Eight Dimensions, Zero Facts: Auditing the Silence Inside a Cricket Data Pipeline

On a Wednesday afternoon in the Bengaluru office, a report opened on my screen. It had eight dimensions. Under each dimension there was a table, each table had five to eight cells, and each cell carried a tidy entry. The first dimension was format and match analysis. Its first cell read: 'N/A — insufficient information, cannot assess.' The second cell said the same. So did the third. I scrolled down. Eight dimensions completed, six risk categories, a five-point governance checklist, an eight-segment transmission map. And in the entire document there was not one cricket fact — no team, no player, no scorecard, no venue, no date. In one place only, inside the domain label, two words flickered: cricket_asia.

This was the report my own desk had produced. And it was also the report that any man sitting upstairs could read and conclude that the work had been done. The formatting was clean. The tables had borders. The categories sat where categories are supposed to sit. What was missing was not merely the match. What was missing was cricket itself.

I opened the venue ledger, and this time the par line did not breathe.

I thought of the summer of 2026, when I rebuilt every IPL venue's par score from 2026–17 ball-by-ball data — Chinnaswamy 189, band 176 to 197; Wankhede 172; Hyderabad 165. That season my mean absolute error on first-innings forecasts fell from 18 runs to 11. Since then every preview opens with a par score, a range and a stated confidence level. Today, in that same ledger, the par-score cell was empty because the venue itself was unknown. And when the par line is absent, everything downstream — the valuation of the innings, the context for the bowling, the arithmetic of the chase — becomes meaningless on its own.

This piece is an accounting of that emptiness. It is not a match story, because there was no match in front of me. It is the story of the moment a data desk caught hold of its own report and asked — this thing that is so neatly arranged, what exactly is in it?

The ledger complete, the page blank

A cricket data desk works in three layers. The first is raw material — scorecard, over-by-over log, pitch report, weather line, toss, DRS decisions. The second is what we call information points — the five or six sentences extracted from the raw material that can carry the spine of a story: who won, at which over the game turned, which rule changed mid-match, which player's claim collides with which number. The third layer is analysis — distributing those points across eight boxes: format, player, team, league, governance, risk, narrative and industry.

The report that stopped me was a third-layer document, and its second layer had arrived completely empty. The list of information points contained nothing. The list of entities contained no names. Title unavailable, source unavailable, article type 'unclassified', time sensitivity never assessed. The bridge over which a story was supposed to reach the third layer had collapsed — and the third layer had fitted a clean, well-painted railing onto the fallen bridge.

In my trade there is one name for this: a false negative. A 'no' that is not actually a 'no' but a 'don't know', and which has been reported in a way that reads like 'no'. It is the same error that cost me three straight bets in Russia in 2026. Before that World Cup my penalty base rate rested on four tournaments: 0.26 penalties per match, 13 awarded in Brazil 2026. Russia produced 29. I refused to re-base the rate mid-tournament — that is my habit; I need thirty matches before I move a number — and lost three straight under bets in the round of sixteen. At the time I thought I was being honest with the data. Later I understood that I had simply failed to recognise a changed regime in the refereeing environment. After that day I built the 'regime flag': a separate switch for when the rule set changes, kept apart from the noise.

Today's report carried the same disease from the opposite direction. In Russia I had misread data that existed. This time the report had presented the absence of data as 'no risk found'. The difference between the two errors is that the first costs money and the second costs trust. And the bill for trust arrives late — after somebody upstairs has already made a decision on the strength of that document.

Eight boxes, forty-one cells, and one lonely word

I read the report again, this time in order. The first dimension was format, and it said: Test, ODI, T20 or The Hundred — undeterminable. This is the most expensive ignorance available, because format is the door without which no cricket number holds meaning. A strike rate of 135 is extraordinary for a Test opener and below ordinary for a T20 finisher. One number, two universes. This is why I never pool Test, ODI and T20 metrics in a single analysis — average, strike rate and economy carry different units, and merging them is not analysis but a bookkeeping error.

