How to read this

The rule that matters most

An inference is never allowed to become a fact. Every claim is filed under one of four bases, and they never render identically. Merging them is precisely how a research project becomes a conspiracy chart.

The evidence control in the toolbar sets the floor. Raising it to documented-only is the most informative thing you can do here: large parts of the network simply dissolve, because much of what circulates about Tunisian power is reported interpretation rather than established record. Feeling that difference is worth more than any disclaimer.

Two of these categories carry extra obligations enforced by the build. An inferred claim must state its reasoning and what evidence would refute it, or the build warns. A claim at the weakest confidence grade must name who is making it, or the build fails outright. There is no way to slip an unattributed insinuation into this dataset.

Why unsubstantiated claims are kept

Most projects delete them. That is a mistake. Deleting a widely-held claim does not make it go away; it just means the map cannot show anyone why the popular account is wrong. So they are recorded, labelled, attributed to where they circulate, kept out of the graph by default, and paired with what the evidence actually shows. Several of the ones on record here are contradicted by this project's own data — that is the most useful thing a category like this can do.

Dates are ranges, not points

Historical personnel records are rarely precise. A source may say an officer was appointed on 1 June 2018, or that he was in post by at least December 2025, or that he took over around 2017. Collapsing all three into one timestamp would invent precision the sources do not support.

So every span resolves to four values: earliest and latest possible start, earliest and latest possible end. On the Chronicle, the documented core of a tenure is drawn solid and the uncertain envelope is hatched. A bar that fades at its right edge means the officeholder was confirmed in post at that date and may well have continued; it does not mean they left.

One stated assumption: an open-ended <=2013 bound, meaning "in post by 2013, start unknown", is bounded to an eight-year window rather than opened back to 1956. Left unbounded it would silently place officers in every earlier era and inflate every tenure-based metric. Eight years is a plausible ceiling for a senior Tunisian command; the longest documented run in this dataset is about seven and a half.

Successions are derived, not asserted

Nobody hand-maintains the command chains. They are computed from the position records at build time, which means a hole in a chain shows up as a hole. The build currently detects 43 gaps and 5 overlapping tenures, all published on the evidence page. Some gaps are real history: the post of Chief of Staff of the Armed Forces genuinely lapsed for roughly two decades. Others are missing records, and saying which is which is the research agenda.

When sources disagree, the disagreement is the record

If one source dates a tenure to 2018 and another to 2019, this dataset does not pick the better-cited one and move on. It stores both, names who holds each, and states whether either was adopted and why. There are currently 7 recorded contradictions, including one where five different prison sentence lengths circulated across outlets for overlapping proceedings and the dataset declines to guess how many distinct cases there were.

There is no single power score

Five indices are computed separately and shown side by side. Four are pure functions of the data. One is an editorial judgement, and it is published below rather than buried in code.

• Formal authority. The constitutional and command weight of the most senior post held. From a published, editable weight table. An editorial judgement, openly stated.
• Proximity to the executive. How few institutional steps separate this person from the presidency. Computed: shortest path to the presidency node over documented edges only.
• Tenure and survival. Years in senior post, weighted by how many regime ruptures were survived in office. Computed purely from position intervals. The most defensible index here.
• Brokerage. How much this person connects parts of the network that would otherwise be separate. Computed: normalised betweenness centrality. Surfaces intermediaries over headline names.
• Cross-layer reach. How many of the analytical layers — security, political, economic, media, judicial, civil society, foreign — the person operates in. Computed from layer membership, as a share of the layers currently enabled. Narrowing the layer filter changes the denominator.

The rankings page offers weight sliders so you can combine them according to your own theory of how power works. The ordering changes when you do. That is the intended lesson: who counts as most powerful depends on what you decide to measure, and there is no view from nowhere.

A high brokerage score means the dataset records someone connecting otherwise separate networks. It is not evidence of wrongdoing, coordination or hidden control.

What the layers mean

Seven analytical layers: security, political, economic, media, judicial, civil society, and foreign and international. Anyone carrying more than one is a candidate bridge, and bridges are what this project is hunting. The most overlooked category of power in Tunisia is probably not the richest businessman or the highest-ranking general, but whoever connects security, presidency, capital and political access. The network view puts each layer in a fixed lane so those crossings are visible as the edges that cross lanes, and the investigate page can enumerate them directly.

The foreign layer is not decoration. An atlas of Tunisian power containing only Tunisians reads as an indictment of Tunisians. External actors shape migration policy, macroeconomic policy, security assistance and civil society funding, and they are held to the same evidentiary standard as everyone else.

Two structural facts worth stating plainly

The National Guard is not part of the armed forces. It sits under the Interior Ministry while performing military-style gendarmerie duties. Conflating the two makes the Tunisian security system look far more unified than it is, and it is a common error in secondary accounts.

The Chief of Staff of the Land Army is not the Chief of Staff of the Armed Forces. Secondary accounts routinely upgrade the former to a joint command that has existed for only about two years since 1991.

The published authority weight table

These weights drive the formal authority index. They encode constitutional and command authority, not real-world influence, which is why a presidential security chief scores lower here than an interior minister even though reporting suggests the reverse may hold in practice. Keeping formal and reported power in separate columns rather than averaging them is the whole reason the table is this shape. Disagree with a number and the argument is about the number, which is how it should be.

Source hierarchy

Sources are ranked, and the rank is visible on every citation.

• Tier 1. Tunisian official decrees and gazette entries, government portals, official military records.
• Tier 2. Institutional research and peer-reviewed academic work.
• Tier 3. Established international journalism.
• Tier 4. Reputable regional media.
• Tier 5. Leads only. Never sufficient on its own, and used here purely as an index to be corroborated elsewhere.

Every citation carries an archive link. Tunisian media and gazette URLs rot quickly, and a source-backed site whose sources return 404 in two years is worthless.

Machines propose, humans verify

The intended pipeline is discovery, extraction, entity resolution, evidence attachment and contradiction detection — all automatable — followed by human verification, which is not. Only the human step determines whether any of this is worth anything, so the review trail is stored per record and the coverage figure is published on the about page even though it is currently 27 of 719. Publishing the real number is the only way that commitment means anything.

Note also what the investigate page does not do: it does not ask a language model to answer your question. Answers are computed by traversing the graph, and every row carries its sources. An assisted research layer can sit on top of that later, but it must never be what produces the answer.

Corrections

The dataset is plain text under version control, so every factual change is a diff with a date. If something here is wrong, the useful correction is a primary source: a decree number, a gazette reference, an official biography, a corporate filing. The open questions are the current research agenda, and the fastest way to improve this map is to close one of them. The full graph is downloadable from the data page so the reasoning can be audited rather than trusted.

What this is not

Not a list of people secretly running Tunisia. Not a claim that wealth implies corruption. Not an adjudication of the prosecutions recorded here: a sentence is a documented fact, the allegations behind it and the criticism of the proceedings are contested, and the site stores those as separate competing framings with attribution rather than picking a winner.