Where memory fades, power escapes.

Memory is made of records. Records are made of information.

For centuries the excuse was scarcity. Records were incomplete, decisions were taken behind closed doors, and power was difficult to understand and harder to challenge. That excuse is gone. More information is produced in a single day than entire generations once created in a lifetime.

And yet understanding is more fragmented than it has ever been. Facts scatter. Records disappear. Connections stay invisible. Public memory fades — and wherever memory fades, power escapes accountability.

Deep Tunisia exists so that Tunisia remembers: seventy years of appointments, institutions, ruptures and relationships, held in one place, in the open, in a form anyone can audit.

A company means little until its ownership is understood. A decision means little until it is placed in the era that produced it. Deep Tunisia exists to connect those fragments into a public understanding that no single document can provide.

The join that makes this possible is deliberately unglamorous. A position is (role · holder · start · end) — an interval, not a fact about a person. You cannot ask who held power in 1994 from a list of people. You can only ask it from a set of intervals.

Dates are ranges, never points: earliest and latest start, earliest and latest end, plus a status. The timeline draws the documented core solid and the uncertain envelope hatched, so a span nobody has confirmed can never quietly pass for one that has been.

Successions are derived, not asserted — computed from the intervals themselves, so a hole in the record shows up as a hole. There are currently 39 succession gaps and 9 overlapping tenures, and all of them are published.

There is no scraper filling this graph and no language model writing into it. Knowledge enters through one door, in a fixed order, and the build refuses anything that arrives out of order.

The rule that matters most: an inference is never allowed to become a fact. Every claim is filed under one of four bases, and no two of them are ever drawn the same way.

The stated architecture of this project is that machines propose and humans verify. Almost nothing has yet been through a second pair of eyes, and the honest number is published on the site rather than kept in a drawer — because that is the only way the commitment means anything.

Worse than the total: all 24 reviews so far have landed on the best-evidenced claims. Of the 4 unsubstantiated claims, the 44 attributed ones and the 83 inferences — the records most capable of harming a named person — not one has been independently checked. That is the single highest-value work outstanding, and it is why funding this project buys verification before it buys anything else.

A nation also needs a place where its people can think together. Deep Tunisia is more than an archive — it includes Tunisia's digital Agora: a discussion layer where citizens can understand institutions, question them, and improve the public record together. Not a replacement for institutions, but a space for holding them to account.

The difference from every other forum is one design decision: the conversation is anchored to the record, with the record still in view. You are not transported to a generic thread and expected to remember what you were discussing. The entity's trajectory, offices, documented connections and sources stay beside you while you argue about them.

Elsewhere you get an opinion. Here the discussion has an object — and if you think the object is wrong, there is a mechanism for saying so that ends in the graph changing.

Citizens should never be forced to choose between participation and personal safety. So there is no account here. Your identity is a keypair, and the only thing that can rebuild it is twelve words generated on your device.

The words are generated, never chosen, and that is the load-bearing decision. Because a public key is public, anyone can attack every identity on the site at once, offline, with no rate limit to hide behind. A phrase a human invents carries perhaps 20–30 bits of real entropy and falls in hours. Twelve words drawn from a 2,048-word list carry 128 bits. It does not fall.

They are shown once. Never transmitted, never stored, never permitted into a URL — which is why recovery can never be a link someone emails you. The same twelve words rebuild the same identity on any device, forever. There is no recovery process, because there is nothing on any server to recover from.

Every other rate limit in this project exists to stop an attacker. This one does not. It exists because a post that took forty seconds and a post that took forty minutes currently render identically — and this whole project is an argument that things of different standing must never look the same.

Scarcity is the instrument. A person with two posts a week spends them on something they have thought about, because they can feel the cost. A person with sixty an hour reacts. No amount of guidance in the composer produces the first behaviour; a budget produces it without saying a word.

And the budget falls on assertion, not on evidence. A post carrying a source that resolves costs half of what the same post costs without one — not because cited posts are true, but because the cost of a post should track how much work it hands to everyone else.

Bourguiba, Ben Ali, Saied, Ennahda, the opposition, the army, the police, the business families, the unions, the EU, France, Italy, Algeria, the United States, Russia, the World Bank and the international NGOs are all recorded under identical rules. An atlas of Tunisian power containing only Tunisians reads as an indictment of Tunisians.

This is not neutrality for its own sake. It is the project's only real protection. Credibility that depends on people agreeing with the conclusions collapses the moment they don't. Credibility that depends on the evidence being accessible and reproducible does not.

A prison sentence is a documented fact; the allegations behind it and the criticism of the proceedings are contested, and both framings are stored with attribution rather than adjudicated here. Wealth is not corruption — the dataset distinguishes business families that predate a regime from ones that regime created, because collapsing that distinction is the most common error in popular accounts.

Deep Tunisia belongs to no party, no government and no ideology. It is open source under version control, its exports are downloadable in full, and its methodology, corrections history and funding are published on the site rather than described in a brochure.

It is built so that it survives its authors. There is no database to seize, no account table to subpoena and no private archive to lose: roughly 1.3 MB of JSON ships with the page, and the canonical record is plain text anyone can fork.

The binding constraint is not data. It is verification capacity. Adding records is cheap and getting cheaper; a second pair of human eyes is neither. Support is spent against that, in this order.