The second dimension named no player, so no role could be fixed. And without a role there is no benchmark — an opener's benchmark is not a finisher's benchmark, and a Test new-ball bowler's economy target is not a death-overs yorker specialist's. The third dimension had no teams, so not one of the three ICC tables could be opened — and it is worth remembering that Test, ODI and T20I rankings are three separate documents, not one. The home-away differential, the largest structural variable in cricket, cannot be extracted without knowing the host nation.

The fourth dimension was league and commercial ecosystem. There was not a single rupee in it — no broadcast rights, no franchise valuation, no auction price. Here my oldest habit surfaced. From the 2026 auction beat onward I have priced players by matches actually available rather than by highlight: retention arithmetic, purse efficiency, injury history, workload. All of it is a single question in disguise — how many matches of the season will this player genuinely be available for? And because not one auction figure appeared in this report, the whole discipline of this dimension — separating commercial value from sporting value — sat there untested. A high bid is not international strength; that is among my harsher lessons.

The fifth dimension was governance, and its checklist had five boxes: revenue distribution, playing-rule controversies, integrity, eligibility and selection, political and geopolitical factors. All five returned the same answer — insufficient information. This is where I had to be most careful, because integrity matters in cricket are the most source-sensitive of all. A board statement and an aggregator's copy are not the same sentence, and here the source field itself was empty. The 2026 Cronje affair, the 2026 Pakistan spot-fixing case, the 2026 IPL — all of them stayed in my reference frame, and not one triggered.

The sixth dimension was risk, and that is where the biggest decision had to be taken. Six categories — sporting, personnel, commercial, rules and integrity, public opinion, systemic — and 'N/A' against all six. The report declined to issue an overall risk rating, and I think the refusal was correct. Had it written 'Low', that would have been a straight false negative — saying the story is low-risk when the truth is that the entire risk surface is invisible. The one thing that could be rated with confidence was analytical risk, and that was High: the risk that a reader of this document mistakes structural completeness for substantive coverage.

The seventh dimension was public narrative. There was no narrative, so no heat-cycle position could be assigned — germination, acceleration, climax or backlash. This was that dimension's costliest omission, because the widest gap between expectation and reality opens precisely at the climax and overhype stage. The eighth dimension was industry transmission, and its map — youth development to national teams to broadcast and derivative markets — was drawn in full, but every arrow was inert, because transmission requires a named shock: a rights deal, a star transfer, an Olympic or calendar decision. There was no shock, so the map was only a picture.

Eight dimensions, more than forty-eight cells, and one single word in the whole document that looked like cricket — cricket_asia.

The lie of the risk rating: 'low' versus 'unscreenable'

One of the basic conditions of my work, one I repeat in writing, is this: the absence of an integrity signal in a report does not mean the absence of integrity risk. It means the absence of evidence. In cricket that distinction matters more than anywhere, because lower-tier leagues have historically been fixing-sensitive and anomalous odds movement is the first signal that never shows up in a scorecard. When a payload is so empty that odds movement is not even mentioned, the honest statement is — screening did not happen. 'Came out clean' is wrong. 'Was not checked' is right.

The same applies to injury and workload. In that report the personnel-risk cell was empty, and a reader could easily have concluded that no injury news existed. The truth was that no player's name had been entered at all. This is the difference between a wicketkeeper and a scorecard: if the keeper did not appeal, it does not follow that there was no edge — it follows that he did not hear it or did not say it. The biggest mistake a data desk makes is to mistake its own silence for a certificate of clearance.

The most dangerous cell in a risk matrix is not the one marked 'high'. It is the one marked 'N/A' that the reader reads as 'fine'.

Which is why the one thing that report did well was to refuse an overall rating. I know the desk did not like the refusal. The desk always wants a direction — the man upstairs asks 'so what is it?', and 'don't know' does not sound like an answer. In 2026, when Mumbai Indians had made 129/8 and my model had them at 58% at the innings break while the desk wanted them pushed to 70%, I refused and logged the divergence. Mumbai won by one run. That day I learned that a model's job is not to win, it is to stay honest. Today the same lesson applied in reverse — the model wanted a number like 70%, and it did not even have a match to work with.

cricket_asia: a label that drifted off the schema

There was exactly one thing in the whole document that I could call a signal, and it was the domain label: cricket_asia. I read it two ways. First: the source article plausibly concerns Asian cricket — an ACC or Asia Cup fixture, an Asian bilateral series, or a subcontinental T20 league. Second: it is a labelling-pipeline fault. The second possibility struck me as stronger, because the label had drifted from the schema's expected value of 'Cricket'. A non-conforming label accompanied by empty information points is not two separate problems — it is two symptoms of the same illness.

