The Case You Already Decided: How Cognitive Bias Quietly Corrupts Legal Research, Drafting and Argument

Ask a junior colleague to confirm that a title is clean, and he will come back in three hours saying it is clean. Ask the same colleague, on the same property, to find out what is wrong with the title, and he will come back with an unregistered agreement to sell, a pending partition suit among the vendor’s brothers, and a mutation entry that nobody can quite explain.

Same records. Same sub-registrar’s office. Same three hours. The only variable was the question.

That gap — between what the file contains and what you walked in expecting to find — is where cases are lost. Not in the courtroom, where everyone is watching, but months earlier, at a desk, in silence, when a competent professional quietly stopped looking because he had already found what he came for.

The law has a word for bias, and it means something narrower than this: the interested adjudicator, the pecuniary stake, nemo judex in causa sua. The other kind has no procedure attached to it — the systematic, predictable, thoroughly ordinary errors that every mind makes, including yours, including mine, including the judge’s. You cannot recuse yourself from those.

First, what a bias actually is

A cognitive bias is not stupidity, and it is not prejudice in the moral sense. It is a shortcut that usually works.

The mind processes an impossible volume of information on a tight budget. To cope, it compresses. It fills gaps with assumptions, it treats the vivid as the frequent, it treats the familiar as the true, and it treats a conclusion it has already reached as a fact it has already verified. Most of the time this is a feature. You do not deliberate about whether the chair will hold you. You sit.

The trouble is that the shortcut does not announce itself. Bias does not feel like bias from the inside. It feels like judgment. It feels like experience. It feels, very often, like being right — which is precisely why seniority offers no protection and sometimes makes things worse.

There is a finding about this, and it is the least comfortable one in the field. Emily Pronin and her colleagues described common biases to people and asked two questions: how susceptible are you, and how susceptible is the average person? Respondents placed themselves below average with great consistency — after having the bias explained to them, and in some studies immediately after demonstrating it. The bias blind spot is the bias that conceals the rest. We detect distortion in others by watching what they do, and in ourselves by consulting how we feel, and since bias does not feel like anything, the introspection comes back clean.

It also predicts what will happen to this article. Most readers will finish it with a particular colleague in mind. If one has already occurred to you, that is not a failure of attention. It is the finding, working.

---

Part I — Six families of bias

1. Biases of belief: the mind defends what it already holds

Confirmation bias is the tendency to seek, notice, and remember information that supports what we already think, and to subject contrary information to a standard of proof we never apply to the agreeable kind.

You see it when someone decides a new colleague is arrogant, and every subsequent interaction becomes evidence for the theory — the short reply was curt, the long reply was showing off. You see it in medicine, where a doctor who forms a diagnosis early orders the tests that would confirm it. You see it every day in politics, where the same press conference sends two audiences home vindicated.

Its close cousin, motivated reasoning, adds an incentive: we do not merely believe what we expect, we believe what suits us. And belief perseverance completes the set — once a belief is formed, discrediting the evidence that produced it often does not dislodge the belief.

2. Biases of number: the first figure owns the room

Anchoring is the mind’s habit of latching onto the first quantity it encounters and adjusting insufficiently from there. The anchor need not be relevant. It need not even be plausible. Experiments have shown that spinning a wheel of fortune before asking an unrelated numerical question shifts the answers.

In daily life the anchor is the crossed-out MRP on the sale tag, the “was ₹4,999” that makes ₹2,499 feel like a rescue. It is the estate agent’s opening quote. It is the salary a candidate names first, which will constrain the entire negotiation regardless of what the role is worth.

Framing works on the same weakness from a different side. The same fact described two ways produces two decisions: a procedure with a “90 percent survival rate” is accepted far more readily than one with a “10 percent mortality rate”. Nothing changed except the sentence.

3. Biases of memory: the vivid, and the inevitable-in-retrospect

The availability heuristic makes us estimate probability by how easily an example comes to mind. So we fear air travel more than the drive to the airport, worry about shark attacks and not about the stairs, and believe crime is rising because we saw three reports of it this week. Vividness masquerades as frequency.

