Our standards for accuracy and transparency.

Case selection

Case Law Digest covers notable Indian court judgments published in the indexed corpus. We prioritize:

• Landmark decisions — Cases that establish or clarify important legal principles.
• Recent rulings — Freshly decided cases that reflect current judicial thinking.
• High-impact matters — Cases with broad implications for legal practice.
• Diverse coverage — Across courts, legal areas, and jurisdictions from the corpus.

Writing standards

• Accurate case details — Case name, court, date, and bench are verified against the judgment text and the case record.
• Clear legal analysis — Issues, holdings, and reasoning are explained in plain language, with quoted passages reproduced verbatim rather than paraphrased where the holding matters.
• Attribution — Every article links to the full judgment text in the corpus and, where available, to the official court record.
• Author & timestamp — Each digest entry carries its author and last-updated date, so readers know who wrote it and when it was last verified.
• AI transparency — AI-generated summaries are clearly distinguished from original judgment text. The model writes only from what was retrieved; it never recalls cases from memory.

Verification checks before publishing

Case details and generated content are checked against the judgment text and the case record before a note is generated, and again before it is published:

• Case name — must be a clean cause title found in the judgment, not an OCR-garbled or incomplete one.
• Court — must resolve to a known court and match the court named in the judgment's own caption.
• Decision date — must be consistent with the case record and with any date stated in the judgment itself.
• Bench — judge names must be found in the judgment text; a name that isn't is dropped, not shown.
• Citation and case number — must be printed in the judgment's own caption.
• Quotations — must be verbatim; a paraphrase presented inside quotation marks is rejected.
• Precedent treatment — a label other than "referred to" (relied upon, followed, distinguished, overruled) needs supporting language near the citation, and a High Court note can never say a judgment "overruled" the Supreme Court.
• Acts, sections and notifications — must be named in the judgment text.

A failing field is simply dropped from the note rather than shown unverified. A wrong court, decision date, or case name is treated as more serious: it blocks the note from being published at all. Existing, already-published notes are re-checked the same way, and any field that no longer verifies is hidden rather than left to go stale.

What we don't publish

• We don't publish content based on unverified or fabricated judgments.
• We don't present AI-generated analysis as legal advice.
• We don't omit material facts that would change the reader's understanding of a case.
• We don't overstate corpus quality: OCR-derived text is treated as OCR-derived, and stated limitations keep pace with the corpus.

Correction policy

Errors happen — particularly in OCR-derived text. When a reader can show us a mistake, we fix it fast and visibly:

• How to report — Email [email protected] with the page URL, what is wrong, and your source (e.g., the official judgment record). Corrections are also accepted via the contact form.
• What we fix — Verified factual errors in case names, court, dates, benches, quoted holdings, statistics, and broken citation links.
• How we fix it — Material corrections are applied to the corrected content and the page is re-verified against the corpus or the official record. The article's last-updated date is advanced with the correction.
• Transparency — For significant corrections that alter a reader's understanding, we note the correction on the article itself rather than silently rewriting history.
• No takedowns for critique — We correct errors; we do not remove content because it is critical of a party, advocate, or institution.
• OCR text is flagged, not silently 'fixed' — Garbled source text in the underlying corpus is corrected through the ingestion pipeline; we do not silently alter judgment text to match a desired reading.