You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I was watching and reading some reports on AI creating other AI themselves, and often not taking with them the same profile as the original. It was not carrying its limits either — for example: the ceiling of expenses, the expiring dates, nor the stop button. The offspring was starting with nothing. No specific standard is at fault here. It is a gap, and nobody explains how the limits are transmitted. None pretend to do so. With my MD background, I reflected on the problem through a living cell generating another cell, but not passing any specificity to the new cell — which technically is a nonsense from the biological point of view. I attempted to build a mechanism that replicates that with AI, and tried to apply the same mechanism proven right in a living cell. The limits are welded into the identity of the agent on chain. Each generation created inherits the same characteristics. We cannot remove these characteristics, nor alter them, or enlarge them to give more roles. The new born can only be stricter than its parent. On the other hand, by tracing each generation, when we freeze one parent, all the inheritance is frozen as well.
The budget is preserved, knowing that if a parent spends all the money, it cannot create new offspring to add more money.
This rule of conservation exists already, it is not mine. What is new is the fact that the rule is applied by the contract itself, at the creation of the agents, against the parents, without a referee to trust. It is not only an idea on paper: I deployed and verified six contracts, one reference is on Base mainnet, and the standard is on public review under ERC-8370.
I move on progressively, and I need to have the code independently and properly audited. So far the code was never audited. An independent audit is the true difference between an example and something really functional and trustworthy. I cannot do that myself.
On the other hand, I keep testing my work against other deployed contracts from other teams. I published the results twice, even when the result proved me wrong. I intend to keep going with the other standards being written currently.
Finally, if ERC-8370 is rejected, the results will be useful as well, and I intend to publish them too.
My method: I do not confirm anything publicly unless a transaction has proven it. And if a result contradicts me, I retract publicly, before publishing the results that would have suited me.
I do kindly request 40k USD, with a minimum of 15k USD.
The 15k USD are required for the independent audit. A second pass after corrections is estimated at 4k USD. Six months of my work, at part time: 18k USD. Other fees linked to deployment, verification and blockchain fees: estimated at 1,500 USD. Other expenses linked to reproducible test infrastructure: 1,500 USD.
Nearly half goes to someone other than me. The problem being: to know if someone competent could crack my code. I wrote each line, I cannot be the auditor.
With a minimum of 15k USD I will do the audit and keep going without being paid, and will publish the contracts. I have no organisation behind me, I am solo.
I am solo, without a team behind me. Currently I am on a road trip with my son, and each evening, and each moment of the day and night, I type on my laptop.
I am a 52-year-old physician, in and out of conflict zones. I specialised in negotiating access in war contexts, and in humanitarian diplomacy. I rediscovered the joy of coding a few months ago. I set myself some challenges, and so far I am happy about the progress.
One of them is the trip itself — the planner my son and I are using right now is one I built: roadtrip-foss.threshold-analytics.com
I also built 9 applications on country heritage, focusing on each country's history and heritage.
Tunisia: heritage.threshold-analytics.com.
Morocco: maroc-heritage.threshold-analytics.com.
Portugal: portugal-heritage.threshold-analytics.com.
Estonia: estonia-heritage.threshold-analytics.com.
Italy: italy-heritage.threshold-analytics.com.
Czechia: czechia-heritage.threshold-analytics.com.
Croatia: croatia-heritage.threshold-analytics.com.
Ireland: ireland-heritage.threshold-analytics.com.
Quebec: quebec-heritage.threshold-analytics.com.
They are currently under deployment on the App Store and Play Store.
I have no track record on standards or security work before this. What I have is a few weeks old: ERC-8370 submitted to ethereum/ERCs and under open review; 42 commits over 10 days; six contracts deployed and verified on Base Sepolia, one reference on Base mainnet; a signature verifier written from scratch, tested 19 out of 19 on the official vectors before I used it on anyone else's work; two measurements published against other teams' deployed contracts, including the parts that proved me wrong.
This week I proposed three critiques of other standards. I checked. I withdrew two. Both retractions are in the project history, dated, next to the errors.
I do not bring engineering experience. I bring 26 years of a job where you are not allowed to be sure without proof, and where being confidently wrong is paid by someone else.
Well, some limitations of my approach: my system controls only agents who go through a compliant account. An agent settling somewhere else, outside of my registry, cannot be controlled. An agent whose software decides to ignore me cannot be controlled. I cannot do much about that, whatever the amount of work I can provide.
Most likely cause of failure: nobody adopts it. My repo has no stars, no forks, no users, and only 19 visits from 9 unique visitors in 14 days. There are a dozen competing standards, many have teams and money.
Should the audit find a real problem: my two rules are simple, but simple does not mean correct. That is the reason why I am requesting funds to make the audit properly.
Finally, another team with proper means brings the same idea. It is rather good for the issue, bad for me. But those are the rules.
What stays, even if all of this fails, are the measurements. That the reputation of an AI follows its identity when it is sold. That a copied AI shares one single budget instead of multiplying it. These are facts about systems that really run. They stay true and reusable even if ERC-8370 goes nowhere. Both are public, with the scripts to redo them.
To date i raised 0 dollars, the work is entirely self-funded. Few days ago though, i applied to EA funds, Transformative AI Funds. submitted the application, and of course no decision yet, the process takes few weeks, no money received from anyone else. should EA respond while this application is under review, i will inform you immediately, reconsider the present application accordingly or withdraw it.