feat: refonte dynamique de la page Perspectives

[m4xim1nus]

Summary

• Remplace le contenu statique de la page Perspectives par des données temps réel issues de PostgreSQL
• KPIs dynamiques : collectivités (total vs soumises à la loi), marchés publics, subventions, couverture, scores A/B
• 3 sections de graphiques Recharts : comparaison par type de collectivité, évolution des scores, volumes
• Filtre les communes < 3 500 habitants (non soumises à la loi pour une République numérique)
• Encart sur le classement européen de la France en open data

TODO (prochain commit)

• La section "Bons et mauvais élèves" (ScoreByTypeChart) apparaît vide — à investiguer
• Filtrer les montants aberrants de subventions

Test plan

• Vérifier que la page /perspectives charge correctement avec les données de la BDD
• Vérifier les KPIs (nombres exacts, couverture < 100%, scores A/B cohérents)
• Tester les filtres par type de collectivité sur le graphique d'évolution des scores
• Vérifier que les graphiques de volume s'arrêtent à l'année courante
• Tester le responsive (mobile / desktop)

🤖 Generated with Claude Code

[github-actions]

Preview is deploying...

Please wait while the preview is deploying on Clever Cloud...

ℹ️ Name	🔗 Infos & links
🌱 Deploying commit	d069141
👁️ Preview	Generating preview app...

[cyrilledaily]

Great work @m4xim1nus — this is a big upgrade from the static page. The dynamic KPIs, the Recharts charts, and the editorial rewrite are all well-structured. A few things to address before merging:

---

🔴 Critical: Outlier amounts will show absurd numbers

The KPI query sums montant_du_marche_public_par_titulaire and montant without any outlier filtering. I checked production data:

	Raw total	Clean total	Ratio
Marchés publics	3 213 T€	791 Md€	99.97% from outliers
Subventions	8.78e18 € (!!!)	277 Md€	essentially all outliers

The page would display something like "3 213 T€ de marchés publics" which is nonsensical.

Fix for marchés: filter on montant_aberrant = false (or IS NULL) in mp_stats and mp_yearly:

WHERE (montant_aberrant IS NULL OR montant_aberrant = false)

Fix for subventions: there's no montant_aberrant column on subventions yet, so add a threshold filter:

WHERE montant <= 1000000000  -- 1 Md€

This is the same approach as our insights dashboard — 780 subventions above 1B€ account for 99.997% of the total.

---

🟡 LATEST_SCORE_YEAR = 2024 is hardcoded

This constant is used for the KPIs and passed as year={2024} to ScoreByTypeChart. It will silently become stale when 2025 data is loaded. Consider fetching it dynamically:

SELECT MAX(annee) FROM bareme WHERE annee <= EXTRACT(YEAR FROM NOW())

---

🟡 No caching — 4 heavy queries on every page load

The KPIs query alone takes ~1.2s on production. With all 4 queries in Promise.all, the page load is bound by the slowest one. Since this data changes only after ETL runs (not per-request), consider adding:

export const revalidate = 3600; // cache for 1 hour

at the top of page.tsx. This is a standard Next.js pattern for ISR.

---

🟡 SCORE_COLORS and TYPE_LABELS are duplicated

SCORE_COLORS is defined identically in both ScoreByTypeChart.tsx and ScoreDistributionChart.tsx. TYPE_LABELS is partially duplicated too. Extract these to a shared constants file (e.g., perspectives/constants.ts).

---

🟢 Minor: montant_du_marche_public_par_titulaire inflates totals

Using the per-titulaire amount means multi-titulaire marchés are summed multiple times. For example, a 1M€ marché with 3 titulaires (each getting 333K) counts correctly. But if you wanted to show the total value of contracts, you'd use montant_du_marche_public grouped by id_mp first. Worth a comment at minimum to explain the choice.

---

🟢 Minor: Text changes are good

Replacing "seules 10% le font" (old Anticor sample estimate) with "la grande majorité ne le fait pas" is more accurate and defensible. The wording in ProjectDescription.tsx and WhyChallenge.tsx is good.

---

Overall this is a very nice piece of work — the page will be much more compelling with real data. The outlier filtering is the only blocker IMO; the rest can be follow-ups.