Power, on the record.
What DeepTunisia is, the engine it runs on, the research it carries out, and the ideas that hold all of it together.
Open sourceCode MITData CC BY 4.0
In one page
A public record of who has held power in Tunisia since 1956, and how we know it.
Every president, minister, governor, security chief, party, business group and foreign actor we can document sits on one connected map. Every line on that map says what kind of evidence stands behind it.
Some lines rest on an official decree. Some rest on a press report. Some are our own reasoning, and say so. Some are rumours that circulate without proof, and those are kept too, clearly marked, so the map can show why they fail.
Underneath runs DeepEpisteme, an engine we built that refuses to publish a claim without its evidence. It is the part of the project we care about most.
Around the map sit the parts people use: the atlas itself, open research such as the Police Satisfaction Index, long form investigations, and Agora, a pseudonymous space for discussion that opens soon.
Everything is open. The code is MIT licensed, the data is CC BY 4.0, and every change to the record is public and dated.
- People
- 2,476
- Institutions
- 312
- Dated positions
- 3,056
- Relationships
- 2,551
- Sources
- 2,564
- 7 eras
- 1956 → 2026
Counted by the build that published this page. When the record changes, these change with it.
01/Philosophy
Freedom is not declared. It is enabled.
A society can hold intelligent, educated, ambitious people and still be unable to change itself. The missing variable is rarely intelligence. It is infrastructure.
DeepTunisia is the first project of DeepOntology, an open research organisation built on one premise. The systems through which people understand, decide and govern are themselves made things. Made things can be designed, inspected, tested and improved, in the open. Four ideas follow from that.
I
Institutions are infrastructure
Rules that govern people work much like software. When rules are written down, versioned and public, anyone can check them, and anyone can propose a better version.
II
Legibility is power
Who can see whom, who can identify whom, who can change the record. Whoever answers those questions holds real authority. We want citizens to see the state clearly without the state seeing them.
III
Truth needs structure
When evidence and assertion are mixed together, public debate becomes a contest between stories backed by power. No civic system is stronger than its relationship with evidence.
IV
Build with Tunisia, not on it
Tunisia is where this has to survive real conditions: scarce trust, uneven institutions, real legal risk. The people in it are participants, critics and co-builders, never test subjects.
We do not design the perfect society. We research bases on which societies can design themselves, more transparently.
What we hold ourselves to
These are not values on a wall. Most of them are enforced by the software, so they do not depend on anyone remembering them.
An inference is never allowed to become a fact.
Every claim says what kind of claim it is. A guess can be useful, but it must look like a guess, carry its reasoning, and say what would prove it wrong.
We do not start from a conclusion.
This is not a site exposing a “deep state”. That framing assumes its own answer. We ask a question the evidence can answer: which institutions, people and economic groups keep their influence when the regime changes.
We test 6 hypotheses in public. At the moment the evidence fails to support 2 of them, including the strong version of the deep state thesis. The results are on the evidence page.
Everyone gets the same standard.
Bourguiba, Ben Ali, Saied, Ennahda, the opposition, the army, the police, business families, unions, the EU, France, Algeria, the US, Russia and the World Bank all sit under identical rules. Credibility that depends on people agreeing with us collapses when they stop. Credibility built on evidence anyone can check does not.
Rumours are recorded, not deleted.
Deleting a popular claim does not make it go away. It only means nobody can see why it is wrong. The claim that the army secretly governs Tunisia sits next to the record of the tri-service command left unfilled for two decades.
We publish our own weaknesses.
Of 8,764 records that can be reviewed, 52 carry an editorial review note, and none has yet been independently verified. That number is on our About page because it is the only way the promise means anything.
Nothing is claimed built until it can be inspected. Everything ships open.
What survives belongs to whoever forks it.
02/The structure of everything
One organisation, one engine, one country, four doors.
Parent organisation
DeepOntology
open research on how societies know, decide and govern
The engine · how we know
DeepEpisteme
refuses any claim without its evidence
Second engine · conception
APM
civic agency without giving up privacy. No code yet.
First deployment · live
DeepTunisia
the engine running on a real country
Next deployments
Anywhere
another place forks the engine and changes one settings file
Live
The Atlas
the knowledge graph, its maps and timelines, and the open data
Live
Research
open studies of how people live with power
Live
Media
investigations told from the same evidence
Opening soon
Agora
pseudonymous discussion, corrections and proposals
Think of it in layers. DeepOntology is the school of thought. DeepEpisteme is the machine that turns that thinking into rules a computer enforces. DeepTunisia is the first place the machine runs on a real country. The four doors are how people actually meet it.
The layers are separate on purpose. Nothing in the engine is specific to Tunisia, so another country, city or community can take it and inherit the same discipline. A second engine, APM, asks how citizens could take part in public life without giving up their privacy. It is an idea on paper, with no code yet.