I did not hang any conclusion on that signal, and that was a deliberate decision. The Asia Cup has been played in ODI and T20I formats; Asian domestic leagues are T20; and within the subcontinent, India, Pakistan, Sri Lanka, Bangladesh, Afghanistan and Nepal have profiles so different that talking about any one team on the strength of 'Asian cricket' would be dishonest. India has a full employment pyramid; Nepal has a growing limited-overs programme; Afghanistan has a spin-based model. Collapsing all of them into one label and then issuing a tier claim would be the same error as printing 'batting-friendly surface' in a preview. I gave up that habit.

But the label did do one job: it told us which part of the pipeline broke. If the information points had come back empty while the label conformed to 'Cricket', we could have said the article contained nothing. When the label also arrives malformed, the conclusion changes — it becomes more rational to believe the article contained something and the extraction machinery failed to lift it. Those are two different worlds. In the first, the fault lies with the article; in the second, with our own tool. And twenty years tell me that in ninety per cent of cases people blame the data when the query was wrong.

The non-comparability of formats: the door that cannot be crossed

The hardest rule of this entire framework sits at the very top — the non-comparability of formats. Test, ODI and T20 numbers are not translations of one another. In Test cricket, average is the question of how long you lasted; in T20, average is close to meaningless and strike rate is everything. ODI is the middle house where the weight of the two shifts with the situation. In bowling, economy is a strange measure in Tests, useful in ODIs, and most meaningful in T20 death overs — and even there only when accompanied by wicket context.

So when someone tells me a player 'averages 45', my first question is: in which format, in which role, across how many matches. Combining three formats into a single 'career' number is the exact confusion I have never committed in my writing. This is why, if the first dimension of that report is empty, the emptiness of the other seven is not their failure but their consequence. No format, no player benchmark; no role, no benchmark; no team, no ranking; no venue, no home-away differential. One brick came out and the whole wall stayed standing, with the ground removed from beneath it.

One point deserves recording here, because data desks routinely forget it: when the rule set changes, rates must be re-based. The revised DLS target, DRS 'umpire's call', powerplay overs, the difference between old and new regulations — all of these belong to the regime flag I built after Russia. At the 2026 Asia Cup in Dubai, dew made the chase decisive and 2026 toss data was worthless there. I wrote that the rate was 'built on X matches under rule set Y'. That sentence should have appeared in this report too, and it appeared in no cell, because there was no rate to qualify.

The minimum viable payload

Now the part that is the real product of this whole episode. If the desk is to run this analysis again, it must return at least five things, and these must be treated not as a 'nice to have' list but as a 'nothing works without this' list.

First: the article's title, its publisher or outlet, and its publication date. This unlocks at least two dimensions — narrative analysis and the freshness rating. Without a date, no cricket report can be located as recent or old, and without that distinction the regime flag cannot be set.

Second: at least five information points, each tied to a specific sentence in the source. The word 'tied' matters. Information points you cannot walk back to a sentence in the source are not facts but inferences — and seating an inference in a fact's chair is precisely the sin this whole framework exists to prevent.

Third: format and match nature. Test, ODI, T20, The Hundred — and bilateral, ICC event, league or warm-up. This is non-negotiable. Proceeding without it is like announcing a par score without looking at the pitch.

Fourth: named entities — teams, players, coaches, league, venue, umpires if they are in dispute.

Fifth: a source-quality grade — official board or ICC, then ESPNcricinfo or Cricbuzz, then general media, and aggregators at the bottom. In cricket this grade matters most because governance and auction claims are presented entirely differently across those three tiers.

Without these five, what gets built is not analysis but an empty container — and the greatest danger of an empty container is that it looks full.

I have a rule in my own writing that the first line of any report should state which variable I trust least. Since 2026 every preview carries a par score, a band and a confidence level. Today that rule took its second form — when there is no par score at all, the variable I trust least is the report itself.

