them yet. Commercialization Gap: high.
3. Signal No. 2: pilot → production — 8/10 in the adjacent market
This is perhaps the strongest signal. In June 2026, the Mexican government included the Safehub network in the national parametric insurance program for earthquakes and volcanic events. This is no longer a laboratory or proof-of-concept, but a government program. (safehub.io)
And on August 13, 2026, Liberty Mutual announced the first real insurance payout triggered by Shake Network data after an earthquake in Peru. (libertymutualre.com)
This means the transition: sensor → data → financial decision → real money. And this is a very important precedent for TRON. The buyer is already ready to trust an external sensor network enough for a financial action to occur based on its data.
4. Signal No. 3: infrastructure is ready — 9/10
And here 2026 is really very different from 2011.
Animals: CattleEye already shows that an existing camera can continuously extract behavioral features without a wearable device. Over 200k animals are already under such AI observation. (cattleeye.com)
Seismics: USGS has already built a full-fledged commercial ecosystem on top of ShakeAlert: there is a formal path Pilot License → Performance Review → License to Operate, after which a partner can sell commercial products on top of government seismic data. (shakealert.org)
Kazakhstan: For the author's proposed region, this is especially interesting: since 2025, Almaty is building another 70 automated early detection stations; 65 of them are already built, the rest are planned by the end of 2026. (gov.kz) There is also an agreement between Kazakhstan and China on the integration of seismic monitoring and early warning systems. (gov.kz)
That is, it is not required to first create a "digital environment." It is appearing right now.
5. Signal No. 4: procurement — 8/10
The technology has a much more important sign of commercialization than just articles or investments: there are real budget lines of buyers. Money is already allocated for: seismic stations; early warning; disaster resilience; business continuity; catastrophe-risk modelling; parametric insurance; automated emergency management.
For example, Mexico already uses Safehub in sovereign disaster financing, and in 2025 the total limit of its multi-peril parametric cover was about $270M. (theinsurer.com)
This does not mean that $270M is available to a data provider. But it shows the size of the financial process into which the technology can be integrated.
6. Signal No. 5: real revenue events — 8/10
In the adjacent market, the most important event for the commercialization radar has already occurred: sensor data directly caused a monetary transaction. The Safehub/Liberty Mutual payout in Peru is a particularly strong signal. (libertymutualre.com)
In addition, in July 2026, Munich Re launched a parametric earthquake cover for corporate clients in Japan, using data from seismic networks. (munichre.com)
A commercial stack is formed: sensor data → risk model → threshold → decision → money.
TRON is potentially able to add one more layer before it: precursor probability → risk escalation → preventive action. This is much clearer to the buyer than the abstract "we will predict an earthquake."
7. Signal No. 6: vendor marketing — 8/10
Market marketing is already changing noticeably. They are not selling "seismology." They are selling: business continuity, automated response, financial resilience, localized risk data, parametric payouts.
Early Warning Labs sells automatic gate opening, equipment shutdown, elevator management, personnel alerting, and other actions. USGS explicitly allows commercial partners to have such machine-to-machine scenarios. (shakealert.org) Safehub sells sensor data through insurers. (safehub.io)
This is a favorable marketing environment for a new technology. The buyer no longer needs to be explained: why do I need another source of earthquake intelligence? They need to be proven: why does your additional source improve my result? This is a much easier sales task.
8. But core TRON is only at the pre-commercial stage
Now the main limitation. The original article wants €200k for a six-month test in Kazakhstan and declares the ultimate goal to be an accurate forecast of magnitude, location, and time. (tron.ru)
From the point of view of our radar, this corresponds approximately to: TRL / Commercial stage: Research → validation, and not: Pilot → production. That is, the market around the technology is at 7–9/10 readiness, but the core technology itself is at about 2–3/10. Therefore, this is not a technology commercialization bet in the spirit of: "the market will appear in 2027." This is: "in 2027, the first commercially significant proof may appear, after which the market will begin to form." The difference is huge.