03/The engine
The epistemology is the compiler.
When a programmer writes code, a compiler checks it before it runs. One broken rule and the program does not build. Nobody has to remember to check. It simply cannot ship broken.
DeepEpisteme does the same for claims about the world. Our rules about evidence are not a style guide that editors are asked to remember. They are written as code, and every time the record is rebuilt, the engine checks every claim against them. If a single record breaks a single rule, nothing is published. Not the map, not the data files, not one page.
Why go this far? Because rules enforced by people alone wear down. They bend under volume, fatigue and political pressure, and now AI can produce convincing fake citations at almost no cost. A record about power in Tunisia will be attacked. It has to hold without relying on anyone’s good intentions.
The engine refuses
- a claim with no source
- an inference without its reasoning and what would prove it wrong
- a weak claim that hides who makes it
- a rumour that hides where it circulates
- an impossible date, or a term that ends before it begins
- an influence claim tied to nothing documented
Every rule keptthe record is published
One rule brokennothing is published
The boundary, stated plainly
The engine checks that every claim keeps its contract with the evidence. It does not decide whether a claim is true. That judgement stays with people, and it stays visible. In the words of our paper, the dataset cannot be quietly wrong.
Try the gate
Build a claim and send it through the engine. These are the real rules, in miniature.
✕
Build fails. Nothing is published.
- An inferred claim must state its reasoning and what would falsify it.
Every claim travels with its evidence
Each claim in the record, whether it is a person holding an office, a family tie, or an event, carries the same nine fields of evidence. They are bundled together and can never be separated.
| Field | In plain words |
|---|---|
| Sources | where it comes from, ranked from tier 1, an official gazette, to tier 5, a lead |
| Kind | documented, reported, inferred or unsubstantiated |
| Strength | how solid the evidence is, graded A to D |
| Verification | how it was checked |
| Said by | who makes the claim, if it is weak |
| Reasoning | why we infer it, if we do |
| Falsifier | what would prove it wrong |
| Disputes | who disagrees, and with what source |
| Review | who reviewed it, when, and how |
Four kinds, drawn four ways
- Documentedan official record states it
- Reportedcredible outlets report it, with attribution
- Inferredour reasoning, with a falsifier attached
- Unsubstantiatedcirculates without proof, hidden by default
A reader can raise the evidence floor to “documented only” and watch much of the network dissolve. It is one of the most revealing things the site does.
Dates are ranges, not points
may have startedcertainly in officemay have ended
When nobody published an exact date, the uncertainty is drawn as uncertainty rather than rounded into false precision.
Raise the evidence floor
This is the record as it stands today. Choose the weakest kind of claim you are willing to see, and watch how much of the map remains.
5,800 100% 5,800 of 5,815 claims stay on the map
- Documented 2,237
- Reported 3,493
- Inferred 70
- Unsubstantiated 15
We attacked it
A self audit found seven ways to slip past the rules. Each was closed and pinned by a test. Then we broke the engine on purpose 26 different ways. It caught 25.
It is reproducible
The same data always builds the same record, byte for byte. Each release publishes a fingerprint. This is the one behind this page:
9be9c150ecb31ef9…
It is portable
Everything specific to Tunisia lives in one settings file. A fictional test country called Ruritania runs through the same checks to prove it. The engine is being extracted into its own repository now.
04/The atlas
70 years of power, on one map.
The atlas is the window onto the record. Each view answers one question, and every line in every view still carries its evidence label.
| View | The question it answers |
|---|---|
| Chronicle | Who held which post, and when, drawn to scale. Bar length is time in office, so continuity is visible at a glance. |
| Now | Who holds each office today, how recently that was confirmed, and how long they have lasted against the usual tenure. |
| Network | Who connects to whom. Institutions are boxes, people in post are seats, and a line appears only for a sourced tie. |
| Atlas | How the structure looks era by era, and who survives from one regime into the next. |
| Rankings | Separate measures of structural position, with weights the reader controls. There is deliberately no single “power score”. |
| Map and World | The governorates and strategic sites, and Tunisia’s trade, debt and treaties with the rest of the world. |
| Investigate | Structured questions over the graph. Answers come from following the records, never from a language model. |
| Evidence | 6 hypotheses tested in public, 26 open questions, 78 contradictions between sources kept side by side. |
126
gaps in who held an office, found by the engine and published rather than hidden.
78
recorded contradictions. When sources disagree, both are kept with who holds each view.
236
records still waiting for a primary source, listed in full as a public worklist.
Every change to the record appears on the Corrections page, generated from version history so nobody chooses which changes are flattering enough to show. The interface works in Arabic, French and English, and is designed first for phones, where most of our readers are.
05/The research pipeline
Measuring how people live with power.
The atlas documents the state from the record. The research program asks the people who live under it.
