# Method, and what it cannot do

Everything below is measured from the corpus as it stands, not estimated. The limits are stated because a reader deciding whether to rely on a page needs them more than the successes.

Canonical: https://unisonlegal.in/methodology

## Reading the page

Every PDF here ships with a usable text layer, so extraction is a text read rather than OCR. That text is nonetheless imperfect: these are scans of printed volumes, some of them seventy years old, and the layer carries the scanner's mistakes. A title reading "STA TE OF PUNJAB" is one of them. The scanned page is on every document page, and it is the authority — the text is what makes the page findable, not what you should quote.

Paragraphs, headings, quotations and printed-page breaks are reconstructed from a flat text layer. The reconstruction is checked against the original on every release: the rule is that the structured text must be the original text with whitespace changed and nothing else, so the reading view cannot silently edit a judgment.

## The citation graph

- **467,941** citations extracted from the judgments' own text
- **109,316** of them (23%) resolve to a judgment this corpus holds
- **25,002** judgments are cited by at least one other

The unresolved majority is not a failure of the extraction. It is that a citation can only point at something we hold, and most Indian judgments are cited by reporters this corpus does not have:

- **SCC** — 243,557 mentions, 0 resolved — not held, so nothing to point at
- **SCR** — 138,199 mentions, 99,530 resolved
- **AIR** — 62,562 mentions, 0 resolved — not held, so nothing to point at
- **ILRA** — 19,274 mentions, 8,977 resolved
- **INSC** — 4,349 mentions, 809 resolved

## Provisions

- **453,164** references to a statutory provision, extracted the same way
- **295,037** (65%) resolve to a provision this corpus holds

Only the acts listed under [sources](https://unisonlegal.in/sources) are loaded, so a reference to the Evidence Act or the Limitation Act is recorded with its name and number and linked to nothing. Where that is so, the page says so rather than offering a dead link.

## What is not here

- No volume of SCC or AIR, and no judgment known only by one
- No High Court other than Allahabad
- Nothing decided after the last volume ingested
- **No machine-written text.** No summary, headnote or description on this site is generated; where a machine-written summary ever appears it is labelled as such and is never presented as the court's words.

Found something wrong? [Tell us](https://unisonlegal.in/corrections).