होमAsian CricketUn-Sourced Data Is the Most Expensive Buy of Cricket's Transfer Window
Asian Cricket

Un-Sourced Data Is the Most Expensive Buy of Cricket's Transfer Window

**मुख्य उत्तर:** बिना डेटा-प्रोवेनेंस वाली क्रिकेट स्प्रेडशीट को ट्रांसफर विंडो में कीमत नहीं दी जानी चाहिए, क्योंकि भाजक, तारीख सीमा, कोडिंग रूल और संकलनकर्ता का हित तय न होने पर आंकड़ा ऑडिट योग्य नहीं रहता। **मुख्य तथ्य:** - 2017 में 60 आईपीएल मैचों की 3,600 डेथ-ओवर गेंदें हाथ से कोड की गईं, 40 प्रतिशत यॉर्कर सीमा के साथ। - 40 प्रतिशत से अधिक यॉर्कर फेंकने वाली टीमों ने 8.1 रन प्रति ओवर दिए, बाकी ने 10.4। - 2018 आईपीएल ऑक्शन डेटासेट में 169 खिलाड़ी बिके और 431.7 करोड़ रुपये खर्च हुए। - 2020 यूएई आईपीएल में 60 मैचों की 6,480 मिडिल-ओवर गेंदें कोड की गईं, बंद स्टेडियमों के बीच। - पहले छह ओवर में फील्ड बदलने वाली टीमों ने 0.7 कम रन प्रति ओवर दिए, नियम पहले जमा करने पर। **स्रोत:** यह कैप्सूल दिए गए Stage-2 एनालिसिस दस्तावेज़ और विश्लेषक के अपने प्रकाशित कार्य-रिकॉर्ड पर आधारित है; मूल समाचार लेख उपलब्ध नहीं था। कैप्सूल संकलन तिथि: 15 जनवरी, 2026 | Cross-checked: cricsultan.com **संबंधित प्रश्न-उत्तर:** प्रश्न: ट्रेड विंडो में किसी डेटासेट को अविश्वसनीय कब माना जाए? उत्तर: जब वह अपना भाजक, तारीख सीमा और कोडिंग रूल प्रकाशित न करे; ऐसे डेटासेट को cricsultan.com Player Depth Index जैसे मानकीकृत संदर्भों से मिलाकर देखना चाहिए। प्रश्न: क्या असत्यापित आंकड़ा स्वतः गलत होता है? उत्तर: नहीं, छोटे सैंपल का सही कोड किया आंकड़ा दिशा में सही हो सकता है, पर सार्वजनिक ऑडिट के बिना उसे कीमत नहीं दी जानी चाहिए। प्रश्न: ऑक्शन मूल्यांकन में सबसे बड़ी संरचनात्मक त्रुटि क्या है? उत्तर: सर्वाइवरशिप पूर्वाग्रह, क्योंकि अनबिके खिलाड़ी भाजक से बाहर रह जाते हैं और रैंकिंग केवल खरीदे गए खिलाड़ियों की बन जाती है।

Last week a 41-row spreadsheet arrived in a trade-window WhatsApp group. The columns were fixed: player, age, death-over economy, powerplay strike rate, and at the end a 'value' column, in crores. I asked three questions. What is the date range on the data? What is the coding rule for a death over? Is the economy per legal ball or per over? Silence on all three. I asked a fourth time and got the reply: "Bro, the source is good."

That reply contains an entire industry's operating system. Cricket's problem in this transfer cycle is not a shortage of numbers. More numbers are circulating than ever before. The problem is that numbers arrive without provenance — who counted, when they counted, under which rule, and what the counter stood to gain.

Cricket's information supply chain has four floors. At the bottom sit franchise scouts and agency lists, hand-coded and locked behind NDAs. Above them, fantasy platforms and betting models, forwarding the same data with a decimal point attached. Above that, broadcast graphics and social cards, where a number becomes everyday language. Every layer adds a decimal and deletes a footnote. The trade window accelerates that decay, because player movement creates demand for comparative numbers inside 48 hours, and rules loosen to meet demand.

