Administrative Law, Water Access & the Limits of Judicial Inclusion — Global Water Law Judicial Decisions Dataset (v0.3.0)
other_water residual reduced 89.2% → 8.3%. NWR filter precision 99.79% (207-case gold-standard, pop.-weighted). Cohen's κ = 0.568 three-label; binary WATER/NOT_WATER κ = 0.932 (n = 59, excludes UNCERTAIN).
Read Audit ↗
Validation Folder ↗
| Category | Count | % | Distribution |
|---|
| Category | Count | % | Distribution |
|---|
indigenous_water — an almost certain undercount. First Nations water insecurity generates substantial political and administrative contestation, but Title-only CanLII records rarely contain the substantive language needed for coding. The Legal Data Hunter semantic search component was designed partly to surface these cases; the near-zero count signals a limitation of the available data, not the absence of legal activity.| Category | Count | % | Distribution |
|---|
| Dimension | Brazil | Netherlands | Canada |
|---|---|---|---|
| Total decisions | 11,724 | 68,654 | 3,218 |
| Data period | 2016–2026 + 1997–2015 (TJSP) | 2016–2025 | 2016–2025 |
| Genuine water decisions | 11,686 (99.7%) | 989 (1.4%) | 421 (13.1%) |
| False positives (v0.3.0) | 38 (0.3%) | 67,665 (98.6%) | 2,797 (86.9%) residual |
| Dominant dispute | Tariff (48.5%) | Flood protection (21.4% of water) | Fisheries/riparian (8.4% of water) |
| Connection refusal | 10.9% (1,275) | 0.012% (8) | 0% |
| Informal settlement | 0.75% (88) · 0 user wins | 0% | 0% |
| HR language | 1.9% (226) · 0 in informal settlement | 0% | 0% |
| Win/loss coded | 4,481 outcomes coded | Not available | Not available |
| User win rate | 31.2% (of coded outcomes) | — | — |
| Utility win rate | 26.0% | — | — |
| Pre-litigation absorption | Low | High (Awb/Geschillencommissie/Ombudsman) | Medium (fragmented) |
| Legal Last Mile model | Rights-declared, admin-deficient | Pre-litigation absorption | Fragmented opacity |
| Courts covered | 8 state courts (of 27) | All 11 Rechtbanken + appellate | CanLII databases (multi-province) |
From: The Legal Last Mile: Administrative Law, Water Access, and the Limits of Judicial Inclusion — Preliminary research, 2026.
Administrative law functions as a gatekeeper that determines whether physical access to water infrastructure translates into legally enforceable entitlement. Bureaucratic procedures — documentation requirements, property registration, formal address systems — filter who can assert rights before courts are ever reached.
Populations excluded from formal administrative systems become invisible in the judicial record. Brazil's informal settlement cases (<1% of decisions) are not evidence of fewer legal problems — they are evidence of exclusion from the formal system. The absence itself is the finding.
A legal system can be simultaneously rights-saturated and access-deficient. Brazil has one of the world's strongest constitutional water rights frameworks, yet HR language appears in only 1.9% of decisions — and the populations most needing rights-based protection are least represented as plaintiffs.
Where administrative systems are robust and universal, courts address systemic water governance (flood protection, spatial planning, waterboard regulation) rather than household access. The Dutch model absorbs service disputes before courts through the Awb complaints procedure and Geschillencommissie.
Where jurisdiction is fragmented between federal and provincial levels with no universal service obligation, water disputes are processed through opaque administrative procedures with variable accessibility. The judicial footprint reflects resource governance, not household access — but the absence of access disputes may reflect invisibility, not resolution.
The jurimetric distribution itself is evidence: 48% tariff vs. 10% connection refusal in Brazil; <0.1% connection refusal in NL; 0% in Canada. These ratios confirm the comparative MDSD analysis — the Legal Last Mile operates differently in each system, but the pattern of who doesn't appear is consistent across all three.
The three cases were selected on the basis of maximum variation on the key independent variable (administrative law model and pre-litigation architecture) while sharing the dependent variable of interest (formal constitutional or statutory recognition of water access rights):
| Dimension | Brazil | Netherlands | Canada |
|---|---|---|---|
| Admin law tradition | Civil law, fragmented | Civil law, comprehensive Awb | Common law, no general code |
| Water rights basis | Constitutional (Art. 6, 225) | Statutory (Drinkwaterbesluit) | Statutory (provincial + Safe Drinking Water Act) |
| Universal service obligation | Partial (urban utilities) | Full (KWR/Vewin) | None (federal) |
| Pre-litigation routes | Weak (PROCON, ANS) | Strong (bezwaar, Geschillencommissie, Ombudsman) | Variable by province |
| Urbanisation / informality | High (15–20% informal) | Very low | Low (urban) |
| Dataset coverage | Partial (8 of 27 courts) | Near-complete (all 11 districts) | Multi-provincial via CanLII |
Brazil (8 courts): Custom scrapers for ESAJ POST (TJAC, TJRR, TJSC, TJSP), Elasticsearch REST (TJDFT), ASP.NET WebForms (TJRJ), Rails GET (TJPI), PHP+Solr GET (TJTO). 19 of 27 courts blocked by CAPTCHA, authentication, or defunct APIs.
Netherlands: Rechtspraak.nl Open Data XML API covering all 11 Rechtbanken + Raad van State + CBb + GHARL.
Canada: CanLII REST API (keyword search + 113 extra databases) supplemented by Legal Data Hunter semantic search across the full CanLII corpus (94,502+ documents).
AI assistance: Claude (Anthropic) assisted with scraper development, data pipeline, coding engine design, and report generation. All research design and interpretation: Claudio Klaus Junior.
21 governance categories in Portuguese, English, Dutch, and French (expanded from 12 in v1.0). Applied via jurimetric_coding.py (GitHub) to decision summaries and titles.
New in v0.3.0: Broad water-core vocabulary filter applied to ALL decisions (replaces narrow NL immigration filter). Three Brazil rescue patterns (tariff, connection, pipe). Result: genuine other_water reduced from 89.2% → 8.3%; 67,703 decisions reclassified as not_water_related. Validation infrastructure added: residual audit, kappa calculator, XLM-RoBERTa transformer classifier, second-coder protocol.
7 variables per decision: hr_language · sust_language · governance_cat · win_loss · mp_involvement · indigenous_water · public_interest
Dataset citation:
Klaus, C (2026) Global Water Law Judicial Decisions Dataset (v1.0) Zenodo <https://doi.org/10.5281/zenodo.19836413> accessed 3 May 2026.
Also archived at:
Harvard Dataverse: doi:10.7910/DVN/C9PEFS ·
DANS SSH: doi:10.17026/SS/RVDBUF ·
OSF: osf.io/admrq
Scholarly inspiration: This research is inspired by the work of Professor LaDawn Haglund (water governance, judicialization of water and sanitation rights in São Paulo, Brazil). See: Haglund (2014) Water Policy; Haglund (2019) in Economic and Social Rights in a Neoliberal World (Cambridge); Haglund (2019) Sustainability 11(19).