9. Where the transition gap arises
Here the idea becomes especially interesting for the radar. The existing market knows how to do:
Existing Market Stack:
EARTHQUAKE HAS STARTED
↓
sensors detected
↓
calculated expected shaking
↓
10–60 seconds
↓
automatic action
The commercial stack is already built. But a layer is potentially missing:
Potential Transition Gap Stack:
NOTHING HAS STARTED YET
↓
anomalous group of independent signals
↓
probability of event increased
↓
hours
↓
soft preventive action
This is the potential Transition Gap. Moreover, it is not necessary to immediately evacuate the city. You can start with reversible actions: increase the readiness of emergency services; transfer personnel to increased readiness; check backup systems; change the scheduling of dangerous operations; temporarily increase monitoring; notify risk managers; prepare emergency liquidity; change the regime of critical infrastructure. Here the price of a false alarm is much lower.
10. Pilot → Production → Scale
The most useful part of the commercialization radar looks like this here.
PILOT: What hurts: "Can we even show a statistically repeatable signal?" Cameras/microphones, animal data, synchronization with seismic events are needed. Pilot price: условно €50–250k. Buyer: university / seismological center / government program / insurer / innovation budget.
PRODUCTION: The pain changes completely: "I see an anomaly score. Can I use it when making a decision worth millions of dollars?" Issues arise: false alarms; SLA; continuity of observation; independent validation; model drift; liability; integration with existing EEW; audit trail. And here a real commercial product arises: Pre-Seismic Risk API. Not: "there will be an earthquake tomorrow." But: "based on an independent additional sensor layer, the probability of the event has moved from baseline to elevated state."
SCALE: The problem: "Does the model work the same way in Almaty, Japan, California, Mexico, and Italy?" Then the moat becomes: a global longitudinal dataset + regional calibration + evidence history. And it is here that the product turns into a platform.
11. Who really pays
I would completely remove the consumer from the first stage.
- Buyer No. 1 — reinsurers / parametric insurers. The most interesting. Because they are already buying new unique data streams. The Liberty Mutual + Safehub precedent shows this. (libertymutualre.com) Budget: catastrophe analytics / modelling / product development. Attractiveness score: 9/10.
- No. 2 — government seismic centers. Especially where modernization is happening right now. Almaty is a good example. (gov.kz) Budget: civil defense / seismic monitoring / R&D. 8/10.
- No. 3 — critical infrastructure. Oil and gas, transport, ports, manufacturing, nuclear, data centers. Their existing solutions already automatically switch equipment to a safe state. (shakealert.org) If instead of 30 seconds you get even 30 minutes with acceptable reliability, the value can be huge. 9/10 after validation, 3/10 now.
- No. 4 — EEW vendors. Perhaps the most underestimated channel. Do not compete with Early Warning Labs / SeismicAI / Safehub. Sell them: one more input signal. This is a much more realistic commercialization.
12. Cheap entry vectors
The strongest option here is not to create your own alternative earthquake system.
- Additional sensor layer — 10/10. existing earthquake platform + TRON precursor API. No replacement of the incumbent.
- Shadow mode — 10/10. The system generates predictions for a year, but no one does anything based on them. After the event: forecast → actual outcome → score. This is almost the ideal way to commercialize high-risk technology.
- Using someone else's observation infrastructure — 9/10. Do not install a million cameras. Connect: farms; dairy AI; veterinary networks; zoos; pet cameras; existing CCTV. CattleEye proves that infrastructure of this type already exists in production. (cattleeye.com)
- Entry through insurance — 8/10. Do not sell a warning. Sell an additional risk signal. This lowers the regulatory/liability barrier.
- Entry through seismic network modernization — 8/10. Almaty is currently in a transition moment: new infrastructure is being installed and integrated. (gov.kz)
13. Competition
A paradoxical market. Direct competitors for: "global AI network of animals as a precursor sensor" are almost absent. Competition: 2/10. But substitutes are very strong: ShakeAlert; Early Warning Labs; SeismicAI; Safehub; government seismic networks; satellite/geophysical sensors; conventional probabilistic seismic hazard models. Therefore, on the real budget line, competition is: 7–8/10. The buyer will say not: "who else is looking at cows?" but: "why should I add your signal to the five existing sources?"
14. SAM → SAMconc
Here you cannot honestly count the huge TAM as "all losses from earthquakes." For the first 1–3 years, I count only the really accessible market.
