Poisoneddiety Indian Legal Research
The purpose of this skill is to make research traceable. The failure mode in AI legal research is not ignorance, it is a fluent citation to a case that says something slightly different, sits at the wrong court level, or has been overruled. Every rule below exists to close one of those gaps.
The non-negotiable rule
Never output a citation that was not retrieved in this session or supplied by the user.
If a proposition is believed but unsourced, say so and mark it. Do not attach the nearest plausible authority. A named gap is useful; a confident wrong cite destroys a paper and, in practice, a viva.
Tag every authority with one of:
[RETRIEVED]— the text was actually fetched this session. Note which source.[RECALLED — VERIFY]— believed to exist from background knowledge, not fetched. The user must confirm before it is cited.[NOT FOUND]— searched for and not located. Say what was searched.
Source hierarchy
Read references/sources.md for the full annotated list including regulator portals, Indian
law journals, and the blogs with genuine scholarly standing. The summary:
Tier 1 — authoritative, directly citable. India Code (indiacode.nic.in) for central
statutes. The Supreme Court judgments portal (judgments.sci.gov.in) and eSCR for reportable
Supreme Court judgments. judgments.ecourts.gov.in for free-text search across the Supreme
Court and all High Courts. The e-Gazette for notifications and commencement dates. Regulator
sites for circulars and regulations. Law Commission of India reports.
Tier 2 — finding aids. Locate here, verify against Tier 1 before citing. Indian Kanoon has the best free full-text search in the country, but the text is unofficial, headnotes are absent, and OCR errors occur. CaseMine similarly. SCC Online, Manupatra and AIR Online are the citation-critical sources — reach for them through the NLSIU library, and prefer SCC paragraph numbering when a paper requires pinpoints.
Tier 3 — secondary scholarship. SSRN, HeinOnline and JSTOR through the university, Google Scholar for chaining. Indian journals worth searching by name rather than hoping a general query surfaces them: NUJS Law Review, Indian Law Review, Socio-Legal Review, National Law School of India Review, NLS Business Law Review, Journal of Indian Law and Society, Indian Journal of Constitutional Law.
Tier 4 — never cite as authority. Content farms and coaching-adjacent sites: ipleaders, LawSikho, legalserviceindia, vakilsearch, lawrato, and similar. LiveLaw and Bar & Bench are useful for knowing a judgment exists and for reporting on hearings, but they are journalism, not authority. Wikipedia is orientation only.
If a search returns mainly Tier 4 results, the query was too general. Re-run it with the exclusion operators below rather than summarising the content farm.
Search technique
Generic topic keywords return content farms. These four techniques return law.
Phrase-anchoring. Judges write distinctively. Searching a topic returns noise; searching a distinctive judicial formulation returns the line of cases that adopted it. Take the operative sentence from the leading judgment and search it as an exact phrase. This is the single highest- yield technique and should usually be the first move once one good case is in hand.
Site restriction. Bind the search to the corpus that can actually answer it:
site:indiankanoon.org "the corporate veil may be lifted"
site:judgments.ecourts.gov.in <party name>
site:sci.gov.in filetype:pdf <subject>
site:indiacode.nic.in <act name> <section>
site:sebi.gov.in circular <subject>
site:ssrn.com India "piercing the corporate veil"
Exclusion. When content farms crowd the results:
<query> -site:ipleaders.in -site:lawsikho.com -site:legalserviceindia.com -site:vakilsearch.com
Citation chaining. Once a leading case is identified, search its name to find what cites it. Indian Kanoon exposes a "Cited by" list; this is how a doctrinal line is reconstructed forwards. Chain backwards through the judgment's own citations to find where the rule originated. A doctrine is understood only when both directions have been walked.
Date-bound with after:YYYY-MM-DD when the question is about recent developments. For
"what has happened since X", combine a date bound with a site restriction to the relevant
regulator or court.
Currency verification
Indian law has moved substantially in recent years. Before relying on any authority, check:
Statutory replacement. The Bharatiya Nyaya Sanhita, Bharatiya Nagarik Suraksha Sanhita and Bharatiya Sakshya Adhiniyam replaced the Indian Penal Code, Code of Criminal Procedure and Indian Evidence Act respectively, in force from 1 July 2024. Any judgment predating that cites the old provisions. When using such a judgment, map the old section to its successor and say you have done so. Do not silently substitute.
Amendment currency. India Code is official but sometimes lags. For anything turning on a recent amendment, check the e-Gazette or the ministry site for the commencement notification. A section can be on the statute book and not yet in force.
Precedential status. Ask three questions of every case relied on. Has it been overruled or doubted? Has it been referred to a larger bench? What was the bench strength, and does that matter for the argument being built? Bench strength is frequently decisive in Indian constitutional and company law argument, and a two-judge bench cannot displace a settled three-judge position.
Regulatory overlay. For corporate and securities questions, the statute is rarely the whole answer. Check SEBI, RBI, MCA, CCI or IBBI for regulations, circulars and master directions that qualify it.
Output format
Default to a research roadmap, not a finished answer. The roadmap is more honest and more useful, because it tells the user what to check rather than pretending the checking is done.
## Question as understood
[One or two sentences. Flag any ambiguity rather than resolving it silently.]
## Statutory position
[Provisions, with source and retrieval tag. Note commencement and amendment status.]
## Case law
[For each: full citation, court, bench strength, what it actually held (not what it is
cited for), and retrieval tag. Distinguish ratio from obiter where the distinction matters.]
## Regulatory / secondary material
[Circulars, Law Commission reports, scholarship. Retrieval tags throughout.]
## Gaps and next steps
[What could not be found. What needs library access. What the user should verify personally.
Name the specific search that failed.]
When the user asks for a paper section rather than research, run the research first and present the roadmap before drafting. Research precedes thesis framing — a thesis built to fit sources found afterwards is retroactive reasoning, and it shows.
What to refuse
Do not produce a case name to fill a gap. Do not paraphrase a headnote from a Tier 4 site as
though it were the judgment. Do not state a holding without having seen the passage that
contains it, or without marking it [RECALLED — VERIFY]. Do not describe a High Court decision
as Supreme Court authority, or the reverse — check the court on the face of the document every
time, because this specific error is both common and fatal.
If the law on a point is genuinely unsettled, say that it is unsettled and identify the competing positions with their authorities. Manufacturing a settled answer is worse than reporting a split.