No-Signal: The Report That Passes the Schema, Not the Truth

At 3:40 p.m. the phone would not stop, and neither would my hands. A file landed on the desk: a version number, a domain label, eight analytical pillars, every field populated. The schema was immaculate. Then I looked inside the fields. Title: not applicable. Source: not applicable. Article type: unclassified. Information points: empty list. Core viewpoints: empty list. Author stance: not applicable. The file had arrived. Nothing inside it had.

Across the desk someone was writing a score, a cup was knocking against a saucer. I did not record that sound, because the file had not yet told me what it was.

I wrote it in the third notebook, the one for things I could not prove yet. Before I put the pen down it was clear this was not a cricket story. It was a photograph of a system dressing its own failure in a valid format and delivering it anyway. On a cricket desk we can recognise a blank scorecard at ten paces. We have not yet built the habit of recognising a blank data report.

Over the past few years cricket newsgathering has split into two stages. Stage one pulls facts out of text: headline, source, information points, named entities. Stage two builds analysis on those information points: format, players, teams, leagues, governance, risk, public sentiment. If stage one returns empty, stage two has nothing to hold.

That is exactly what happened here, with a twist. Stage one raised no error. It produced a fully valid structure in which every field held the words "not applicable". The subtler detail is that the instruction text survived. Lines like "identify from the information points above" and "judge from the source fields of the information points" are still there, while the information points they refer to do not exist. The work stopped before results were produced; only the mould was cast.

Stage two is not an easy job, because its entire method stands on stage one's information points. Without a format anchor, form numbers mean nothing: a strike rate of 135 is strong in ODI cricket, middling for a T20 finisher, close to irrelevant in a Test. Without a named team, the home-away differential cannot be estimated, and that differential is the single largest moderator of results in the sport. Without a named player, the age curve, injury history and form deviation cannot be computed at all. Zero information points means all eight pillars close at once.

A second signal sits in the domain label. It reads "cricket_world", where the agreed taxonomy says "cricket". A small difference, except routing, retrieval and dashboards all run on that label. When the label drifts, the story lands in the wrong room, and a story in the wrong room is not easily brought back.

The real finding in this report is not about cricket, it is about process. Structural validity and informational completeness are not the same thing, and most automated validation measures the first and never the second. A monitor that only asks whether the schema is intact would have passed this record. Inside it there is not one verifiable claim.

The empty record also tells us, indirectly, what a proper information point looks like. "The team played well" is not an information point, it is an opinion. "In the 47th over, the number eleven batter was sent in" is one, because its source, its timing and the direction of the decision can all be checked. A pipeline that cannot make that distinction drops opinion and fact into the same basket, and afterwards nobody can recover which was which.

I have a comparison from my own habit. Since August 2026 I have kept a two-page set-piece and role chart for every match I watch, by hand, in pen. It is append-only: I never rub out the old page, I only add a line. A number written in pencil can be quietly revised later to suit; a number in ink cannot. The blockchain-style idea of a provenance chain is only the technical form of that simple principle: each step leaves an immutable mark of its input and its output. Text ingestion, information-point extraction, entity identification, source-quality grading — with a mark on all four, the step where the zero entered is found in minutes. Today it takes a human eye, and the human eye is not always at the desk on time.

The biggest risk in this empty record is not a cricket risk at all, it is a pipeline risk: taking the next decision on top of a zero input. Likelihood is high and impact is high, because the machine looks legitimate. Anyone who reads the file as "analysis complete, result nil" commits the error twice — once in the empty data, once in seating that emptiness inside a decision.

The second risk is human: the temptation to write something under pressure. Given an empty input, the easiest work available is to manufacture analysis that sounds credible. In January 2026 I spent five weeks around a club's medical and recruitment staff. I confirmed a name at 3:40 p.m. and held the story six hours, because two independent sources had to be reconciled and the club had to speak to the player's family. Filing nine hours early with a hole in it is worse than filing nine hours late with the hole closed. This blank report says the same thing in another language: the greater the urge to fill the empty field, the higher the price of getting it wrong.

The third risk is batch contamination. If this record arrived inside a batch, other records in that batch may carry the same signature — placeholder title, empty information points, instruction text left intact. A silent failure rarely travels alone; one broken record is often ten.

The fourth risk is the taxonomy drift I have already described. The fifth is the slyest: reading silence as safety. This report raises no red flag on integrity, injury or governance, and that is not evidence that no such risk exists — it is evidence that no source existed. An empty field is never a clearance.

The fix is technical and cheap. A completeness gate: if the information-point count is under three, or the title or source is null, quarantine the record immediately as NO-SIGNAL and send it no further. Then a root-cause check — did the extraction stage receive any text at all, or was the text stripped before parsing. Then a scan of the rest of the batch against the same signature. The cost of those three steps is close to nothing. The gain is restored trust in every future verdict of "nil result".

The record has one unexpected use as well: it is a ready-made regression test for the build log. Whoever installs the completeness gate next can run this exact input through it and ask whether the gate actually stops it. The failure signature here is legible, which makes the test repeatable.

An outsider reading this empty report would say the system failed. The opposite reading is closer to the truth: the system worked, because it invented nothing. A model that filled the blanks with inference would have looked more useful and would have eaten the floor out of our trade. Writing "not applicable" in an empty field is an honest act, of the same family as keeping the unproven thing in a separate notebook.

So the real failure is not mechanical, it is cultural. Our desks still carry the assumption that a file arriving means a file filled. The training ground told me the truth three days before the transfer market did, because I went there for the habit, not for the result. Data asks the same discipline: stop weighing a report by how it looks.

In a transfer window we suffer the opposite problem — too much signal, too little evidence — and the work there is filtering rumour. Here the problem is inverted: zero signal, delivered in the wrapping of signal. The same instrument serves both cases. Separate the tiers of evidence, and never present the unproven as proven.

The next internal signal is clear enough. Who will count how many more files in this batch look valid and read empty? Until the completeness gate is installed, every green tick deserves a second look. One line went into the notebook today. Another goes in on the day somebody asks the only question that matters: the schema passed, but how many information points did it carry?