Hindsight bias — the “I knew it all along” effect — quietly rewrites what we believed before we knew the answer. After a market crashes, everyone recalls having seen it coming. After a marriage ends, the warning signs are obvious. The knowledge of the outcome contaminates the reconstruction of the decision, and it does so invisibly: people sincerely misremember their own earlier predictions.

Its close relative is outcome bias, and the two are worth keeping apart. Hindsight bias is about memory: I knew this would happen. Outcome bias is about evaluation: it turned out badly, therefore the decision was bad. The first rewrites what you believed; the second condemns a choice that was sound when it was made. Almost the whole of negligence litigation runs on the second, which is why the law has had to build a defence against it.

Add recency (the last thing you read weighs more than the best thing you read) and you have a memory system that is excellent at storytelling and unreliable at accounting.

4. Biases of people: my circumstances, your character

The fundamental attribution error is the habit of explaining our own conduct by circumstance and everyone else’s by character. I snapped because I have been up since four; you snapped because you are difficult.

The halo effect lets one strong impression colour every unrelated assessment. A well-dressed, articulate person is rated as more competent, more honest and more likely to repay a loan, on no evidence at all. Its inverse — the horn effect — does the same damage in reverse.

In-group bias and affinity bias tilt us towards people who resemble us: same college, same city, same mother tongue, same golf handicap. This is the family of biases the law has actually theorised, and we will return to it.

5. Biases of commitment: throwing good after bad

The sunk cost fallacy is the compulsion to continue because of what has already been spent. It is why people sit through terrible films they paid for, finish books they hate, and stay in ventures long past the point of rescue. The money is gone in both branches of the decision; only the future differs. The mind refuses to accept this because abandoning the project means admitting the earlier spending was wasted.

Status quo bias and the related omission bias make inaction feel safer than action even when the risks are identical. A harm you allowed sits more comfortably than an equal harm you caused, which is why the caveat gets added to the opinion and the clause does not get renegotiated.

Overconfidence does the rest. Ask a room of professionals to rate themselves against their peers and the average lands comfortably above the middle. The finding that holds up best is Philip Tetlock’s: across two decades of expert political forecasting, confidence rose far faster than accuracy, and the most confident forecasters were not the most accurate ones.

6. Biases of the invisible: what the data does not contain

Survivorship bias is drawing conclusions from a sample that has already been filtered by success. The familiar illustration is Abraham Wald’s wartime work on aircraft armour: the returning bombers showed damage in particular places, and the reason those places did not need armour is that planes hit there returned.

The same blindness has a numerical form. Told that a test is 99 percent accurate for a condition affecting one person in ten thousand, most people read a positive result as near-certainty. It is not: the base rate swamps the accuracy, and the great majority of positives are false. What is absent from the sample governs the answer, and nothing in the sample says so.

---

Part II — What this does to a lawyer’s work

Everything above is ordinary human machinery. The reason it deserves a lawyer’s attention is that our work is unusually good at concealing its effects. A biased title opinion looks exactly like a sound one until the day someone challenges it. A biased research note reads beautifully. Nothing squeaks.

In legal research

Research is where confirmation bias does the most damage, because the tools are built to cooperate with it. Type a proposition into any database and it will find you cases. It will always find you cases. The database has no view on whether your proposition is correct; it simply matches strings.

So the sequence goes: the client states the problem, you form a preliminary view within about four minutes, and everything that follows is a search for support rather than a search for the answer. You find eleven judgments. You do not find the Full Bench decision that reads the section the other way, because you never searched for the other way. Nobody in the room knows what is missing — that is the defining property of what is missing.

Anchoring compounds it. The first authority you read frames every subsequent one; cases that fit the frame feel “on point”, cases that do not feel “distinguishable on facts”. The word distinguishable is doing enormous unexamined work in a great many opinions.

