Analysis, Income Tax

Anatomy of an Indian Tax Case

A judgment, ideally, is a prime example of how one should structure any piece of text, especially legal texts. It is supposed to be coherent and concise, and importantly, follow a trail that takes the reader smoothly from one point to the other. While a judgment may simply seem like a collection of legalese to a lay person, it is possible to delineate certain elements within it. Understanding these elements, and their characteristics, can be of immense help to a reader, making the text much easier to read.

At Riverus, we apply machine learning to the cases to highlight and extract information in a manner that makes understanding and reading a case much simpler and faster. A quick analysis of the judgments available in Riverus’ database revealed that a legal judgment appeared to follow this universal unsaid format:

1.Opening– A quick introduction to the case, details of where it was originally filed and other pieces of information regarding the case history and parties.

2(a). Issues – The matter at hand, it’s points of interest and the foundation stone of a case. Issues are usually a dispute over a point of law or a fact. An issue determines the answer that a judgment is likely to give – the court usually frames an issue question and then proceeds to answer it. For instance –

“The basic question which was put forth for our consideration is whether the arrangement which has been worked out between the assessee company and the distributor (Agency) claimed by the Income Tax Department are covered under the provisions of Sections 194H and /or 194 J of the Income Tax Act.”

2(b). Grounds –Grounds are the reasons that a party gives for filing an appeal or a case. It is usually framed to show that a violation of law had happened which has adversely affected the party and hence they have approached the court for redress. For example –

In ITA No 921/JP/2011 for A.Y. 2004–05, Revenue has filed appeal on the following grounds: 1. That the ld. CIT (A) has erred in law as well as on the facts and circumstances of the case in deleting the disallowance the claim of exemption u/s. 10(23c)(iiiad) of the IT Act, 1961.”

2(c). Prayer – Prayer is the relief sought by a party from the court.

Together, grounds, issues, and prayers can be termed as Submissions before the court.

  1. Facts – Facts are what transpired in a case. Arguments and reasoning of a case rely on the facts. While both parties are usually in agreement about the facts, if there is a dispute it gives rise to a question of fact, which the court must resolve.
  2. Arguments –  Contentions put forth by either party to support or defend their stance are arguments. Counsel usually refer to the law and precedents in supporting their arguments. An example of an argument would be –

“It was submitted by ld. AR of assessee that the impugned order passed by CIT(A) is ex-parte qua the assessee.”

  1. Holdings – The verdict, inclusive of the directive part (if any) which indicates the action to be taken by the parties.

“We accordingly allow the appeal and answer the question referred by the Tribunal in favour of the assessee and against the Revenue. The Revenue will pay to the assessee costs throughout.”

  1. Outcome – A condensed form of holdings as a verdict, to sum up the case.

In the result, both the appeals of the assessee for the assessment years 2008–09 and 2009–10 are allowed in part and whereas appeal of the department for assessment year 2009–10 is rejected.”

Ideally, in cases that address more than one issue of law, points 2 through 5 should be repeated till all issues are dealt with.


We used text classification models to predict most of the aforementioned elements in the judgments on our database. This helps us not only in highlighting the essential elements of a case but also in making fairly accurate predictions as to where a particular element might occur in a case.Knowing the approximate location of an element even before knowing the case can substantially increase a person’s judgment reading speed.

Assuming the start of the case is 0 and the end is 1, the approximate locations of these elements as predicted by our model is as follows:

  • Submissions: 0.29
  • Arguments: 0.46
  • Holdings: 0.76
  • Outcome: 0.82

Plotted to a graph, the results look something like this :


On the x axis, we have various elements of a case. The y axis has been inverted for easy understanding of the chart. On the y axis we have location of an element in a case, i.e., the extent to which one would have to traverse a case to reach a particular line in the text (0.2 = 20% into the case), starting from the first line.

We were also able to glean certain other interesting insights using this prediction model. For instance, we found that Submissions, Holding and Outcomewere clearly located at certain specific points in the case, whereas Arguments and Reasons are scattered almost evenly throughout the case.

On comparison, it also emerged that High Court and Supreme Court judgments tend to be the most structured and coherent. They start by clearly describing an issue, and then discussing it in detail before proceeding with another issue. ITAT judgments were more scattershot in terms of structuring.

Interestingly enough, we were also able to map out the relative positioning of the elements over the years.


There is a discernible trend which shows that judgments have moved from being fairly structured in terms of elements to being much more loose with the structuring as time progressed. In the above graph, the bulbousness of the bar represents density of the elements at a particular place in the judgments. We can see that the progression goes from being heavily delineated (more bulbous) to a more diffused concentration (less bulbous), showing that the elements are now more spread out in terms of positioning in judgments. 

While many say that judgment writing is an art, more often than not the exercise of reading a judgment also becomes one. With greater access to relevant data it is possible to gain unique insights into the anatomy of a judgment and consequently make the process of extracting data from it that much more uncomplicated. Using machine learning, we have automated the process of identification of issues, arguments and holding in cases. These portions are highlighted to enhance readability, and armed with the knowledge of how courts tend to structure judgments, reading a case has never been more efficient than it is now at Riverus.

(Written by Varun Chitale, Data Scientist @ Riverus)

Also Read: Unreported judgments: Why is it not required for tax cases?

Share your thoughts using the comment section below

%d bloggers like this: