The Empty Stage-1: When a Data Pipeline Becomes Cricket Analysis's Biggest Error

One cricket analysis pipeline's Stage-1 came back empty. No title, no source, no timestamp, not a single fact. The system worked exactly as designed — and that is the problem.

In cricket we know this position well. When the Snickometer shows no spike on a DRS review, the third umpire does not say 'not out' — he says 'inconclusive'. The difference is that in cricket, inconclusive has a protocol: the original call stands. Analytics pipelines have no such backstop. An empty output means every downstream layer — Stage-2, Stage-3, the final report — will stand on absent data.

What an empty Stage-1 actually tells you

On the surface this is a technical failure. Deeper down it is a diagnostic signal, and the signal points to one of three distinct illnesses. First: the crawler never reached the page — paywall or robot wall. Second: the crawler arrived, retrieved the HTML, but the content extractor's selector slipped — the system opened the right room, found the right door, and put the key in the wrong lock. Third: the content arrived but the model declined to classify it as 'analysable'.

Each has a different cure. The first needs a source access change, the second a selector maintenance pass, the third a classification prompt review. One symptom, three diseases. A team that reruns Stage-1 without a diagnosis will get the same empty output two times out of three — because it did not treat anything, it merely re-rolled the dice.

The real sin of a data pipeline: the silent failure

The best failure is the one that makes noise. The most dangerous is the one that quietly wears the face of success.

A stage that returns JSON — empty but valid — shows green on every monitoring dashboard. Schema validation passes. No alert fires. And the next layer sits down to write a report whose foundation is sand.

The precise analogue exists in cricket scoring: a ball recorded as a dot instead of a bye. The scorecard will look mathematically correct, the overs will complete, the totals will tally. Only when the match turns on that single run does anyone discover when the system lied.

In our work the lie surfaces the moment you question the report built from Stage-1's empty box. Asking the question does not mean the report is bad — it means only the person who refuses to assume the system is 'correct' is actually auditing it. I don't assume. I audit. And the first rule of the audit is never read an empty ledger as zero — an empty ledger means 'nothing was entered here', which is an entirely different claim.

What the data layer is actually for in cricket analysis

Even if the pipeline had run cleanly, a question would remain: what does this data actually tell us about the cricket?

From years of watching matches I have found that cricket's biggest lie hides inside the word 'control'. I do not trust possession until I know which phase it was earned in. Forty runs in the powerplay and forty runs in the death overs are not the same number — nor is the field geometry behind them the same. When I build a powerplay ring map, I often find that the team appearing to 'dominate' is actually converting balls outside the thirty-yard circle into singles while the required rate climbs on the scoreboard.

That is why I never confine statistics to the broadcast camera. While the camera holds a close-up of the batsman's trigger movement, the field geometry is playing out off-screen. It is not the bowler's length but the fielding position that decides a chase — and that position escapes the frame. The half-space ledger starts where the broadcast camera stops.

The same applies to a data pipeline. If your pipeline has stalled at an empty Stage-1, you are holding the same angle that shows the batsman's feet and not the fielder who shifted to third man. You have ball data but not field data. You know how many deliveries were bowled, but not how many of them landed where the batsman could not afford them to.

How to enter 'empty' into the model

I have a principle that came out of football analysis and translates into cricket: the crowd-context variable. When stadiums were empty, pressure did not disappear — it changed where pressure could be heard.

An empty Stage-1 output should be treated the same way. Read the empty output not as 'no data' but as a distinct variable. The data is absent — that itself is a data point. The question is not 'why did the information not arrive' but 'is the source this information was supposed to arrive from trustworthy for our system at all'.

If the answer is no, the whole plan changes. Because a report built on that source will carry its original limit — however polished the report looks.

I take the same lesson from auction planning. When an IPL team's purse looks empty at auction, the fool says 'they had no choice'. The careful analyst asks: 'which retention option was live that they declined?' An empty purse is not an outcome, it is an accounting of decisions. An empty Stage-1 is the same — not a result, but the ledger of choices made by the crawler, the selector and the classifier.

So when Stage-1 returns empty I ask for three things — no, one request in three forms: show me the raw HTML (did the crawler arrive), show me the extractor's selector path (where did it slip), show me the classification log (why did the model decline). Without those three, 'rerun' is not a solution, it is preparation for another empty output.

The point where the model bends

There comes a turn where the auditor has to accept that the model is incomplete — not wrong.

Here the model did the right thing: it did not fabricate data, did not guess, did not invent a title. It said honestly: I have nothing. That is not a 'bad' output, it is an honest one. And honest emptiness is always better than dishonest completeness.

But the question arises right there: is the audit's job finished? No. The auditor's job is not to report that 'the system is empty' — the auditor's job is to report 'why it is empty, and what will change so it is not empty next time'. Reporting emptiness and raising your hands is not an analyst — it is a clerk.

We know this well in cricket. A batsman faces twenty balls for three runs and is out, and the commentator says 'he was not in rhythm'. I say: three off twenty means he played seventeen balls on which no runs came — the question is not whether he was in rhythm, the question is what field setting those seventeen balls were bowled to. The answer is there, away from the camera, in the field map.

The empty Stage-1 is the same: zero facts returned is not a number, it is those seventeen balls on which the system did nothing. The diagnosis is there, not in the rerun button.

Make the system's limit part of the system

My suggestion is simple, though hard to implement: add an 'absence field' to every stage of the pipeline.

When Stage-1 runs, record this too: how many sources were checked, how many failed, which one failed and why. When the output is empty, the system will have a reason — 'paywall', 'selector mismatch', 'classification reject'. That reason opens the route for the next attempt.

It is the same reasoning I use in tournament planning. Calendar first, team second. Because the calendar is the constraint that determines who can play and who cannot. If you do not look at the calendar's constraint before picking the team, your team will look strong on paper and tired on the field.

The same applies to a data pipeline. If you do not look at the source's constraint before Stage-1, your report will look like analysis on paper and read as empty.

What to watch next

The next time a pipeline returns empty, do not ask first 'how do we fill it'. Ask first: 'can we even read this source?'

If the answer is no, choose a different source. If the answer is yes, fix the selector. If the answer is maybe, run the data model with that same maybe — and write the maybe into the report.

Because cricket's best report is not the one that knows everything. The best report is the one that states what it does not know — and what can be drawn from that absence.

Stage-1 came back empty. The question is not how to fill it. The question is whether that emptiness is telling us something a full report never shows.