Then there is authority bias — the tendency to weigh a proposition by who said it rather than what was decided. A stray observation of the Supreme Court is treated as binding when it is obiter; a considered ratio of a Division Bench of one’s own High Court, squarely on the point and binding, is passed over because the paragraph is less quotable. Every practitioner has read a note that cites a two-line quotation from a judgment whose facts had nothing to do with the case at hand.

In investigating facts and conducting due diligence

This is my own daily bread, so let me be concrete about the trade rather than about any file. A title search conducted to confirm title is not a title search. It is a ritual.

These are the failures the work is prone to. An encumbrance certificate is taken for thirteen years because thirteen years is what one takes, and the transaction that created the defect sits in year sixteen. Link documents get read for names and dates, which match, rather than for recitals, which do not. The revenue record shows the vendor’s name and the search stops there, although the extent in the record and the extent in the deed differ by a few guntas and nobody reconciles them. The vendor is a respectable man with a well-appointed office, which is the halo effect writing a legal opinion. Not one of these requires carelessness. Each one requires only that the question asked was the wrong question.

The countermeasure is structural, not attitudinal. Do not resolve to be careful — you already are careful. Change the question. Instead of “is this title good?”, ask “if I were acting for a rival claimant, how would I attack this title, and with which document?” That single reframing recruits the same confirmation machinery to work in the opposite direction, which is the only reliable way I know to use it. It is the change I made in my own practice, and it is the reason the list above is written from experience rather than from theory.

In drafting

Drafting is where the curse of knowledge arrives. Once you understand what the clause means, you lose the ability to read it as someone who does not. Ambiguity becomes invisible to its author. This is why the drafter is systematically the worst person to review the draft, and why “I’ve read it four times” is not reassurance.

Optimism bias shapes the substance. A contract drafted while the parties are pleased with each other quietly assumes performance. Termination is a formality, dispute resolution is boilerplate, force majeure is copied from the last file, and the indemnity is capped in a way that made sense for a different transaction. Every one of those clauses exists for the day the parties hate each other, and it is drafted on the day they do not.

Framing decides more than lawyers admit. “The Purchaser shall not be liable unless” and “The Purchaser shall be liable only if” allocate identical risk and are argued as though they do not.

In argument and in court

By the time a matter reaches argument, months of investment have accumulated, and the sunk cost fallacy has quietly become a doctrine of the file. The weak ground stays in because it survived three drafts. The hopeless appeal is filed because the suit was fought for eleven years. Nobody performs the only correct calculation, which is prospective: from today, ignoring everything already spent, what is this worth?

Overconfidence sets the odds. Ask any set of opposing counsel to state their chances of success and the two figures will comfortably exceed 100 percent. Both are sincere. Both have read the same record. Each has assessed it through a lens the other cannot see and neither can remove.

And the narrative bias — the mind’s craving for a coherent story — is what actually persuades. A tidy account of events that explains everything feels true in direct proportion to its tidiness, which is a warning as much as a technique. Good advocacy exploits it. Good preparation guards against it, by asking which parts of your own story are load-bearing and which are decorative assumptions you have started to believe.

In reading witnesses

Credibility is routinely assessed from demeanour — hesitation, evasiveness, the eye that will not meet yours — and appellate courts defer to those findings precisely because the trial judge saw the witness and they did not. The empirical position is unkind to the practice. The largest synthesis of the evidence — Charles Bond and Bella DePaulo’s meta-analysis of 206 studies covering some 24,000 observers — found average accuracy at telling lies from truths of about 54 percent, barely above chance, and that professional experience and training confer almost no advantage. Its most awkward finding for courtroom practice is that lies are detected rather more accurately from what is heard than from what is seen. Nervousness looks identical whether it is produced by lying, or by being cross-examined in a language you handle badly, in a building you have never entered before, about events you would rather not revisit. Add to that the well-documented unreliability of eyewitness identification, which degrades further with stress, delay and suggestion in the manner of questioning.

None of this means the practice should be abandoned; a trial has to run on something. It means knowing how much weight the inference will carry before resting a submission on it, and being slow to concede an adverse credibility finding merely because the trial court used the word “demeanour”.

