Every filtering policy removes more than it intends to. We simulated five stylised policies over 103,367,510 categorised domains from our register to measure exactly what each one blocks — and how much education, health, government, news, science and civic content goes with it.
Set one DNS address on your router and every device is filtered: CIPA categories blocked, a staff PIN for adult research, and reports for your records. Nothing to install.
Peer review is still ahead of it; the working paper is open on SSRN and free to read and cite.
Every filtering policy removes more than the law requires. This paper measures how much more, by simulating five stylised policies over WebFilteringDatabase’s full register of categorised domains and counting exactly which education, health, government, news, science and civic domains each one takes with it.
Blocking the categories the statute actually names costs almost nothing in legitimate content. The cost appears when a policy grows past that core.
“Protected” means a domain carrying an education, health, government, news, science or civic label — 16,250,357 of the 103,367,510 domains. The last column is over-blocking relative to the statutory core.
| Policy | Categories blocked | Domains blocked | Protected domains blocked | vs statutory core |
|---|---|---|---|---|
| Statutory core | 1 category (Adult) | 0.92% | 0.048% | — |
| Conservative | 9 categories | 2.41% | 0.11% | 2.3× |
| Typical district | 17 categories | 4.59% | 0.30% | 6.3× |
| Strict | 23 categories | 11.4% | 2.9% | 60× |
| Restrictive | 29 categories | 19.1% | 7.6% | 158× |
Each policy is computed under three rules for domains carrying several labels, which 9.0% of them do. The paper reports all three; the figures above use the middle rule.
Aggregate counts only, with no domain names: category prevalence by link-authority band, the full 57 × 57 co-occurrence matrix and pair statistics, policy simulations by rule, domain class and popularity, a cost-ordered blocking curve, the co-label context of sensitive categories, and shares by country and domain suffix. A codebook defines every term. Released under CC BY 4.0 and deposited in five repositories.
Repository DOIs are added here as each deposit is minted. Until then, cite the SSRN paper.
Samo Plibersek (2026). Where Over-Blocking Comes From: Simulating CIPA School Filtering Policies on 103 Million Categorised Domains. SSRN Working Paper. https://doi.org/10.2139/ssrn.7638439
Every figure in the paper is an aggregate; no individual domain is named. The text and the dataset are CC BY 4.0, so both can be quoted, charted and redistributed with attribution. For the methodology, a custom cut of the data or comment, get in touch. Our 49,000-domain research dataset is a free sample of the register the paper is built on.