How a Scenario Is Written

By the end you can read one scenario and explain what its successful execution does and does not establish.

The family’s question

The Garniers ask: “How do you check a rule before applying it to our file?” Predict whether a successful scenario proves every possible household correct.

Use the Moreau birth-grant card from Chapter 2 for a small example. Article L. 531-2 CSS, pinned December 2020 edition, uses ressources ne dépassent pas un plafond: resources do not exceed a ceiling. Scénario → worked test situation → a .lawtest file. The equality boundary is a useful first scenario because the phrase includes equality.

Facts, question, expectation

This is the complete equality-right test in the Moreau case package; the file’s language, package and namespace header is omitted:

test "PlafondEgal-droit" {
    given {
        context { legal_time @2025-07-15; decision_time @2025-07-15T12:00:00Z; knowledge_time @2025-07-15T12:00:00Z; timezone "Europe/Paris"; }
    assert "grossesse": fr.pf.paje::attente_naissance(entity_ref("urn:law:examples:fr:moreau:menage"), entity_ref("urn:law:examples:fr:moreau:enfant")) { origin case_input; }
    assert "base": fr.pf.paje::plafond_base_taux_plein(entity_ref("urn:law:examples:fr:moreau:menage"), 30000 EUR) { origin case_input; }
    assert "bmaf": fr.pf.paje::bmaf_applicable(entity_ref("urn:law:examples:fr:moreau:menage"), 400 EUR) { origin case_input; }
    assert "examen": fr.pf.paje::date_examen_situation(entity_ref("urn:law:examples:fr:moreau:menage"), entity_ref("urn:law:examples:fr:moreau:enfant"), @2025-07-01) { origin case_input; }
    assert "ressources": fr.pf.paje::ressources_annuelles(entity_ref("urn:law:examples:fr:moreau:menage"), 35841.00 EUR) { origin case_input; }
    }
    evaluate truth(fr.pf.paje::droit_prime_naissance(entity_ref("urn:law:examples:fr:moreau:menage"), entity_ref("urn:law:examples:fr:moreau:enfant")));
    expect truth_status == TRUE_ONLY;
    expect evaluation_status == COMPUTED;
}

Read it in three passes. given supplies the dated situation and the origin of each fact. evaluate asks about the right. TRUE_ONLY requires support for the proposition and no support for its opposite. COMPUTED means that evaluation completed; it does not certify legal correctness. The examination date is a supplied file fact here, not an invented reply to a judgment request.

Recorded on 11 October 2026 with law 0.1.12. These are artificial teaching inputs, as explained in Chapter 2.

The declared world

A test must also say which model it runs against. The manifest names the suite and the world, starting with the case package:

[[tests]]
family = "examples.fr.pf.paje.prime_naissance_moreau#cas"
regression = false
suites = ["tests/prime-naissance.lawtest"]
world = ["examples.fr.pf.paje.prime_naissance_moreau", "fr.pf.paje"]

The case package imports the benefit canon; it does not copy the grant rule. The saved law ask questions and the declared scenarios cover the same family files.

Change one thing

The neighbouring test adds a cent above the ceiling. Another removes annual resources. Both ask the same proposition. They check different risks: a boundary comparison and an incomplete file. A single successful equality test would leave both risks untested.

A second reading and a hand computation

The case’s hand computation records the arithmetic independently of the result: the ceiling comes from the base and statutory rate; the award comes from the BMAF and grant rate. A reviewer can recompute those steps with fractions, then inspect the answer’s applied rules and source anchors.

A useful method is a blind second reading: give another reader the pinned text and case facts before showing the expected answer. Compare their reading, including exclusions and dates, with the proposed test. Agreement on arithmetic is a different check from agreement on the legal construction.

The package’s check.py also checks the rule identifiers in the proof and the annual-resource premise in the why_not graph, then runs the declared tests. Testing the status alone would miss an answer reached on unintended grounds.

What this gives the fund

A reviewable example connects a legal phrase to its boundary, supplied facts and expected outcome. The fund can preserve that expectation through a change and explain why a new result should be investigated. Passing it does not establish legal correctness for every file or certify the entire benefit model.

Where Catala stands on this

Catala has a mature testing workflow: clerk test discovers scope tests and command-output tests, with backend checks available (Catala book, ch. 3.3). This is shared ground: both approaches preserve examples and expectations. Arxo adds built-in query statuses and a why_not graph whose missing premises the check can inspect. Catala 1.3.0 has no dedicated built-in form for the “not established, missing this premise” query answer; an application can define its own checks for absent inputs. For the fund, a scenario can preserve the reason a file is incomplete as well as the amount on a complete file.

Run the scenario

From a local checkout with law and Python 3.12+ installed, start at the repository root:

cd corpus/laws/fr/pf-paje/examples/prime-naissance-moreau
law test . --json
python3 check.py --law "$(command -v law)" --out "$(mktemp -d)"

These local commands use the case package from Chapter 2 and its installed checkout dependencies. The declared tests preserve expectations; the saved questions preserve full answer documents for inspection.

Exercise

1. Which block holds the family facts?
2. Which assertion covers the equality boundary?
3. Why inspect the applied rule in addition to the truth status?