Stage I: Validation/Data pilots. Assumption: 20–40 potential early adopters worldwide: seismic centers; insurers/reinsurers; EEW vendors; critical infrastructure innovation teams. Average contract: €50–150k. We get: SAM ≈ €1–6M/year. Competition for these budgets: условно 5/10. According to our formula: SAMconc ≈ €0.5–3M. Realistic SOM of the first 12–18 months: €150–600k. This already allows a small team to exist.
15. If the precursor signal is confirmed
Then the market changes radically. Assumption: 100 large institutional buyers × €100–500k/year → SAM ≈ €10–50M ARR. With Competition = 7/10: SAMconc ≈ €3–15M ARR. And then completely different markets may appear: critical infrastructure; aviation/rail; energy; insurance; sovereign risk; industrial automation. And then the TAM becomes much larger. But this is a conditional second market, existing only after the signal is proven.
16. Commercialization scorecard
| Radar signal | Score |
|---|---|
| Capital movement | 7/10 |
| Real production deployments of the adjacent market | 9/10 |
| Real procurement | 8/10 |
| Revenue / money events | 8/10 |
| Infrastructure readiness | 9/10 |
| Vendor marketing activity | 8/10 |
| Clear budget owner | 8/10 |
| Regulatory/adoption readiness | 5/10 |
| Component readiness | 9/10 |
| Core TRON technology readiness | 2–3/10 |
| Ability to MVP with a small team | 8/10 |
| Probability of full commercialization in 1–3 years | 5/10 |
17. Conclusion on the commercialization radar
I would rate: TRON / AI biological earthquake precursor sensing. Commercialization Radar: WATCH / EARLY SIGNAL. Not BUY NOW. Moreover, the reason is quite specific. This is not a situation: good technology, but the market is not ready yet. Here it is exactly the opposite: the market is ready, the infrastructure is ready, budgets exist, adjacent products are already in production — but the core technology has not yet been proven.
That is why I would put TRON in the category: "Potential Discontinuity 2027–2029" with a very specific trigger for moving to the top of the radar: the appearance of a prospective result where the additional animal/biological signal statistically improves the forecast relative to the seismic-only baseline with an acceptable level of false alarms.
Before this event, primarily the technology verification infrastructure is commercialized. After such an event, commercialization can go very quickly, because a significant part of the downstream stack will not have to be built: sensor networks, APIs, alerting, building automation, catastrophe modelling, parametric insurance, and institutional buyers already exist. Therefore, in the radar, I would put 5/10 now, but with an unusually high optionality of 9/10: one well-designed positive prospective trial can transfer the technology from the research category to a new commercial vertical in 12–24 months.
User
conduct a deep and extensive search, Which structures: funds, family offices, private investors and philanthropists can fund such R&D?
AI
Yes. I would consider this not as a search for an "investor in an earthquake prediction startup," but as a separate market for high-risk R&D funding. As of September 6, 2026, I have found several quite real routes, and some of them are much better suited than classic venture capital. Main conclusion: the first €200–500k here is more rational to raise predominantly through grants, scientific philanthropy, and corporate sponsored research, and to involve equity investors after the first prospective result. If you go straight to VC, the project will be evaluated as an extremely early startup with a binary scientific risk and they will either refuse, or the capital will turn out to be too expensive.
