Zero Information Points: When an Analysis Pipeline Fails Silently

2:40 a.m. at the Khulna desk. I opened the file — a Stage-2 brief, an input-integrity notice sitting at the top. I scrolled. Information points: zero. Entities: zero. Article title: N/A. Source: N/A. Time sensitivity: not assessed. Source quality: not assessed. One cell in the entire grid was populated — Domain label: cricket_world.

Since walking into The Daily Star's sports desk in 2026 I have opened a great many files. Scorecards, ball-by-ball traces, xG sheets, PPDA logs. This was the first document I have held where the subject of the analysis was absent while the architecture of the analysis stood fully built. Fifteen tables, eight dimensions, and under every one of them a neatly aligned N/A. The most dangerous failure in a pipeline makes no sound at all.

In 2026, when I left the match-reporting desk in Dhaka and started Expected Truth from Khulna, I wrote one rule on the first page of my notebook: every claim carries its information point beside it. An information point is an atomic, verifiable fact — a score, a spell, a date, a decision. Stage-1 deconstruction does exactly that job: it pulls information points, core viewpoints and named entities out of an article. Stage-2 stands on that foundation. With zero foundation there is no wall, only a picture of a wall.

In that first year I tracked Abahani Limited Dhaka's title run — 34 goals from 26.8 xG, a plus-7.2 overperformance — and logged their PPDA in a 2-0 win over Sheikh Jamal Dhanmondi Club. Four thousand subscribers, one syndication deal. At the 2026 Russia World Cup, Croatia scored 14 goals from 9.6 xG across seven matches, plus 4.4, with Luka Modric covering 72.3 km. My model gave France a 58 percent win probability before the final. That experience turned pre-registration into habit — writing the hypothesis down before kickoff. In 2026, across 83 empty-stadium Bundesliga matches, home teams' points per game fell from 1.54 to 1.21 and average goals from 3.1 to 2.7, while Bayern Munich's PPDA tightened from 7.2 to 6.4. Perfecting the Empty Stadium Index cost me two publication windows. I hired a freelance editor and made deadlines mandatory. The lesson holds: the pursuit of perfection and the absence of data are different problems with the same outcome — no report ships.

Now I walk the evidence chain. Seven of Stage-1's eight mandatory fields are blank. Only cricket_world survives — a label with resolution so low that a match report, an auction story and a boardroom-politics piece all land in the same basket. The classifier returned "Unclassified," which means the text it received was close to empty. So where did it fail: summarisation or extraction? Summary, stance and purpose all going blank together is the signature of an empty or unreadable source document, not of a weak summariser.

Why does format come first? Because format is a gatekeeping variable, not a descriptive one. A strike rate of 145 is elite in ODI middle overs, par for a T20 finisher, irrelevant in Test cricket. Without format the number stays uninterpretable, and an uninterpretable number is not analysis — it is decoration.

The second piece of evidence: no venue, no pitch type, no weather, no dew, no DLS. Home-ground bias, toss luck, DRS controversy — not one variable comes under control. Third: "time sensitivity not assessed" means that if I quoted an ICC ranking, I would not know which dated table it came from. Rankings are format-specific and time-stamped; missing either makes the citation an irresponsible assertion. Fourth: source quality is ungraded, which removes the very instrument that discounts rumour against reporting. Agent-driven pre-auction hype and a paywalled page look equally credible in this state.

Fifth: the commercial layer is dormant. Which league — IPL, BPL, Big Bash, The Hundred, PSL, SA20 — is undetermined. Yet separating commercial value from sporting value is this dimension's central function: a high auction price is a market-clearing outcome, not a performance verdict. With no named transaction that distinction cannot be applied. And to be explicit — I will not drop financial figures in as proxies. Quoting broadcast value or franchise valuation without a source manufactures false precision.

The numbers didn't break the model; they exposed where the model was blind.

Here is the subtlest trap: silence is not clearance. The report does not say "no integrity risk was found." It says no scan took place. A null finding and an extraction failure are not the same thing. The first is a valid negative result from a completed scan; the second is the absence of any scan at all. Miss that distinction and a reader concludes there is no injury risk in the squad, no commercial stress in the league, no transparency problem at the board — when not one of those three questions was ever asked.

The instinctive response is: just process the document again. Correct, and that is the cheap remedy. The real damage sits downstream. The template looks complete; the blank cells are neatly aligned. To an automated consumer this reads as finished analysis. The costliest failure is the one that passes validation.

A second counter-intuitive reading: the defect is sequencing, not content. One sentence survives in the output — "identify entities from the information points above." The template rendered but was never executed against a document. Repeat probability is therefore high; without fixing the handoff, the next document produces the same result. Index overfitting is the neighbouring hazard — adding variables is easy, but without a pre-registered baseline a holdout test means nothing.

Third, and largest: the risk here is epistemic, not cricketing. Under pressure an analyst can invent a plausible team, a plausible player, a plausible strike rate, a firm conclusion. I could have. A fictional series, a believable spell, a decisive closing line — no reader would have caught it. So the correct action is to stop. I don't chase outliers; I follow them until they confess — and this rule runs in reverse here: there is no outlier, only a zero, and a zero cannot be followed, only acknowledged.

My instinct is to draw recovery maps after a shock — checkpoints, failure thresholds, repeatable routines. This file allowed no recovery narrative. No collapse occurred; an absence occurred. Crisis-recovery framing applies when there is evidence of a fall; an absence is a re-reading problem, not a repair problem. The data-supremacy trap is live for the same reason — no player interview, no coach interview, no ground report, not one of triangulation's three legs.

Three signals for the next round. First, re-extraction success: at least one named entity and at least three information points. Second, source-text integrity: a character count near zero, or non-English content, confirming the root cause. Third, format identification: Test, ODI, T20 or The Hundred. Clear those three and all eight dimensions can be populated with evidence-anchored conclusions.

Expected truth is not a verdict; it is a standing question with a timestamp. The only achievement of this report is a diagnosis, not a cricketing ruling: there was no foundation, therefore there was no analysis. When the next file lands on the desk, the first question will be the same — how many information points?