Eight Dimensions, Eleven Tables, Zero Information Points: Auditing a Cricket Data Pipeline's Silent Failure

Zero.

The report arrived on my desk fully dressed: eight analytical dimensions, eleven tables, a six-row risk matrix, a four-column star rating, and a Comprehensive Assessment at the end. On paper it carried everything a disciplined cricket report should. On page one I found the number I have trusted for forty-six years, except this time it was sitting in the wrong place: Information Points — zero.

This was not a half-finished note. It was a frame that satisfied every formal rule and lost every piece of content. Stage-1 breaks a source article into information points, core viewpoints, entities and metadata; Stage-2 raises the analytical scaffold on that frame. Stage-2 cannot exceed the ceiling Stage-1 built. Here the ceiling was on the floor and eight storeys had been erected on top of it.

The Eden Gardens ledger taught me to reconcile before I speak. In 2026 I keyed the first manual ball-by-ball feed — 37 matches, roughly 22,400 deliveries — and on the night of 13 March in Kolkata I opened a private dBase book: every ball, run rate, dot ball, phase split. Four years later that book isolated fourteen matches in which a side's run rate between overs 15 and 40 fell 1.8 below its own baseline while wicket loss stayed flat. The forty-one-page printout ran under someone else's byline. The lesson held: the number is the deliverable, not the name above it.

So when this report came in, I read it as a ledger, not as news.

The fields that survived are the loudest clue. The domain label was populated — cricket_world. The article type was there — Unclassified. Additional notes were present. What died were the fields that needed content: information points, entities, viewpoints, source quality, time sensitivity. That asymmetry is the diagnosis. If the document had stalled somewhere — paywall, truncated text, failed fetch — you would normally salvage partial text and lose the domain label instead. Here it inverted: the schema instantiated, the population step failed silently. That is the most dangerous class of failure, because validators pass it. Emptiness makes no noise.

Which is why I filed this as a process report rather than a cricket report.

Format identification here is a blocking condition, not a descriptive nicety. Test, ODI and T20 do not share scales. Strike rate, economy rate, average, even the meaning of a good performance are format-specific and non-transferable. Where the format is undetermined, powerplay-middle-death splits, new-ball behaviour and home-ground effect cannot be asserted. I did not assume a format, because assuming one is the cardinal sin of this framework: metrics from one format can be pasted onto another to manufacture any story, and that story will be false.

The player dimension is empty for the same reason. No name, no role classification — opener, anchor, finisher, pace, spin, all-rounder. And I did not fill the gap from memory. Forty years of watching leaves plenty in the head, but a figure recalled is not a sourced figure; it is a borrowed number pretending to be part of the report.

The repair order is fixed: entities first, then format, then figures, then narrative. Entities and format alone unlock dimensions one, three and seven qualitatively. Figures come last because the commercial layer is the most expensive to rebuild — rights values, franchise valuations, auction prices do not survive in an analyst's memory with verifiable precision. Every auction has a ceiling; the trick is finding who built it. Here the ceiling itself was missing, and without it any read on the bid is guesswork.

One more thing stung: provenance loss is worse than text loss. Publisher, URL, byline and timestamp were all unavailable. Text can be re-fetched; provenance may not be recoverable. On a day when all four are absent, we cannot even say whose story it is or what date it belongs to. A smaller, more irritating slip sat alongside: the domain label returned cricket_world rather than Cricket. That is taxonomy drift, and downstream routing rules miss cricket content silently because of it.

The risk matrix carried six categories. Five were null. One — conclusions issued on a null evidentiary base — was rated high, and had already been realised. I read that row twice.

The hazard is not inside the report; it is in the report's tailoring. A fully rendered eight-dimension document reads as substantive even when every input field is blank. That cannot be buried; it belongs on the first page. The second counter-intuitive point is that absence is not clearance. No integrity flag appeared in the report, and someone could read that as a clean chit — when the truth is that no field was populated at all, so no flag could have risen. A null means unknown, not clean.

The third is the coldest. Of the eight dimensions, the most inference-heavy — industry transmission, the chain from youth development to broadcast and capital — is exactly where an analyst is most tempted to fill from memory. This report did not fill it, and that is its most honest work. I know my own trap: after years of being right about the counter-intuitive finding, the deviation becomes a habit of voice rather than a result of the ledger. I found the deviation before I found the story — that is my signature and my risk. Which is why a piece must occasionally conclude that the conventional number was correct. Here the conventional instinct — fill the empty space with recall — was wrong, and it should not have been filled.

A table is not a conclusion; it is a trail of evidence. The eleven tables in this report say one thing: the article they were built on was never read.

The work ahead is technical and cheap. Every Stage-1 payload needs a hard validation gate: zero information points means rejection, an alert, no pass-through. Re-extraction should emphasise named entities and format detection; that is the highest-yield repair. And if this payload was part of a batch, the batch needs auditing — silent failures of this kind usually recur.

The question left hanging on my desk: how many reports in the archive look complete and audit nothing? The day that answer arrives in a table, this report will have done its job.