In July 2025, Tunisia’s Ministry of Economy and Planning announced the launch of Tartib 2.0, a platform for programming, tracking and prioritising public investment projects. From the outside, it’s a date in a press release. From the inside, it was months of building a tool while part of its rules were being decided at the same time as we wrote the code.
Building while the rules are being decided
The project was run with Expertise France on behalf of the ministry. The overall logic was clear from the start: enter projects, check their admissibility, evaluate them, rank them. The details kept moving until the end: which criteria, what weight for each, which indicators for which sector, how the sectoral and central evaluations combine.
Waiting for a frozen specification would have meant never shipping. Three choices let us keep moving:
- Rules in data, not in code. Grids, criteria, indicators and weights are tables. When a weight changed in a meeting, it changed in the tool the next day, without a new release.
- Short, regular demos. Instead of long documents, we showed working screens in the ministry teams’ own vocabulary. A remark made in front of a screen is far more precise than a remark on a document.
- A clear order of priority. Project entry and evaluation first, because nothing else makes sense without them. Reports, dashboards and the project map afterwards.
Honestly, what helped me most was writing every decision down somewhere, with the date and the person who made it. Under pressure, you quickly forget why a criterion counts double.
Around a hundred users, several ministries
The platform brings together staff from different ministries, who mustn’t all see the same things. A ministry enters its projects and follows their evaluations; the central team sees everything and can adjust scores. Permissions were designed from the start rather than bolted on at the end, because fixing them later on real data would have been risky.
We also had to think about users who don’t spend their day in web applications: forms split into steps, labels in the administration’s own terms, and screens that say clearly what is left to fill in.
Deploying at the CNI
Production runs on the servers of the Centre National de l’Informatique, Tunisia’s national computing centre. That makes sense for a government platform, but it changes how you ship: you aren’t deploying to infrastructure you fully control, and every intervention goes through the centre’s procedures and teams.
So we prepared each release as a package to hand over rather than as continuous deployment:
- Docker images built and tested before each delivery;
- configuration entirely through environment variables, with nothing tied to our own machines;
- Liquibase migrations applied at startup, so the database always matches the application version;
- a written, step-by-step installation procedure, tested on a clean environment;
- a demo instance on a VPS, to validate each version with the ministry before presenting it to the CNI.
That demo instance was very useful: feedback came in on a version identical to the one going to production, and it avoided last-minute surprises.
What I’d do the same way
- Put business rules in data as soon as they might change.
- Show screens early, even unfinished ones.
- Write decisions down, with their date.
- Prepare the deployment alongside the application, not in the last week.
And what I’d do better: plan time for user training earlier. A platform delivered on time but poorly understood is no use to anyone, and training is often the part that gets planned last.