The Zero Row: Who Writes the Result of a Match That Has No Ledger

There is a row in my spreadsheet numbered 4,812. It exists. The columns exist too — ball number, batter position, shot coordinates, run value, wicket probability. Every cell is empty. The scoreboard above it says Abahani Limited Dhaka won by five wickets. The ledger says nothing at all.

I built the first xG ledger in Sylhet, and those numbers rewrote the accepted history of the game. 132 matches, 14,800 shots, two junior loggers trained to record coordinates rather than opinions. That season, Abahani finished 14.2 runs above their expected total. Fourteen point two is a small number, but across 132 matches it separates clinical finishing from luck. From that ledger I learned the first lesson of this trade: the scoreboard and the process are two different documents.

This piece is not about the 14.2. It is about row 4,812 — the row that stayed empty and still survived inside the system as a valid record.

The record that looks complete and contains nothing

Data systems produce something called a null payload. The schema is intact, every column sits where it should, the format is flawless — and every field is either blank or a placeholder. This is not an error message. An error message is honest; it says I failed. A null record is dishonest; it says I am finished.

In cricket that difference is dangerous. If a scorecard scrape fails outright, we know the match data is missing. If the scraper instead returns an empty shell that passes validation, the system concludes the match happened and everything is fine.

The failure arrives in three shapes.

First, ingestion failure. Modern scorecards render in JavaScript. The crawler opens the page, snapshots before the scripts run, and captures a skeleton with nothing inside. Or the page returns a paywall stub, or a redirect interstitial. Cricket sites are especially prone to this, because live scores often sit on a separate subdomain that blocks automated access.

Second, template echo. A model is instructed to extract facts from a scorecard and instead returns its own instruction text — identify from the information above. It never read the document. That is an ingestion failure wearing the costume of an output.

Third, and most common, substituting unknown for neutral. No venue data, so we assume a neutral venue. No pitch report, so we assume normal behaviour. No rain or Duckworth-Lewis reference, so we assume a full-length match.

Not knowing and being neutral are not the same thing, and confusing them is the most ordinary crime in cricket data.

What a dropped feed does to a season model

Suppose seven of 132 matches lose their ball-by-ball feed. The results exist; the shot-level data does not. Model the remaining 125 and your expected runs are computed on an incomplete season. The trouble is that those seven matches may have been the decisive ones — the semi-final, the tie-breaker, the rain-shortened game.

Your model is then simultaneously accurate and wrong. Accurate because nothing in the 125-match calculation is false. Wrong because the question you asked — who was best — required the missing seven.

I first met this problem in strike rates. A batter had 40 innings of data and six innings missing. In two of the missing innings he was dismissed for zero. Removing them made both his average and his strike rate look better. The number in the report was true. The story was a lie.

This is why I print the sample size next to every number. Not just the figure, but how many rows it stands on.

Auction valuation, where small samples command the highest price

Cricket's largest data market is the auction, and its largest risk is misreading an insufficient sample.

A young player has eleven domestic T20 innings, a strike rate above 180, and his price doubles. By international convention, a 180-plus strike rate marks an exceptional finisher — but across eleven innings the confidence interval is so wide that prediction is impossible.

In my ledger I write three things separately. First, what the number is. Second, the sample size and its limits. Third, the conditions that produced it — powerplay, death overs, or an easy pitch. Without the third, the first two are close to meaningless.

Auction premiums come in four kinds: the local young star, the all-rounder, the scarce position, and panic bidding. The first three are analysable. The fourth is not, and it moves the most money.

Price and international performance correlate far more weakly than we like to admit. A big fee is a big bet, not a proof. The transfer market is not a bazaar; it is a probability engine with agents.

The number travels; its assumptions do not

In 2026 I worked live xG for a regional broadcaster at the World Cup. France beat Croatia 4-2 in the final. My model put expected goals at 2.1 to 1.8, with France's PPDA at 12.4 — they let Croatia control midfield. France won on clinical finishing, not dominance. Across the tournament I logged 64 matches and 1,872 shots. The biggest surprise was Croatia: 1.8 expected goals from only seven shots on target.

What reached the screen was four numbers. What did not reach the screen were the assumptions underneath — how shot quality was weighted, which model, which era, which league benchmark. The next day the argument was whether France were weak or strong. Nobody asked for the error bar on 1.8.

Cricket has the same problem in a sharper form, because public graphics routinely display win probability — 78 percent, 63 percent — with no model card attached. Which model, which innings state, whether Duckworth-Lewis was incorporated: all absent.

When a number detaches from its assumptions and travels alone, it stops being information and becomes propaganda.

No evidence is not the same as no match

This is where my real fear sits.

An empty record does not mean the event did not happen. It means we have no document of it. A system that cannot hold that distinction will make two errors at once: it will look away where a problem genuinely exists, and it will manufacture suspicion where none exists.

Cricket's integrity history shows how costly this is. Hansie Cronje in 2026, the 2026 spot-fixing case, the 2026 IPL betting scandal — in each there was a period when people knew and the record did not. Treating an empty record as neutrality leaves those periods permanently dark.

To be explicit: those cases are my reference framework, not allegations. Nothing here accuses any specific match or player. But the methodological lesson survives. Zero information and proof of innocence are not the same thing, and the distance between them is the real risk.

One rule, immediate effect

The simplest fix is a schema rule: reject any record whose information-point count is zero. Do not let it enter the system.

One rule. No new model, no budget. A single condition — zero information points, record discarded. Silent failures become loud ones.

The next step is an input-integrity pre-flight: before analysis begins, build a table checking title, source, information points, time sensitivity. If three or more of ten fields are empty, halt.

This works outside cricket, but cricket needs it most, because its data arrives from broadcasters, boards, franchises and scorecard sites with no shared standard.

Where I distrust my own method

The scoreboard is not my enemy. Splitting process from result carries a seductive trap: process smugness. Holding a ledger makes it easy to declare that the winning team did not deserve it. That is dangerous, because results feed back into process. Confidence is a performance variable. A won series changes a bowler's length in the next one.

A second trap is scalability hubris. I built a ledger in Sylhet, but it does not copy directly to other markets. Data availability differs, logger training costs differ, cricket culture differs. A model that works in England distorts a domestic tournament when transplanted blindly.

A third is market signal overreach. Market-implied probability and my process model are separate instruments. The market talks in money; the model talks in runs. Blend them and analysis becomes betting, and the analysis loses its value.

Silence is also data

I have logged many matches in empty stadiums. Silence is not zero. It is a separate measurement requiring its own interpretation. Without crowds, home advantage shrinks, umpiring pressure eases, and expected home runs fall. To know that number, you must log the emptiness as data rather than discard it as failure. A desk that throws away null records loses this permanently.

What to watch next

Three things this season. First, whether every public cricket graphic carries a model card — I will not trust a number that cannot explain its assumptions, however large the channel. Second, whether auction prices are printed beside sample sizes; if a young player's fee doubles and nobody states how many innings he has played, the fee is an event, not information. Third, and most importantly, how many null records enter our own pipeline. That is not a rival's fault. That is our audit log.

I do not chase results; I audit the process until it confesses. Today it confessed this: when the ledger holds an empty row, nobody can write that match's result. Only the scoreboard can.