In estimating time, cost and settlement

The planning fallacy is the systematic underestimation of how long something will take and what it will cost, and its signature feature is that experience does not cure it. The same lawyer who has never once finished a matter within the estimate produces the next estimate with undiminished confidence, because the estimate is built by imagining the task going well rather than by consulting what comparable tasks actually took. The correction is not resolve. It is to find out what three similar matters took in that court, and use that figure instead of the one that comes to mind.

Reactive devaluation is quieter and costs more. The identical term is worth less because the other side proposed it — if they are offering it, it cannot be worth having. It is a large part of why a mediator’s proposal succeeds where the opponent’s identical proposal failed a fortnight earlier. Before rejecting an offer, it is worth asking what you would have made of it had your own client thought of it first.

In advising the client

The most expensive bias in practice is the one that operates on advice. Clients do not want probabilities; they want reassurance, and they reward the lawyer who supplies it. The pressure to say “strong case” is enormous, and it is never experienced as pressure. It is experienced as confidence.

A discipline worth adopting: write down, at intake, your estimate of the outcome and the two or three facts that would change it. Seal it in the file. Read it at the end. Most lawyers have never once seen their own prediction record, which is exactly why most lawyers believe it is good.

---

Part III — The newest bias, and what the Supreme Court has done about it

Automation bias is the tendency to over-trust output from a machine, and to relax the scrutiny we would apply to the same statement from a person. It is not new — pilots have been trained against it for decades — but it has arrived in Indian legal practice with unusual speed and unusual consequences.

A generative model produces citations that look correct. The case name has the right shape. The citation has the right format. The paragraph reads like something a Bench would say. Every surface cue our profession uses to assess reliability is present, and every one of them is decorative.

The Supreme Court addressed this squarely in Pooja Ramesh Singh v. Jammu and Kashmir Bank Ltd., 2026 INSC 668, decided on 2 July 2026. A Bench of Narasimha and Aradhe, JJ. set aside orders of the NCLT and the NCLAT in a Section 7 proceeding after examining the six authorities the adjudicating authority had relied on. Three citations did not exist at all. Two were correct citations carrying paragraphs that appear nowhere in those judgments. One was a real decision cited under the wrong cause title, with an invented paragraph attached. The first respondent’s affidavit stated that its counsel had cited none of them and that the tribunal had obtained them through its own research — after which they passed unchallenged through an entire tier of appeal.

The Court’s language is worth quoting rather than paraphrasing. Hallucinated material used as precedent, it held, “is like the release of methyl isocyanate in the province of law and justice: invisible, insidious, and catastrophic by the time anyone notices. It not only contaminates but takes away the very lifeblood of judicial determination.”

The operative holding is short. A decision resting on such material “is no decision in the eyes of the law, irrespective of whether such material had a direct or indirect bearing on the decision-making”, and is to be set aside “even if an iota of fake or hallucinated material enters the decision-making process”. Both orders went, and the Section 7 application was restored for fresh decision on merits.

The Court further declared that citing such judgments without verification “is a misconduct on the part of an advocate”, and directed the Bar Council of India to constitute a committee and prescribe norms carrying disciplinary consequences. That declaration deserves a careful note, of the kind this article has already recommended for anyone reading a quotable line in a judgment: on these facts no advocate had cited the fake authorities at all, so the misconduct proposition was not necessary to the decision. It is a declaration of the Court’s position and a direction to the Bar rather than a holding worked out on contested facts. It will be followed, and it should be. It is still not ratio, and anyone facing a misconduct allegation founded on it should know the difference.

For a practitioner the useful part is not the admonition. It is the ground. If an order against your client — at any tier, from any court, tribunal or quasi-judicial authority — rests even partly on an authority you cannot trace, there is now a Supreme Court declaration that the order is unsustainable, and that it makes no difference whether the phantom authority was central or incidental to the reasoning. The burden that comes with it is practical rather than rhetorical: you must verify from the reports that the citation, or the quoted paragraph, genuinely does not exist, and put that verification on affidavit instead of asserting it from the bar. The test cuts both ways, which is the real lesson. The first thing to do with any adverse order is to check that every authority in it is real. That is a ten-minute exercise which, until this July, had no name.

