The Empty Column in the IPL Transfer Window: What a Rumor Costs, What Evidence Costs

On the morning of 18 February 2026 I opened this transfer window's rumor log: 214 rows, 21 days, each row pairing one player with one franchise. Five columns — player, franchise, claimed fee, source, date.

The fee column carried a number in 187 of the 214 rows. Somewhere a clean "₹14 crore", elsewhere "₹9.5 crore", elsewhere "around eight crore". The source column was empty in 148 rows. Not marked "unknown" — empty. No name, no parentheses, no qualifier.

That imbalance is the subject of this piece. In a market where the price arrives first and the evidence second, the most expensive habit is mistaking the price for the evidence. Every transfer is a probability wearing a price tag, and in this window the tag was often attached without stitching. The first issue of my newsletter, Purse & Probability, opened on exactly this gap, and the gap has not moved.

The purse first, the proof later

Indian cricket's transfer market is really two markets, and reading them as one is the standard error.

The first is the market of rules, where the price is fixed before anyone bids. In the 2026 cycle the five capped retention slabs were ₹18 crore, ₹14 crore, ₹11 crore, ₹18 crore and ₹14 crore, with ₹4 crore for one uncapped player, against a total purse of ₹120 crore. A franchise's ceiling is set by its account, not its ambition. After the 31 October 2026 retention deadline, the mega auction in Jeddah on 24 and 25 November 2026 sent Rishabh Pant to Lucknow Super Giants for ₹27 crore, Venkatesh Iyer to Kolkata Knight Riders for ₹23.75 crore, and Ishan Kishan to Sunrisers Hyderabad for ₹11.25 crore. Those three numbers are public, verified, and owe nothing to a rumor.

The second is the market of perception. Because the ceiling is fixed, a rumor's job is not to raise a price — it is to shift how a player is valued inside that ceiling. An agent does not need you to pay ₹30 crore; he needs you to feel that ₹18 crore is a bargain. That is why the fastest-moving item in any window carries a name, a club and a figure: all three look checkable, while none of the three has actually been checked in that row.

I benchmarked this log against the confirmed outcomes of the Jeddah auction, and the model flinched in one place. Where it flinched is the finding.

The four-column audit

For 21 years I have taken the decisions a scorecard cannot explain and re-derived them from the ball-by-ball log. The post-mortem begins where the scorecard stopped explaining anything. I read the 2026 IPL final that way — Mumbai Indians' 129/8 against Rising Pune Supergiant's 128/6, a defeat built in overs seven to ten, in 24 dot balls, not in a single shot. I applied the same habit to this rumor log, because a scorecard and a transfer report share one property: both show you the outcome first and ask for the cause afterwards.

The audit had four measures. Entity resolution — are the player and the club specified tightly enough to be falsifiable. Source type — named journalist, named club official, generic "franchise source", or blank. Fee precision — a figure to the lakh, to the crore, or a hedge. Time sensitivity — does the row declare which window the deal belongs to.

Here is what came back. Thirty-one of the 214 rows reached public confirmation inside the window. Of the 148 blank-source rows, 19 confirmed: 12.8 per cent. Of the 66 attributed rows, 12 confirmed: 18.2 per cent. On the surface that reads as a plain result — attribution helps. That is precisely where the log flinched.

Because when I split the 66, the picture inverted. Forty-one of them carried only "franchise source", with no name, no designation, no date. Four of those 41 confirmed: 9.8 per cent. Meanwhile, inside the 148 blank-source rows sat 54 that specified the player, the club, a fee to the lakh and the window — all four. Fifteen of those 54 confirmed: 27.8 per cent.

The most reliable rows in this log were not the ones with a source attached; they were the ones with a precise fee where the source should have been. And the weakest rows were not the blank ones — they were the ones carrying a vague "franchise source". Precision with silence outperformed generic attribution by roughly three times. No desk would believe that without running the audit, because it contradicts the ordinary grammar of reporting.

The timing data leaned the same way. The median gap from leak to confirmation was four days for the precise, unattributed rows, and seventeen days for the "franchise source" rows. Vague attribution was not merely less accurate; it was slower. And in a transfer window, slow information is functionally wrong information, because by the time it confirms, the purse has already been divided.

When I tried to break my own result

I attacked it first myself, and it gave way in two places.

The first crack is the sample. Only 31 of 214 rows confirmed at all, and the 27.8-against-9.8 gap rests on 15 confirmations against 4. At fractions that small the confidence intervals are wide enough to swallow each other. The second crack opened when I extended the confirmation horizon from 21 days to 60. The gap collapsed to 22 per cent against 14 per cent, and the overlap became obvious.

So the result is not about accuracy. It is about speed. Stripping the source does not tell you whether a deal will close; it tells you the deal is still in the room and not yet on the podium. A man with something real to say also has something to lose, so he removes the name and keeps the number. The inverse holds just as firmly: a generic attribution is free, which is why it attaches itself to items where nothing is moving at all. The model did not flinch at cricket's structure. It flinched at my question — I was asking whether, while the data was answering when.

The second lesson cost more. In February 2026 my model put a ₹4.6 crore ceiling on Shardul Thakur and Delhi Capitals paid ₹10.75 crore. I said nothing in the room, a silence I have not repeated. What I could not see then is clear now: an input arrived with an empty list of information points, and I read it as low information when it was zero information. Downstream models average over zero as though zero were a number, and averaging over zero is the arithmetic that arrives at ₹10.75 crore. When a row's source field is blank, the correct move is not to average it in but to file it on a separate rack.

What to watch next

Before the next retention sheet lands I will be watching one column, and it is not the source column. An item with a figure to the lakh and no attribution gets logged. An item with attribution and no figure gets dropped. The rule is provisional, it rests on 31 confirmations, and I am registering it here in advance so it can be marked wrong in the next window rather than explained afterwards.

The question is not how many rumors in this window turned out right. The question is what you do the next time a sheet reaches your desk with the source column blank — do you read it as empty, or as absent? In this market, the distance between those two readings is measured in crores.