1. Who I would attack first
| Structure | Type of money | Realistic scale | Fit now | Why it fits |
|---|---|---|---|---|
| Emergent Ventures | private philanthropy / grant | from a small seed grant, no fixed standard check | 10/10 | specifically looking for unusual zero-to-one ideas, accept globally |
| Global Innovation Fund | grant / risk capital | up to $230k Pilot, then up to $2.3m Test & Transition | 9/10 | disaster preparedness + innovations for developing markets |
| Fifty Years Manifest Grants | scientific grant | $25–100k | 9/10 | fast money for translational high-risk research |
| Lloyd's Register Foundation | charitable research funding | now small grants up to £10k, large programs reached up to £1m | 9/10 | infrastructure safety, disaster resilience, early warning |
| Kazakhstan: Science Committee / Science Fund | state R&D | depends on the competition | 9/10 | local scientific base + applied commercialization |
| AXA Research / AXA Foundation | corporate scientific philanthropy | Chair up to €1.5m / 5 years | 9/10 | natural risks, social resilience, insurance |
| Horizon Europe Cluster 3 | European consortium grant | usually hundreds of thousands – several million € per consortium | 8.5/10 | earthquake is directly included in disaster-resilient society |
| W. M. Keck Foundation | scientific philanthropy | usually $1–1.3m / 3 years | 9/10 science / 3/10 access | loves risky ideas that conventional peer review considers too bold |
| NOMIS Foundation | private scientific philanthropy | large multi-year projects | 9/10 science / 2/10 access | high-risk fundamental interdisciplinary science |
| Schmidt Sciences | Eric & Wendy Schmidt philanthropy | multi-million programs | 8/10 | AI + Earth systems + new scientific instruments |
| Patrick J. McGovern Foundation | AI philanthropy | often around $300–750k | 8/10 | AI for public purpose, crisis/resilience |
| Astera Institute | private scientific philanthropy of Jed McCaleb | individually | 7/10 | risky research between government and market |
| Convergent Research | FRO / scientific infrastructure | $20–50m for a mature FRO | 4/10 now → 10/10 later | if you turn the project into global scientific infrastructure |
| SOSV / HAX | equity | initial funding up to about $550k | 7/10 after prototype | sensors + AI + hard tech |
| Root Ventures | equity | usually $3–5m initial | 4/10 now → 8/10 after evidence | deep tech / automation / ML |
| DCVC | equity | venture rounds | 4/10 now → 9/10 after evidence | deep tech, geospatial/risk intelligence |
2. The easiest first chance: Emergent Ventures
This is one of the few sources where I would apply almost immediately. Emergent Ventures at the Mercatus Center specifically exists to fund unusual, experimental, and potentially scalable ideas that have difficulty getting a regular institutional grant. The program is global, the application is relatively low-bureaucracy. (mercatus.org)
It is exactly here that you should not try to present TRON as a mature commercial product. The correct formulation: Prospective experiment: does the collective behavior of a distributed animal population contain additional pre-seismic information absent in standard geophysical sensors?
I would ask for the first money not for a "warning system," but for: pre-registration → dataset → retrospective benchmark → 2–3 test sites → shadow forecasting. Fit: 10/10.
3. Global Innovation Fund — perhaps the best source for €200k+
GIF is especially interesting because it is a much larger and more structured tool. The organization finances innovations from almost any country and uses a stepped model. For early pilots, capital of about up to $230k is possible, and at the Test & Transition stage — up to $2.3M. (globalinnovation.fund)
Moreover, the fund's theme directly includes disaster preparedness, prevention and response. But there is a nuance here. GIF is focused on improving the lives of people in low- and middle-income countries. Therefore, the proposal: "we will do research near Almaty" will be weaker than: "we are creating a cheap additional layer of disaster early warning, which can be scaled to Central Asia, Nepal, Pakistan, Indonesia, and other regions poorly provided with instrumental networks."
Ideal construction: Kazakhstan is the first site. Then: Kyrgyzstan / Tajikistan / Nepal / Pakistan as proof that the approach can reduce the cost of early monitoring in countries that cannot afford Japanese density of geophysical infrastructure. Fit: 9/10.
4. Fifty Years — a very interesting "science → startup" bridge
Fifty Years has an unusual Manifest Grants program aimed specifically at research that can quickly move from the lab to practical application. Grant sizes are around $25–100k, the application is very short; the program promised a fast decision cycle and does not take IP. (fiftyyears.com)
For us, this is almost the ideal size for: the first statistically rigorous experiment. For example: $80k: 20 farms × existing cameras + audio + a few accelerometers + weather/environment controls + seismic ground truth + independent statistician. Important: it is desirable to have a real scientific PI here — a specialist in animal behavior / seismology / statistical forecasting. Fit: 9/10.