The judgment did not come from nowhere. The Bombay High Court had quashed a substantial faceless assessment order in October 2025 where the assessing officer cited three non-existent decisions, and in January 2026 imposed costs of ₹50,000 on a litigant whose submissions relied on a judgment the Court could not trace. Earlier still, in March 2025, the Karnataka High Court ordered an inquiry against a trial judge who had relied on two Supreme Court decisions that were never delivered. The problem was never confined to the Bar.

Two things about that judgment are worth holding onto, because a summary of it tends to lose both. The first is that it is not a judgment against AI. The Court records its “resolve to adopt AI technology in aid of adjudication,” asserts control over the process “with a human in the loop at every stage,” and says plainly that the problem cannot be solved by judicial orders at all — only by public policy and enforceable rules, which are already being framed. The second is the illustration it chose, from England. In Anthony Malcolm Cork v. Mark Smith, [2026] EWHC 1199 (Ch), a junior solicitor at Pinsent Masons accepted a fabricated statutory quotation produced by the firm’s internal AI; the supervising partners did not catch it; and when the High Court questioned the non-existent text, the firm filed a further letter of explanation also generated by the software, which masked the first error with a second. The firm referred itself to the regulator. The failure was not the hallucination. It was the absence, twice over, of anyone who opened the source.

None of this is an argument against the tools. It is an argument about where verification sits. The reason automation bias is dangerous is not that machines are unusually unreliable — it is that they are unusually fluent, and fluency is the cue we have been trained to trust. Every citation goes to the reporter. Every quoted paragraph is opened and read in the original. No exceptions, and no exceptions for the ones that look obviously right, because those are the ones that get through.

---

Part IV — The law’s own theory of bias

Law noticed bias long before psychology named it, and built machinery against it. But the machinery is aimed at a narrower target than the problem.

The rule against bias — nemo judex in causa sua — is aimed principally at interest. It asks whether the adjudicator had a stake, a relationship, or a prior commitment, and the test is objective. In A.K. Kraipak v. Union of India, (1969) 2 SCC 262, the Court held that the dividing line between administrative and quasi-judicial function was being obliterated, and set aside a selection where the acting Chief Conservator of Forests sat on the board that was selecting him, even though he withdrew when his own name was considered. The point was not that he actually favoured himself. It was that the Bench declined to believe the affidavits of the other members swearing he had not influenced them:

“In a group deliberation each member of the group is bound to influence the others, more so, if the member concerned is a person with special knowledge. His bias is likely to operate in a subtle manner. It is no wonder that the other members of the selection board are unaware of the extent to which his opinion influenced their conclusions.”

That is a Bench in 1969 describing conformity and group influence, and declining to accept sincere denials of it as evidence of anything. The Court added that mere suspicion of bias is not enough — there must be a reasonable likelihood, judged by “human probabilities and ordinary course of human conduct”.

Manak Lal v. Dr. Prem Chand Singhvi, AIR 1957 SC 425, made the standard plain: what matters is not actual bias but the reasonable apprehension of it. The Court adopted R v. Sussex Justices, ex parte McCarthy, [1924] 1 KB 256, where Lord Hewart CJ held that the answer “depends not upon what actually was done but upon what might appear to be done”. The pairing repays attention: in Sussex Justices the conviction was quashed because there had been no waiver, while in Manak Lal the appellant lost precisely because he had known the facts, said nothing, and taken his chance on a favourable report. Same test, opposite outcomes, thirty-three years apart, and the difference was when the objection was taken. Ranjit Thakur v. Union of India, (1987) 4 SCC 611, put it in the sharpest form: the judge is not to examine his own mind and ask, however honestly, whether he is biased; he is to look at the mind of the party before him. And in Justice P.D. Dinakaran v. Judges Inquiry Committee, (2011) 8 SCC 380, the Court did not merely apply the real-likelihood standard, it chose it: for non-pecuniary bias, it held, that test is “preferred over” the reasonable-suspicion test. The question is whether a fair-minded and informed person, having considered all the facts, would reasonably apprehend that the judge would not act impartially — or, put the other way, whether a reasonably intelligent man fully apprised of the facts would have a serious apprehension of bias.