The Police Satisfaction Index
Open since October 23, 2026How satisfied are people with the police they live with? Answers become one number from 0 to 100, published every month like a market index, with live counters during the month.
0 · Police stateGuardian · 100
Trust ↑
Are the police fair, accountable, on the public’s side?
Service ↑
Do they come when called and handle contact well?
Grip ↓
Do people feel watched, afraid, unfree around them?
Harm ↓
Have people or those close to them been abused?
Why anyone should trust the number
- The questionnaire was frozen and fingerprinted, and the method registered in public, before the first answer arrived.
- Every day, the month’s counts are written to a dated public ledger. Once a month closes, anyone can check that no answer was added, removed or changed.
- Every exclusion rule, and how many answers it removed, is published.
- Every number reads “among respondents”. An open online survey speaks for the people who answered it, never for all Tunisians, and we say so every time.
Why it is safe to answer
- No name, email, phone number, account, IP address or device fingerprint is stored. There is no field that could hold one.
- No questions about age, gender or address. The only location question is optional, and results for groups of fewer than 20 people are never shown.
- Each person gets a receipt code and can delete their answers until the month closes.
- Survey answers never enter the knowledge graph. An answer is an observation about one person, never a fact about Tunisia.
A platform, not a one-off survey
We built the research system so that a new study is a configuration, not new software. Each study adds a protocol, a questionnaire and a registry entry, and inherits everything else: privacy, consent, withdrawal, the ledger and publication.
- Designprotocol and questions
- Reviewethics, law, translation
- Freezefingerprint and pre-register
- Fieldopen, with live counts
- Releasemonthly figure and ledger
- Open datarecomputable by anyone
Each wave repeats the frozen questions, so change over time is measured, not guessed.
| Study | The question | Stage |
|---|---|---|
| Police Satisfaction Index | How people live with the police, every month | Open |
| Grading reliability | Do independent people grade the same evidence the same way under our rules? | Ready |
| What democracy means | Which parts of democracy people call essential, and how present each is in Tunisia today | In review |
| Police stories | A second door into the index: a few structured questions, then a personal account | Designed |
| Quality of life | The next national index on the same platform | Planned |
The goal is an independent statistics layer. Tunisia has no public, repeatable and checkable measure of how people experience the state. Anyone can say things are fine or that they are not, and nobody can recompute either claim. Each index is meant to be that record.
Stories are the most sensitive part. A personal account can identify the person who wrote it, and Decree-Law 54 is used against critics. Stories open only once the protections are in place. Later, a partner polling firm or university could run the same questions on a random sample, so findings can speak about Tunisians, not only respondents.
06/Agora Opening soon
A public square that cannot rewrite the record.
Agora is where readers discuss any person, office or claim, report errors, and propose changes. It is built, tested, and kept closed in production by a server switch until it opens as a public beta.
Agora
discuss, report, propose
The same gate as every claim
- it needs a source
- a person reviews it
- the engine’s build must pass
The record
changes, with a dated entry
Identity without an account
- No email, no password, no phone number, no social login.
- Your identity is a cryptographic key created inside your own browser. It never leaves your device, and the server never sees it.
- Twelve words rebuild it on any device. Nothing is stored anywhere for anyone to seize.
- For something risky to say, “post once” makes a throwaway identity that is discarded after use.
- We say pseudonymous, never anonymous. Writing style can identify a person over time, and the composer says so before anyone posts.
Why bots cannot win
- Community posts are never project data. Votes rank discussions. They never grade claims.
- New identities start with tight limits that grow with trust. Moderation rights are never earned automatically.
- So the worst a perfect bot attack can do is reorder a discussion page. The record itself is out of reach.
- That protection comes from the design of the evidence rules, not from checking IDs, which in Tunisia would mean handing identities to the state.
Why it is still closed
Opening Agora is a deliberate decision, not a code change. Its identity design depends on one legal question: whether Tunisian law allows a service to run without being able to identify its users. We are getting a proper legal answer before the first public post, because the people most likely to use Agora are the people with most to lose.
07/Privacy and security
Built for a country where speaking has a cost.
In Tunisia, Decree-Law 54 is used to prosecute critics. A reader looking up a security chief, or a citizen describing a police encounter, should leave no trail that can be turned against them. We designed for that from the first line of code.
Reading
No analytics, no advertising, no tracking scripts, no requests to other companies. Fonts are hosted by us, so even typography tells no one what you read. Our host runs a security check that sets one cookie, and we disclose it.
Answering a survey
No identifiers are stored with an answer. To stop floods, the server uses a salted fingerprint of the connection that changes every day and is kept only for the hour it covers. The server that receives answers keeps no request log.
Taking part in Agora
The identity key is created and kept in the browser. The server stores no IP addresses and holds nothing that could unlock a user if it were seized.
