The N/A Scorecard: The Most Dangerous Error in Cricket Data Is the One That Never Shows Up on Screen

The file arrived at 11:42 pm: eight tabs, eight sections — format and match analysis, player technique, team landscape, league economics, governance, risk matrix, public narrative, industry transmission. Every section carried the same line: 'insufficient information, cannot assess.' No model crashed. The machine did not answer wrongly; it asked the question eight times.

That is the cold discomfort. After twelve years on a pricing desk I know a blank field is not a zero, but nearly every machine we run reads it as one. Cricket has an old name for this — the wicket the scorecard never logged, the bye that crossed the rope while the scorer's pen stopped. At season's end nobody asks why the pen stopped that day.

Our pipeline is a two-storey building, and the profession has stopped counting the stairs. The ground floor is description: scorer entries, ball-by-ball tagging, wicket type, catch distance, field placement, DRS decisions and reviews, toss, dew, pitch age. The first floor is analysis: model, price, win probability, verdict. The whole desk sits upstairs and audits upstairs.

The print era kept the building cheap because the ground floor had little to count: runs, wickets, averages, catches. After 2026 the IPL made the ball-by-ball corpus free and dense, and one thing happened quietly — whatever data was easily available became the beat. Tracking arrived after 2026; I spent the winter of 2026 logging ball-tracking for every wicket. The WPL dataset arrived in 2026, and the 2026 home World Cup win supplied a sample the commentary box acknowledged faster than it modelled.

One finding belongs here as a finding, not a complaint: Indian cricket's densest dataset is the IPL, its thinnest is the Ranji Trophy. Where the game is played most, it is described least. Regional-language coverage and pre-2026 women's cricket share that thinness, and the gap enters our conclusions silently — we simply do not write about what we cannot count, so the map quietly shrinks to the beat.

A prior, with its window: of the desk disputes I have been in since 2026, roughly three in four traced to an input field rather than a model, give or take ten — my own logbook estimate, not a survey. When I write 'insufficient information' I do not mean a negative result. I mean the model has no right to speak.

21 May 2026, Hyderabad. Mumbai Indians 129/8 at the innings break against Rising Pune Supergiant; my ball-by-ball true-run-value model gave them 34%. Krunal Pandya made 47 off 38, the match finished one run short for Pune, and the model lost. Three weeks went into rebuilding the death-overs wicket-value term. What the audit found is the cleanest evidence for today's subject: the model was not wrong — one term was weighted wrongly. A term error makes noise. A field error stays silent.

In 2026 I retired possession percentage and nobody at the desk noticed. That summer Germany carried a 14% prior to fail the group stage; on 27 June 2026 in Kazan they lost 0-2 to South Korea and finished bottom of Group F. Three days later Spain completed 1,029 passes against Russia and went out on penalties. The lesson transfers directly to cricket: a metric that tracks its own outcome instead of its own process becomes decoration. Field tilt, PPDA and progressive passes per shot replaced possession. In cricket the translation is simple — strike rate and economy say what happened; expected runs per ball says what was attempted. The first is a scorecard, the second an intention.

In May 2026 the Bundesliga returned to silent stadiums. I re-ran the home-advantage regression: a 1.07 multiplier across six IPL seasons and 1,200-plus matches. Behind closed doors it collapsed to 1.00. I published before the September UAE restart and it held across 60 matches. A mandatory context-coefficient paragraph entered every piece — crowds, travel, calendar, pitch age — along with a rule that pre-2026 and post-2026 numbers never share a sentence. Empty stands taught me that atmosphere was a variable I had priced at zero.

19 December 2026, Adelaide. India led by 53 and were bowled out for 36. My in-play model had given them 74% to not lose. Thirty-six all out in forty-five minutes, and every model I trusted went silent. That winter I logged ball-tracking for every wicket and changed the reporting rule I still use: collapses are forensics, never character judgments. Where the ball pitched, where the bat went, which frame the footwork stalled in. One innings is not a large enough event to explain a series; if the sample is small I write that it is small and decline the verdict.

12 June 2026, Parken. Denmark–Finland on my second screen, Christian Eriksen down in the 43rd minute. I pulled every Denmark-linked market within ninety seconds and drafted a twelve-clause void protocol the firm still uses. Seven weeks later Tokyo handed me the opposite problem: 3x3 basketball and skateboarding debuting, samples of eight. I built a Bayesian shrinkage model, priced India's men's hockey bronze at 4.5, got the 5-4 win over Germany on 5 August, and published the uncertainty bands beside the pick.

22 November 2026, Lusail. Argentina 1-0 up, my in-play model at 94%, Saudi Arabia winning 2-1. Within forty hours I published not an excuse but a variance ledger — calibration error measured across 5,000 historical matches. The public variance ledger began that week: every flagged edge, win or lose, scored quarterly.

Put these together and the pattern is plain. Predictive error makes noise — it reaches the scoreboard, the timeline, the editor's phone. Descriptive error stays silent, because it never reaches the scoreboard at all. When a side wins from 34%, four people ask why the model was wrong; nobody asks how many of the 47 fields fed into it were actually blank, or who filled them by guess.

Tournament cycles compress emotion, which makes the question sharper. Flags and storylines arrive first, description later and often too late. I read selection as version control: India's real problem is not talent but the machinery that ranks it; the hardest team in world sport is India's second XI, and the newest unassimilated dataset is the women's game. When someone quotes an average, my first question is not whether it is correct but how many of those innings were in this format, how many at home, and how many had information that was never logged. There is a version of this game that only the numbers saw, and I owe it a hearing.

The industry's first reflex is to blame the model, because a model sits on a screen and can be rewritten. Correlation and causation blur here: a wrong output traces upstream to the stairs, not the first floor. My logbook holds cases where one incomplete entry made a team look two percentage points stronger for four weeks and nobody noticed, because the result matched the forecast. A correct result hides a wrong input. That is the quietest dishonesty in this profession, and it comes from haste, not conspiracy.

The over-correction is just as dangerous. Not every blank means zero; sometimes 'unknown' is the honest answer and must stay unknown. The real risk is the imputation that does not announce itself — the meeting-room line 'let's adjust for it', with the adjustment method recorded nowhere. Editorial pressure pushes that way: printing 'insufficient information' on deadline is courage, printing a decimal is a career. Two editors dropped me over the habit; the readers who stayed were the ones who paid.

On the commentary box, care is required. When a wrong number goes to air, the temptation is to discuss the person. That is the lazy route. What broke is not someone's understanding but someone's field population — an unversioned scorecard still climbing on air. Audit the claim, not the speaker; a wrong number is a dataset problem and its fix is a better number, not a better sneer. The day I retired a metric and the desk said nothing, I had two things I could write: the metric, or the story of being ignored. I chose the metric. The correction is the story, not the silence that met it.

Next round I will ask for something that is not a model: a third column beside every published number, the way an error term sits beside a coefficient — field completeness. How many fields returned blank, who filled them, by what rule, and how much uncertainty that fill carries. Readers will find it unattractive and ignore it for a year, exactly as they ignored the void protocol, and then quote it in the first dispute.

So my question has changed. I no longer ask only what the win probability is. I ask how many entries in this model returned 'insufficient information', and who filled them. The day that question sits on the scoreboard, cricket's data will finally look at its own real error. Until then, behind every decimal stands a blank field, and it is silent.

Every number I trust carries a scar from a season I got wrong. This piece is the accounting of those scars — and the next scar may be the file that came back empty last night.