It would be wrong to say the doctrine ignores cognition altogether. In State of W.B. v. Shivananda Pathak, (1998) 5 SCC 513, the Court recognised what it called a new form of bias — judicial obstinacy — and held that a judge whose decision has been overruled cannot rewrite it in the same or collateral proceedings, because to do so “would be exhibitive of his bias in his own favour to satisfy his egoistic judicial obstinacy”. The Court was careful about how far that reaches. It noted the distinction between prejudging facts about a particular party and holding settled views on general questions of law or policy, observed that a judge would disqualify himself in the first case and might not in the second, and then expressly left the question open. Nor is the rule absolute: the doctrine of necessity permits an otherwise disqualified adjudicator to act where there is no one else who can.

But the same judgment concedes the whole of the problem this article is about, and it does so by quoting an American judge. Frank J. in In re Linahan, 138 F.2d 650:

“If, however, ‘bias’ and ‘partiality’ be defined to mean the total absence of preconceptions in the mind of the judge, then no one has ever had a fair trial and no one will. The human mind, even at infancy, is no blank piece of paper. We are born with predispositions… Much harm is done by the myth that, merely by… taking the oath of office as a judge, a man ceases to be human and strips himself of all predilections, becomes a passionless thinking machine.”

That passage sits in the Supreme Court Reports. It concedes that the standard the rule against bias appears to demand is one no human being has ever met, which is why the doctrine settles for something narrower and more achievable instead.

What the doctrine does not reach is the honest, disinterested, thoroughly impartial adjudicator whose reasoning is nonetheless distorted by anchoring, hindsight or availability. There is no application that lies against the human condition.

So the law developed a second layer of defences, mostly unlabelled. They are worth recognising as such:

The burden and standard of proof work on the parties rather than on the adjudicator, but they do similar work. They force the side that has already decided what happened to establish it against a fixed benchmark, rather than to accumulate consistency. The familiar caution that suspicion, however grave, cannot substitute for proof is a rule about the difference between a coherent story and a proved one.

The five golden principles of circumstantial evidence are the clearest example, and their origin is worth knowing. In Hanumant v. State of Madhya Pradesh, AIR 1952 SC 343, the Supreme Court held that the circumstances “should be such as to exclude every hypothesis but the one proposed to be proved” and that there must be “a chain of evidence so far complete as not to leave any reasonable ground for a conclusion consistent with the innocence of the accused” — the formulation later systematised as the five principles in Sharad Birdhichand Sarda v. State of Maharashtra, (1984) 4 SCC 116. The Court did not reach it in the abstract. It reached it by quoting a warning given to a jury in 1838 by Baron Alderson in Reg v. Hodge:

“The mind was apt to take a pleasure in adapting circumstances to one another, and even in straining them a little, if need be, to force them to form parts of one connected whole; and the more ingenious the mind of the individual, the more likely was it, considering such matters to overreach and mislead itself, to supply some little link that is wanting, to take for granted some fact consistent with its previous theories and necessary to render them complete.”

Confirmation bias and the craving for narrative coherence, described from an English bench in 1838, more than a century before the psychological literature gave either a name, and adopted by the Supreme Court of India in 1952.

And Hanumant did not merely quote it. It applied it. The Court held that the courts below “have just fallen into the error against which warning was uttered by Baron Alderson”, then dismantled the chain of inference item by item. One circumstance was “of a neutral character”, but the magistrate had been “obsessed with the idea” that it could mean only one thing. Another was innocent, and told against the accused only “unless one starts with a presumption of guilt”. On a third, the magistrate “misled himself completely”. Two men convicted by three courts were acquitted.