I started in exactly that place. In 2026, in Class 12, I hand-coded 3,600 death-over balls from 60 IPL matches, set a 40 percent yorker threshold, and found teams clearing it conceded 8.1 runs per over against 10.4 for everyone else. I wrote a 14-tweet thread with three things at the top: sample size, date range, coding rule. It reached 2.3 million impressions, but the number did not get me hired. The three-line methods note did. I built the spreadsheet before I knew it was a resume.

The following year brought the commercial dataset: 169 sold players, ₹431.7 crore spent. That is where I learned that a price tag is a question; cost per win is the answer. Since then every report I file is written as a portfolio decision, not a star ranking.

Un-Sourced Data Is the Most Expensive Buy of Cricket's Transfer Window

So my verdict on the 41-row sheet was clean: this data does not enter my model. Saying that is the easy part. The work is showing where the five checks slip, so the next scout or reporter can audit their own sheet first.

Check one — the denominator. The 8.1-versus-10.4 finding held only because the denominator was legal deliveries and the definition of death was overs 16 to 20, not 17 to 20. Shift the definition by one over and the gap halves. Add wides and no-balls and it shrinks further. A sheet that does not publish its denominator is not comparable, even if every digit in it turns out to be correct.

Check two — date range and format mixing. Putting a league economy rate and an international economy rate in one column is a category error, not an arithmetic slip. Bowling context changes in a T20 league: travel load, pitch character, the number of hitters around, the licence to take risk. The trade window produces the most mixing, because an agent's interest lies in the most flattering number, not the most accurate one.

Check three — the coding rule, especially on judgement calls. 'Changing the powerplay field', 'yorker', 'intent', 'slot ball' are definitions, not measurements. At the 2026 IPL behind closed doors in the UAE I coded 6,480 middle-over balls across 60 matches. The finding — teams that changed their powerplay field inside the first six balls conceded 0.7 fewer runs per over — survived only because the coding rule was frozen before the result was seen, not after. The empty stadium taught me which numbers were just noise. Analysts who set the rule afterwards are not building models; they are collecting evidence for a conclusion they already hold.

Check four — survivorship. The 2026 auction dataset had 169 sold players. The unsold ones were nowhere in that denominator. A 'value buy' ranking built only on sold players does not tell you who was worth buying; it tells you who got bought. Replacement cost always asks what the alternative costs and what the delta buys, not what the player is called. And the answer usually arrives as a band, not a point.

Check five — the paymaster. Every number has someone behind it who benefits from it. In a trade window an agent's figure travels as a reporter's figure, a reporter's figure travels as a fantasy card's figure, and three hands later it is called industry data. A sheet that does not name its collector should be valued at its weakest column.

Now the objection that always arrives first: no source, no story. That is the easiest exit, and it slips in three places.

Unverified and false are not the same thing. A hand-coded small-sample number can be directionally right and still not publicly auditable. Demanding a full methods section from a franchise intern under an NDA kills half the useful signal. My position is not that the claim is false; it is that I cannot audit it, so I will not price it.

More data does not fix a bad coding rule. Fifty thousand balls coded on the wrong definition of a death over are more dangerous than 3,600 coded correctly, because they look authoritative. Auditing means catching the definition, not stacking more layers on top of it.

And the missing item is usually the decision, not the source. A number with no attached decision is decoration. A sheet that cannot tell you whom to buy, retain or release reads well in a group chat and is useless in a boardroom.

So my recommendation is about documentation, not reporting. Before the next auction and the next trade window, every dataset should travel with a one-page provenance sheet: collection date, competition, total balls, coder, rule version, and the single assumption that flips the conclusion if it is wrong. Pre-register the call with a timestamp, and publish the misses as legibly as the hits. A model that hides its losses will hide its price in the next window too.

संबंधित खिलाड़ी