Paragraph-level treatment, explained.

Take a famous example. Donoghue v Stevenson is “good law”. It has been followed in 847 reported Australian decisions. A case-level citator will tell you that, and stop. What it will not tell you is which paragraph of Donoghue each of those 847 cases actually relied on.

Donoghue is a 1932 House of Lords decision with three separate judgments running across roughly 70 paragraphs of the All England Reports. Lord Atkin’s neighbour principle sits at paragraph 580. Lord Macmillan’s concurring reasoning sits at paragraph 599. Lord Thankerton’s concurrence sits at paragraph 619. Some later cases follow Lord Atkin specifically. Some apply the neighbour principle in its broader Macmillan formulation. Some distinguish Donoghue on the very ground Lord Thankerton emphasised. To a case-level citator, all 847 are “followed”. To a paragraph-level citator, they are 847 different stories.

Why the distinction matters in litigation.

If you are running a negligence argument and your opponent is going to rely on Donoghue, the question is not whether Donoghue is good law. The question is whether the paragraph you intend to lean on is the paragraph that has been most followed since, or the paragraph that has been quietly distinguished by every appellate court for thirty years. Case-level citators cannot answer that. Paragraph-level citators can.

In appellate work, the distinction is even sharper. When you build a distinguishing list, you are not trying to show the case is bad law. You are trying to show the specific reasoning the other side wants the court to adopt has been confined to its facts since. That confinement happens at the paragraph level. A citator that abstracts away from paragraphs cannot surface it.

Why the distinction matters in academic writing.

Doctoral and postgraduate writing in the common-law tradition lives or dies on the granularity of citation. A paper arguing that Lord Atkin’s neighbour principle has narrowed since the 1980s needs to be able to point to specific paragraphs in specific judgments where the narrowing happened, and to specific paragraphs in specific later judgments that adopted the narrower formulation. A case-level citator gives you a wall of citations. A paragraph-level citator gives you a thesis.

Honest acknowledgement: most undergraduate writing does not need this level of granularity. A 3,000-word problem question at a third-year level can be done well with case-level treatment alone. The difference shows up in research-heavy writing, postgraduate seminars, and the long tail of advisory work that practitioners do for high-stakes clients.

Why the distinction is hard to build.

Paragraph-level treatment is not a feature you toggle on. It is an architecture decision that runs through the whole stack. You need a corpus that has been paragraphed accurately (which not every report series does for older cases). You need a citation extractor that captures pinpoints, not just case names. You need a treatment classifier that reads the surrounding paragraph in the citing case and decides whether the citing court followed, applied, distinguished, or overruled the specific paragraph cited. You need to store all of that in a graph that can be queried at paragraph granularity without falling over. And you need a UI that surfaces it.

The traditional Australian citators were built when editorial labour was the cheapest part of a research product, so the case-level abstraction was an acceptable simplification. Capable reasoning models have made paragraph-level extraction tractable at a price that lets a research-individual product carry it. That economic shift is the load-bearing reason CaseSharp can offer paragraph-level treatment at a budget the previous generation of tools never could.

What it looks like in practice.

Open the CaseSharp authority page for Donoghue v Stevenson. The treatment badge at the top says “Good law, followed in 847 decisions”. Click into the citator tab. You see 847 rows. Each row is the citing case, the AGLC4 citation, the treatment classification, AND the paragraph of Donoghue that the citing court pointed to. Filter to “followed at [580]” and you get the cases that actually adopted Lord Atkin’s neighbour principle. Filter to “distinguished at [619]” and you get the cases that have backed away from Lord Thankerton’s reasoning over the years. The distinguishing list for an appellate brief is now a two-click operation, not a week of reading.

Every authority page in CaseSharp is built this way. Treatment by paragraph. Citation graph by paragraph. Legislation linkage by section and subsection. The surface looks like any other research tool, but the underlying data is structurally different.

What this commits us to.

The paragraph-level commitment has implications. It means we need a paragraphed corpus, which is hard for pre-2000 judgments and harder still for older state cases. It means our treatment classifier has to be careful about distinguishing a paragraph-specific follow from a general endorsement. It means our authority pages will sometimes show empty paragraph cells where the citing court did not pinpoint, and we have to be honest about that rather than guess.

Honesty about what the data does and does not show is the price of the methodology. Every authority page on CaseSharp will tell you, on the same surface, what is paragraph-pinpointed and what is only case-level cited. Where a citing court did not pinpoint, we say so explicitly. The user gets to decide whether the citing case is still useful evidence for their argument.

Sources.