Guide · DeepTunisia

The one rule: an inference is never a fact

Every claim in the dataset is filed under one of four epistemic bases. They never render identically, because they are not the same kind of fact:

Documented
An official record, decree, gazette entry or primary document states this directly.
Reported
One or more credible publications report it, with attribution. Not independently checked against a primary record.
Inferred
Nobody states this directly. It is reasoned from documented structure, and carries both the reasoning and what would falsify it.
Unsubstantiated
A claim that circulates without reliable evidence. Recorded so it can be examined, never presented as evidence.

The record and the argument

The top bar splits the interface into two bubbles. The distinction is the project's whole claim, stated before you open anything:

Graph — the sourced record. Positions, relationships, events, and the evidence behind them. Everything here carries a basis and a citation.
Agora — everything the record is not: discussion, proposed changes, reports, and the news feed. No basis, no source grade, no verification. It is deliberately rendered in a register no sourced claim ever wears.

Crossing the line is cheap and safe. The two dots next to the names are load-bearing: filled means "this is the record", hollow means "this is not".

Time is a dimension here

The dock at the bottom is the timeline. Scrub it and every view projects onto that instant — the network shows who was live in 1994, the Chronicle shows who held which post. Two things to know:

The density curve behind the track shows how much of the record exists at each date. It thins before 1987 not because less happened, but because the archive is thinner. That is the project's honesty signal: the map shows the record, not the past.
A position is an interval, not a line in a table. "X held post Y" and "between these dates" are separate claims with separate standing — one can be documented while the other is inferred.

What "verified" means

The project's architecture is machines propose, humans verify. At this stage most records have not been through a second pair of eyes — the real number is published on the about page rather than hidden. A record flagged "awaiting a primary source" is not an accusation; it is a to-do list the project shows you.

Which view answers your question

View	What it answers	When to reach for it
Chronicle	Who held which post, when, drawn to scale. Bar length is tenure, so continuity is visible without reading.	Continuity questions: whether a post survived a rupture, and who held it for how long.
Now	Who holds which office today — split by how recently each was verified, and measured against that office's historical median tenure.	“Who is in post right now”, and which claims about the present rest on fresh verification.
Network	Who connects to whom. Fixed lane per layer, so cross-layer bridges are literally the edges that cross lanes.	Path questions: how two people or institutions are connected, and through which layer of power.
Atlas	Everything at once: structure per era as small multiples, a survival matrix of people against eras, an institution-crossing matrix, and median tenure per office.	Structural questions: which people and institutions survive regime changes, and which do not.
World	Tunisia's place among states and international organisations, and its strategic assets.	Foreign-power questions: what Tunisia depends on, and which states and bodies are in the record.
Rankings	Five separate structural indices with reader-adjustable weights. Deliberately no single power score.	Comparisons: who ranks where on brokerage, continuity, influence and the rest — with the weights in your hands.
Investigate	Structured queries over the graph. Computed by traversal, never generated by a language model.	Specific questions with an answer in the data: who crossed paths, who held which posts, who was connected to whom.
Map	Tunisian governorates and strategic assets with sourced coordinates.	Geographic questions: where things are, and what the record knows about a place.
Evidence	Hypotheses with falsifiers, open questions, unresolved source contradictions, every citation.	Judging the project's claims: what the data supports, what it refuses, and what is still open.
Method	How the dataset is built, and what it refuses to assert.	Trust questions: how a number was derived, and which claim types the project declines to make.
Corrections	Every change to the dataset, generated from version control.	Accountability: what was corrected, when, and by which commit.
Data	The full graph, downloadable under CC BY 4.0.	Reuse: take the dataset, the CSVs, or the reasoning in the editorial queue.
About	What DeepTunisia is, how it is governed, who funds it, and how to correct it.	Governance and provenance questions: who is behind this, and under what rules.
Agora	Discussion, proposed changes and reports — everything the record is not.	Contributing: argue about the record, propose a change, or report a problem.
Feed	Headlines from other outlets, matched to the graph — never merged into it.	The news: what is being published today, clearly separated from the sourced record.

If something 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 closing one is the fastest way to improve this map. The open data page has the full graph, so you can check the reasoning rather than take it on trust.