The requirement that circumstances exclude every other reasonable hypothesis is, in psychological terms, a compulsory falsification exercise. It forces somebody in the room to generate and test the alternatives, which is exactly what the biased mind declines to do — and Hanumant is what it looks like when an appellate court enforces that against three benches which did not.

The warning against hindsight is explicit in professional negligence. Jacob Mathew v. State of Punjab, (2005) 6 SCC 1, was a prosecution under Section 304A IPC. A three-judge Bench adopted the Bolam standard — the ordinary competent practitioner exercising ordinary skill — as the measure of the duty, while holding that criminal liability requires something more: negligence of a gross or very high degree, so that the words “rash or negligent” in Section 304A are to be read as qualified by “grossly”. Within that framework the Court stated the hindsight rule as a formal conclusion: “the standard of care, while assessing the practice as adopted, is judged in the light of knowledge available at the time of the incident, and not at the date of trial.” A doctor who followed a practice acceptable to the profession of that day is not negligent because a better course was also available, or because a different one would, as we now know, have worked.

The same judgment goes further than the doctrine required, and names outcome bias directly. “Care has to be taken,” the Court said, “to see that the result of an accident which is exculpatory may not persuade the human mind to confuse it with the consequence of negligence.” Quoting Merry and McCall Smith’s Errors, Medicine and the Law, it identified the reflex behind most negligence litigation — “a marked tendency to look for a human actor to blame for an untoward event, a tendency which is closely linked with the desire to punish” — and the shape that reflex takes, which is “to blame the last identifiable element in the chain of causation, the person holding the ‘smoking gun’.” Best of all, and worth pinning above a desk: “There is a tendency to confuse the reasonable person with the error-free person.”

Structured quantification pulls in the same direction. The multiplier method systematised in Sarla Verma v. Delhi Transport Corporation, (2009) 6 SCC 121 — since modified by a Constitution Bench in National Insurance Co. v. Pranay Sethi, (2017) 16 SCC 680, which standardised the addition for future prospects and fixed the conventional heads, while expressly preserving the Sarla Verma multiplier table — was adopted to end wide and divergent awards on comparable facts. The Court’s stated object was uniformity rather than debiasing, but replacing unstructured intuition with a formula has the incidental effect of denying any single figure the chance to become the anchor. It sacrifices some individual fit and buys consistency.

It would be too neat to stop there, and stopping there would be the very error this article is about. The same body of law contains rules that entrench bias rather than correct it. Stare decisis is anchoring given institutional status — the first court to decide a question fixes the frame for every court after it, and the entire point of the doctrine is that the frame should be hard to shift. The presumption of regularity attaching to official records is exactly what allows a searcher to stop at the first entry that agrees with him. Limitation and res judicata foreclose the correction of errors that everyone now accepts were errors. A system that debiases in some places and hardens bias in others is not being inconsistent. It is trading accuracy against finality, and finality usually wins. Knowing which trade a particular rule is making is the useful thing.

Read together, a good deal of procedural and evidentiary law is intelligible as institutional debiasing — the accumulated response of a system that learned, case by case, that individual conscientiousness is not enough. The rest of it is the system deciding that some errors are cheaper to live with than to fix.

---

Part V — Working countermeasures

Resolutions do not work. Bias is not defeated by intending to be objective; the intention is fully compatible with the error. What works is procedure — small structural changes that make the mind do work it would otherwise skip.

---

A closing thought

The law’s traditional answer to bias is disqualification: find the interest, remove the person, and the problem leaves with him. For everything else, there is nobody to remove. The mind that misses the encumbrance in year sixteen is the same mind that will draft the opinion, argue the appeal and advise on the settlement. You cannot recuse it. You can only give it work that is harder to do carelessly — a question framed to hunt for defects, a search run in the opposite direction, a colleague instructed to be unpleasant about your reasoning, a prediction on record where you will have to face it.

The best lawyers I have watched are not the ones with the fewest biases. They are the ones who assume, as a matter of routine, that they have them, and who build their files accordingly.

Vivek Sanghi practises law in Hyderabad, with a focus on legal research, property title work, and contract and legal drafting.

---

Sources and further reading