5. Lloyd's Register Foundation — unexpectedly good candidate
This is one of the most interesting funds I found. The mission of Lloyd's Register Foundation is Engineering a safer world. They fund research on safety, infrastructure resilience, disaster resilience, and projects based on the World Risk Poll. In recent large programs, individual grants reached £1M, and now there are also permanently available small grants up to £10k. (lrfoundation.org.uk)
£10k is not enough for an experiment. But they can be used practically as a relationship wedge: £10k → international workshop → prospective forecasting protocol → feasibility report → application for the next large program. This is especially good because the pitch can be built not around the controversial idea of "animals feel earthquakes." It sounds like: Exploring low-cost biological sensing as an additional independent layer for disaster-resilient early-warning infrastructure. Fit: 9/10.
6. AXA — one of the best large strategic funds
Here there is an almost perfect match of interests. AXA funds academic research on risks to people, society, and the planet. The AXA Chairs 2026 program provided funding of up to €1.5M for five years. (institution.axa-research.org)
The current 2026 cycle is already practically closed, so you need to aim for the next one. There is an important limitation: the applicant must be a university / public research organization. A commercial company cannot just come for €1.5M. Therefore, here I would build a model: University PI + seismology institute + animal-behavior group + statistical forecasting group + TRON technology team + insurance/risk partner. Moreover, AXA is especially interesting because the natural downstream market is precisely insurance. That is, the fundamental research program is immediately linked to the buyer. Fit: 9/10.
7. Kazakh money — I would definitely not ignore it
If the first experiment is really supposed to be in Almaty, it is logical to have a local grant recipient. In Kazakhstan in 2026, there were separate competitions for targeted science funding and grants for the commercialization of scientific results through the Science Fund. Most current cycles have already closed, but the mechanism exists and it is logical to prepare for the next round. (gov.kz)
Separately, Almaty held a competition for scientific and technical projects with total funding of 348 million tenge, up to 34.8 million tenge per project, with an emphasis on practical results and commercialization. (gov.kz) 34.8M KZT is just the order of magnitude sufficient for a local validation pilot.
A very good financial combination arises here. Not to ask one investor: €200k. But to assemble:
| Source | Possible order |
|---|---|
| Kazakh scientific grant | €50–70k |
| Emergent Ventures | €30–100k |
| Manifest Grant | €25–100k |
| Corporate sponsor | €50–100k |
| In-kind cameras/cloud/data | €20–50k equivalent |
And get the experiment without selling a significant stake in the company.
8. Horizon Europe: a big opportunity, but a consortium is needed
Horizon Europe Cluster 3 — Civil Security for Society — includes the Disaster-Resilient Society direction. And there, earthquakes, volcanic eruptions, and tsunamis relate directly to the considered natural hazards. In 2026, the budget of the corresponding block was tens of millions of euros; the current deadline for a number of Cluster 3 calls is November 5, 2026. (rea.ec.europa.eu)
Here it will no longer work: "Give €200k to the TRON team." You need a construction like: EU university + seismic institute + Central Asian field site + AI/animal behavior lab + civil protection authority + insurer / infrastructure operator. And most likely the project needs to be described wider: Multimodal distributed sensing for anticipatory disaster-risk management. Animal sensing becomes one of the sensor modalities. This significantly increases grantability. Fit: 9/10 thematically, 5/10 in terms of access complexity.
9. W. M. Keck Foundation — almost the ideal fund for "it sounds crazy, but it can be checked"
Keck is well known for funding research that: is too risky for regular grants; questions accepted concepts; requires new tools; can open a new scientific direction. Typical large research grants are in the range of about $1–1.3M for three years. (wmkeck.org)
In philosophy, this is almost an ideal TRON. The problem: Keck works through American non-profit research institutions. That is, you need, for example: an American university → PI → Kazakhstan as a field site. A startup will practically not get there directly. But if you attract a strong American seismologist or animal-behavior scientist, Keck becomes much more interesting. Scientific fit: 9.5/10. Organizational fit now: 3/10.