Contrarian: the bug is not the hard part, the acceptance is

Now the thing that will be said least in all of this.

Almost everyone who reads about this episode will offer the same opinion: there is a bug in the pipeline, fix it, run it again, move on. I decline. The bug is the least interesting part of the story. Bugs land daily in analytics and get fixed daily. What does not get fixed is the process that judged this report fit to leave the desk.

Notice that the report never hid its own nullity. Every cell said plainly — insufficient information. The problem was not that a lie was told. The problem was that the truth was packaged so well that it began to look like a finding. Eight dimensions, six risk categories, a five-point checklist, an eight-segment transmission map. That scaffolding is itself a statement, and the statement is: we did the work. When the truth was that the thing the work was supposed to be done on never reached the desk.

The feeling of complete structure is not proof of complete coverage — and the data industry's biggest illness is that it sells the two as one report.

This is not new to me. I have watched it repeatedly in cricket. In the Ranji Trophy knockouts there always arrives a day when a match report gets filed because the slot exists, and the report contains nothing beyond the scorecard — no bowling story, no accounting of the pitch, nothing about what changed on the third day. The slot was filled; coverage did not happen. The same thing showed up in the gap between broadcast promise and delivery in the early WPL seasons — a full day in the announcement, only the score in the delivery. And the same thing came back in that report, this time wearing the shape of a spreadsheet.

My second disagreement is with the easiest conclusion available — that the article contained nothing. I refuse it. The conclusion 'the article was empty' and the conclusion 'our machine failed to extract it' are worlds apart, and there is only one way to settle which is right: look at that malformed domain label. The schema asked for 'Cricket'; the answer came back 'cricket_asia'. A small thing, and it pulled the entire report from 'article at fault' to 'tool at fault'. How would a pipeline that cannot even return a conforming label return clean information points?

And a third disagreement, the most uncomfortable one. I suspect this episode is not an accident. If it were an accident, someone would have hit a wall at some turn and said — stop, something is missing here. That did not happen, because nothing in the report looked missing. In a desk where a full, furnished, bordered deliverable counts as 'delivery' and a returned 'we don't know' does not, these reports will keep being produced — even after the bug is fixed.

This is where one scene comes back to me that sits outside the ledger, and I will place it once, here. I learned to price silence when the stadiums emptied — that afternoon at a domestic match where eleven players a side were on the field, the umpires were in position, there was nobody in the commentary box and three people in the stands. I wrote in my notebook that day that the cheapest part of the sport is not what gets sold but what never gets counted. Today I am applying that line to my own report. The most expensive part of that document was its empty cells, and they were the least counted.

Takeaway: three triggers for next time

I end not with a conclusion but with conditions. When the run is repeated, there are three triggers I will treat as the publication threshold, and I have fixed that threshold before reporting begins so that there is no argument later.

First trigger: at least three discrete entries in the information-point list and at least one named entity. Below three, I will still write, but I will say plainly that this is a provisional reading — incomplete, preliminary and falsifiable. I know this makes me slow. Being slow is still better than being wrong in a hurry.

Second trigger: schema conformance of the domain label. Until the label reads 'Cricket', I will not trust any dimension beneath it, however populated those cells appear. It is the same caution I applied when the desk asked me to re-base a rate mid-tournament.

Third trigger: source and date populated. This is the dullest condition and the most necessary. Governance and auction claims cannot be graded without both, and ungraded claims are only noise.

I trust the row of numbers more than the roar of the crowd — but if there is no number in the row at all, the question of trust does not arise, and that is the whole lesson of this story.

Over the next four weeks I will watch two signals from that desk: whether the payload, on its next arrival, brings at least one name with it, and whether the label returns to schema. If both happen, we can celebrate a fixed bug. If only the second happens and not the first, we should understand that the real problem was never in the article — it was in how we read.

The outlier is not noise; it is a door left ajar by the model. This time nobody came through that door — there was only a clean, well-formatted report pointing at its own emptiness while every reader took it for approval. The question is not what was in that report. The question is the eye we read it with — and next time a fully furnished document lands on the desk, whether we look for the scorecard first or the border.