The site itself
The atlas is plain static files with strict security headers. Every interactive service stays closed unless deliberately switched on. Releases are signed and fingerprinted, and a failed build publishes nothing.
What we refuse to do
We will not verify people by phone number, because in Tunisia every SIM is registered to a national ID. We will not use the state’s digital identity service, because it would tell the state which citizen logged in to DeepTunisia, and when.
What we cannot protect, and say so
Like any website, your internet provider can see that you visited. We cannot prevent that, so we tell people plainly that a VPN or Tor hides the visit. We test our own defences adversarially and publish the limits instead of overstating the protection.
08/Built to last
Designed to outlive its own budget.
Public-interest projects usually die when money, attention or a key person runs out. We designed DeepTunisia around each of those failure points.
It runs at close to zero cost.
The infrastructure fits within free hosting tiers, with no payment card on file. If a spam flood hits a limit, the service pauses instead of generating a bill.
The record survives an outage.
The atlas is plain files. If an interactive service goes down, the record stays readable.
Nobody can quietly lose or change it.
The whole dataset is plain text under version control. Anyone can copy it and rebuild the site from scratch, and every change has a date and an author.
Growth does not mean new software.
A new index is a configuration of the research platform. A new country is a fork of the engine with one settings file changed.
Small team, large leverage.
DeepTunisia was built by one person working with many AI models, and we disclose this openly on the colophon. The rule never changes: models propose, people decide, and the engine checks everything either of them writes.
Independent by rule.
Every funder is published with amount, period and purpose. No funder may influence conclusions, research targets or how a claim is classified. The current list is on the About page.
It keeps itself current.
A news feed refreshes every four hours. Office holders without a recent source are automatically marked “last verified” instead of “current”, so the record cannot silently go stale. The figures on this page are counted the same way, by the build that published it.
09/Contributing
Anyone can change the record. Nobody can change it quietly.
The record is meant to be corrected by the people who read it. There are two ways in, and both pass through the same gate as our own work.
Open now
A pull request
for anyone who uses GitHub
Opening with Agora
A proposal on Agora
for anyone, no code needed
The same gate for everyone
- a source behind every change
- a person reviews it in public
- the engine’s build must pass
Accepted
enters the record, with a dated entry on the Corrections page
Declined
the reason is written down and stays public
For developers, today
Anyone can open a pull request right now. It can be a small data fix, such as a corrected date or a missing source, or a whole new feature. The checks that guard our own work run on yours. If a contribution is accepted, it becomes part of the dated history. If it is declined, the reason is written on the pull request and stays there for anyone to read. Nothing is turned away in private.
For everyone, with Agora
Changing the record should not depend on knowing how to code. Agora brings the same power to every reader: propose a correction from the page you are reading, and it becomes public and open to discussion. Anyone can join the discussion from day one. Proposing changes opens as an identity earns trust, which keeps floods of bots out without asking anyone who they are.
A place to investigate together
Many facts about power are never written in a single document. They sit in pieces: a decree in one place, a company filing in another, something a person saw years ago. Agora is meant to be the place where people gather those pieces, compare what they know and coordinate before submitting a change. Put together carefully, pieces that mean little alone can become evidence that holds up, and reveal what was invisible to any one person. Whatever comes out of that work faces the same gate. Discussion is never data. Evidence is.
10/Where we are
Early, honest, and open to everyone.
| Live | The atlas and open data, the Police Satisfaction Index, The Chemical Century investigation, and the engine’s research paper, The Epistemology Is the Compiler. |
|---|---|
| Built, opening | Agora as a public beta. The grading reliability study, waiting for raters. |
| Designed | The democracy study, police stories, a quality of life index, and the engine as its own repository. |
| Imagined | APM, a civic agency engine. More deployments beyond Tunisia. |
This is not ours. It belongs to whoever picks it up.
DeepTunisia is open source by design. It is not meant to be owned by an organisation, and no single person or group should be able to close it, buy it or bend it. The code, the data and the method are public so that anyone can use them, challenge them, improve them or carry them somewhere new. It has no single kind of contributor. Anyone can take part.
Journalists
Test the record against what you know. Cite it, challenge it, and tell us what is missing or wrong.
Researchers
Review and evaluate the method, the engine and the studies. Grade evidence as a rater, review ethics or methods, co-author.
Organisations
Partner on a study, lend expertise, run a survey wave with us, or introduce us to people and institutions we should know.
Creators
Turn records into stories that keep their labels: what is documented, reported, inferred, or only a rumour.
Builders
Improve the code, translate, test the security, or fork the engine for another place.
Anyone
Send feedback, point to a source, share a contact, or tell us where the site is confusing.
If you agree with it, it is already yours.
If you want to own it, you already do.
- Site
- deeptunisia.org
- Code and data
- github.com/deepontology/deeptunisia
- Contact
- contact@deeptunisia.org