10. NOMIS — potentially the best "science patron," but only through reputation
The Swiss NOMIS Foundation finances precisely risky fundamental interdisciplinary projects that are difficult to get through standard academic funding. The fund makes very large long-term bets; for example, recently NOMIS announced multi-year programs for tens of millions of euros. (collegium.ethz.ch)
But there is a problem: a cold application is practically not an access route. NOMIS works through its own scientific network, recommendations, universities, and eminent researchers. Therefore, the path does not look like: TRON → NOMIS. But: TRON → strong scientific advisory board → world-class professor/institute → prospective pilot → publication/preprint → introduction to NOMIS. If the result of the first pilot turns out to be truly unusual, NOMIS is one of the most logical candidates for a €1–5M+ research program.
11. Eric & Wendy Schmidt / Schmidt Sciences
Here appears the class that the user usually means by "family office / private patron." Schmidt Sciences finances: AI for science, new scientific instruments, Earth systems, climate modelling, and large international scientific programs. In 2026, the organization, for example, allocated tens of millions of dollars to international scientific teams. (schmidtsciences.org)
But Schmidt Sciences explicitly states that it does not accept unsolicited proposals and searches for projects itself. (schmidtsciences.org) That is: You cannot do "Hello Eric Schmidt, give me $500k." You need to do: scientific result → respected PI → publication/preprint → scientific conference → Schmidt network / Science Philanthropy Alliance / university development office → invitation. But the thematic fit is quite high. Fit: 8/10. Cold accessibility: 2/10.
12. Jed McCaleb / Astera Institute
A very curious candidate. Astera explicitly talks about financing long-horizon, risky technical work, which is between the capabilities of traditional government funding and commercial capital. Jed McCaleb is known as the founder/financier of Astera. But in 2026, Astera somewhat narrowed its priorities towards: biological and artificial intelligence + AI-enabled life sciences. (astera.org)
Therefore, TRON needs to be shown not as an earthquake company. It is more interesting: Using distributed biological organisms as a sensor system: can ML extract a weak collective physical signal from the behavior of living systems? This is a very astera-like research question. Fit: 7–8/10.
13. Patrick J. McGovern Foundation
This is one of the most interesting private AI funds. The fund finances the use of artificial intelligence in the public interest, including climate resilience, crisis response, public infrastructure, and scientific tools. In recent cycles, there are grants of about $500–750k for individual AI/public-good projects. (mcgovern.org)
But: PJMF does not accept unsolicited grant proposals. That is, again, a warm introduction is needed. After the first proof pilot, I would rate it very high.
14. Science Philanthropy Alliance — not a fund, but potentially the main router
This is a very important structure. The Science Philanthropy Alliance does not issue regular grants to applicants. It advises private patrons and family foundations on science funding. The organization's network includes over 40 major scientific philanthropists, including structures associated with: Sergey Brin Family Foundation, Dalio Philanthropies, Schmidt Sciences, Simons Foundation, Sloan Foundation, Keck Foundation, Moore Foundation, NOMIS, and others. (sciencephilanthropyalliance.org)
Moreover, the Alliance reported that it has consulted dozens of private foundations and helped direct hundreds of millions of dollars of new capital into science. This means a very important thing: Do not search for 50 billionaires on LinkedIn. It is much more effective to build such a path: strong university PI → university development/philanthropy office → Science Philanthropy Alliance network → suitable family foundation. This is exactly how private science philanthropy often works in reality.
15. Which specific private-family directions I would keep on the long list
Not all of them have an open program specifically for earthquakes, so I would not write them a cold pitch. But for a warm-intro campaign, the list looks like this:
| Private capital / philanthropist | Vehicle | Our fit |
|---|---|---|
| Eric & Wendy Schmidt | Schmidt Sciences | 8/10 |
| Jed McCaleb | Astera Institute | 7–8/10 |
| Patrick McGovern legacy | PJMF | 8/10 |
| NOMIS private philanthropy | NOMIS Foundation | 9/10 |
| Sergey Brin-related philanthropy | Sergey Brin Family Foundation | 5–6/10, needs a special approach |
| Dalio family philanthropy | Dalio Philanthropies | 4–5/10 |
| Allen family philanthropy | Allen Family Philanthropies | 5/10 |
| Keck family philanthropy | W. M. Keck Foundation | 9/10, but only US institution |
Here I would consider the last three as network targets, not current open grants.
16. Corporate patrons and strategic sponsors
There is another class of capital that, in my opinion, may turn out to be better than a family office. Insurers and reinsurers. Because they have the opportunity to monetize the result almost immediately.
AXA — The most obvious through its own scientific program.
Swiss Re Foundation — The foundation finances resilience, public systems, and innovations, especially in developing countries. Thematically, pure earthquake is not their central vertical, but disaster resilience + risk infrastructure is already much closer. (swissrefoundation.org)
Munich Re Foundation — Together with UNDRR, it holds the RISK Award, where innovations in disaster risk reduction are funded; the prize size was up to €100k. The current cycle is already closed and has a narrower climate-resilience theme, but this is the right network of contacts for the project. (munichre-foundation.org)
From a strategic point of view, here I generally consider the insurer not as a "science sponsor." But as a future: design partner + data buyer + research co-financier. For example: the insurer gives €100k, the university €100k in grant money, TRON conducts a blinded prospective trial. If the signal is absent — a scientific result. If it is present — the insurer gets early access to a new data layer.
17. Convergent Research — very important if the hypothesis is confirmed
It is too early to apply there now for €200k as a regular startup. But imagine that in two years it is discovered: in three independent regions, the collective biological signal really precedes some of the events. Then the problem becomes completely different: who will build the global scientific infrastructure that simultaneously observes millions of animals, environmental signals, and seismic ground truth? Neither a university nor a startup will do this normally on their own.
It is for such situations that Convergent Research creates Focused Research Organizations — FROs. Such organizations are built around scientific infrastructure/public goods that are too large for a single laboratory, but are still poorly suited for venture capital. Convergent reports raising almost $400M for such organizations; a typical initial FRO may require $20–50M for 3–7 years. (convergentresearch.org)
That is, the potential evolution of TRON: €200k experiment → €2m multicountry validation → $20–50m Global Biological Precursor Observatory. It is at the third stage that Convergent becomes almost ideal.
18. What about regular deep-tech investors?
They are needed — but a little later.
HAX / SOSV. Of regular investors, I would test HAX first. They invest in hard-tech, sensors, robotics, AI + physical world, and the size of initial funding in the current program can reach about $550k. (hax.co) But HAX must see: hardware/software prototype + proprietary dataset + commercial buyer. One scientific hypothesis is not enough. A good time to apply: after the appearance of a working pipeline: camera/audio → animal anomaly model → spatiotemporal correlation → seismic baseline comparison.
Root Ventures. Root works specifically with very early-stage technical companies; their initial investments are usually in the range of about $3–5M. (root.vc) But I would go there after: prospective pilot + 2–3 LOIs from insurers / seismic operators.
DCVC. DCVC is especially interesting later because the fund has long worked at the intersection of: deep tech + AI + geospatial intelligence + risk. Its portfolio already includes companies related to geospatial risk and insurance. (dcvc.com) If TRON ever becomes: Pre-Seismic Risk Intelligence API, DCVC becomes an almost ideal profile fund. But at the current stage — it is too early.
19. What I would NOT do
I would not spend significant time now on standard "AI VCs" and generalist seed funds. The problem is too binary: either the new signal exists, or it does not. A regular VC does not get an advantage for financing the verification of a scientific hypothesis. But it gets a huge downside.
It is a completely different matter when you can already say: "We have registered 1.2 billion hours of baseline behavior, conducted 14 independent seismic events, the prospective model showed X relative to the seismic baseline." Then the scientific risk decreases sharply, and the data moat increases sharply. And the company's valuation can be several times higher.
20. Therefore, I would build financing in four rounds
- Stage A — "Can this even be checked?" €30–80k. Goal: no product. Needed: pre-registered protocol, independent statistician, retrospective test, access to data, first field partners. Financiers: Emergent Ventures + Fifty Years + Lloyd's small grant + in-kind university/cloud.
- Stage B — Prospective shadow-mode pilot €150–300k. 10–30 sites, preferably at least two independent geographies. Financiers: Global Innovation Fund + Kazakhstan + insurer sponsor + scientific philanthropy. It is exactly here that the currently requested €200k looks reasonable.
- Stage C — Multi-country validation €1–3m. You need to prove: signal adds information versus physical sensors. Then: AXA + Lloyd's Register Foundation + Horizon Europe + Keck + NOMIS + McGovern + Schmidt Sciences.
- Stage D — Commercialization / global infrastructure $5–50m+. Here the separation occurs. For-profit: HAX → Fifty Years VC → Root → DCVC → Lux/other deep-tech. Public-good infrastructure: Convergent Research / FRO + scientific philanthropy + governments + insurers.
21. Moreover, I would divide the organization into two parts
This can greatly increase the amount of available capital.
Research / public-good layer. University or non-profit consortium. It owns: scientific protocol, ground-truth benchmark, part of open data, publications. And has access to: AXA, Keck, NOMIS, Horizon, governmental science grants, and philanthropic money.
Commercial layer. Company. It builds: production pipeline, risk API, integration, customer-specific models, SLAs, enterprise tooling. And here enter: HAX / DCVC / Root / insurers. Thus, you do not force a scientific patron to finance a commercial SaaS, and a venture fund — fundamental science. Legally, IP/licensing and conflicts of interest here, of course, need to be оформлены cleanly in advance.
22. And the most important thing — how to sell this to them
The same idea cannot be sent to everyone with the same text.
- To a scientific patron: Not: "We will learn to predict earthquakes." But: "We will for the first time conduct a pre-registered, prospective, falsifiable test of the hypothesis whether distributed biological systems contain additional information about pre-seismic processes."
- To a disaster-resilience fund: "We are checking the possibility of creating an ultra-cheap independent sensor layer for early warning in regions with insufficient instrumental infrastructure."
- To an AI fund: "Is it possible using multimodal AI to detect an extremely weak physical signal by correlated changes in the behavior of thousands of independent biological sensors?"
- To an insurer: "Can a new data layer give measurable additional lead time, allowing to reduce expected loss?"
- To a VC: Only after evidence: "We are creating a proprietary pre-seismic risk data network and API for insurers and critical infrastructure; the unique asset is a longitudinal biological dataset that cannot be quickly reproduced."
This is a very important change in packaging.
My shortlist: who to write to in what order
If the task is to really raise the first €200–300k, my order of actions would be:
| Priority | Target | Action now |
|---|---|---|
| 1 | Emergent Ventures | apply now |
| 2 | Global Innovation Fund | submit initial application |
| 3 | Fifty Years Manifest | prepare 2-page scientific proposal |
| 4 | Lloyd's Register Foundation | small grant + get to the programme team |
| 5 | Kazakhstan / Almaty / Science Fund | find an accredited local scientific partner and prepare for the next cycle |
| 6 | Horizon Cluster 3 | urgently check the exact match to the current call before Nov 5, 2026 |
| 7 | AXA Research | assemble an academic consortium for the next cycle |
| 8 | insurance corporate sponsor | AXA / Swiss Re / Munich Re and analogues |
| 9 | Keck | find a US university PI |
| 10 | NOMIS | only through an authoritative scientific intermediary |
| 11 | Schmidt / McGovern / Astera | warm introductions after the scientific core |
| 12 | HAX | after prototype |
| 13 | DCVC / Root | after prospective evidence |
This also corresponds to the logic of low-cost entry, which we have already formulated for the Superradar: first cheap proof of result and shadow mode, and not building the whole system before the appearance of evidence.
If you set one goal
I would not try to find one patron for €200k right now. I would set the goal: in 90 days to form a €250–350k blended research round from 3–4 sources without significant equity dilution. The optimal construction looks approximately like: €50–100k private scientific grant + €100–150k international impact/disaster grant + €50–100k Kazakhstan + €50k strategic insurer/in-kind.
And in parallel, you need to find one very strong academic PI, because it is he who opens the doors to AXA, Keck, NOMIS, Schmidt, Horizon, and a large part of family scientific philanthropy. This is potentially more valuable than another angel investor.
If the first prospective trial shows a real incremental signal, project financing turns from the problem of "who will believe in a strange hypothesis?" into the problem of "who will be the first to get access to a potentially new category of catastrophic-risk data?" — and this is a completely different negotiating position.
User
and now export the entire dialogue into a docx document, exactly, verbatim, check the correctness and neatness of formatting tables, links, headings.
AI
[End of transcript]
