Ten Times the Numbers Disagreed, and What We Did About It

Introduction

When we assemble figures from many sources, we expect some to disagree. The disagreements are informative: they show where definitions differ, where a source has made an error, and where a reader can be misled by a headline. Below are ten we met while preparing this series. For each we say what we did. We resolved none of them by choosing the number we preferred.

The Ten Discrepancies (1 to 5)

1. A segment table that did not add up (PhysicsWallah). Online revenue of Rs 105 crore and offline revenue of Rs 490 crore total Rs 595 crore, against consolidated revenue of Rs 1,054 crore. The article did not explain the remaining roughly Rs 459 crore. We reported the gap and pointed readers to the investor presentation.

2. Two revenue figures for one company (IHCL). The company's release gave consolidated revenue of Rs 2,419 crore. One outlet gave operating revenue of Rs 2,340 crore with the same profit. We used the release and noted that the lower figure may exclude some income, without claiming that it does.

3. Four estimates of logistics cost. The DPIIT-NCAER study said 7.97% of GDP for 2023-24; a KPMG blog said around 8% with no source; a page on The Statesman's website said 10 to 10.7% for FY2026 with an unnamed report; earlier estimates ran to 13%. We set them side by side, explained the definitional reasons, and did not pick one.

4. An evident typo in units (Indian Railways). The cumulative loading figure was printed as "Rs 833.7 mt". We read it as million tonnes and flagged the error.

5. A garbled ratio (AISHE summary). A secondary summary of the higher-education survey gave a gross enrolment ratio of both 7% and 30.0%. We treated 30.0% as the intended figure because 7% was incompatible with the enrolment count, said that was our inference, and asked readers to check the official report.

“Most mismatches come from the space between a source and a reader.”

The Ten Discrepancies (6 to 10)

6. Year-on-year versus sequential (hotels). An article described April 2026 as weaker than March but stronger than April 2025. Both were true. We explained the two comparisons.

7. Gross leasing versus net absorption (offices). Cushman & Wakefield's net absorption fell 18% on the year while CBRE's leasing rose 6%. The measures differ.

8. Different year-to-date GCC leasing. Cushman & Wakefield said 25.3 million square feet; CBRE said 28 million. Reported both.

9. An internal inflation mismatch. A summary of the RBI statement said headline inflation would average almost 5.8% over three quarters, while its listed quarterly projections averaged about 5.5%. We flagged it.

10. A headline that overstated. An article's headline used "record" for hotel investment while its body did not say so. We did not repeat the word.

Patterns & A Six-Step Method for Conflicting Data

Patterns: Most of the ten involve secondary sources: news summaries of reports, blogs and aggregator pages. Several involve differences in definition between primary sources. None, as far as we can tell, involves a deliberate falsehood. Most come from the space between a source and a reader: rounding, copying, summarising and headline writing.

A method for conflicting data: our framework: 1. Go to the primary source when you can. The company release, the ministry bulletin and the regulator's statement beat a summary of them. 2. Check the arithmetic. Do the parts sum to the whole? Does a percentage match the two numbers it describes? 3. Name the measure. Net or gross, nominal or real, calendar year or financial year, consolidated or standalone. 4. Report the conflict. If two reputable sources differ, say so, and say what each is measuring. 5. Use a label for inference. When you fill a gap, say you did. 6. Re-check developing stories. A number that is right on Tuesday can be revised on Friday.

Publishing Ethics, Reader Rules, and Constructive Tone

What it means for readers: When a number is surprising, look for the second source before you quote it. A figure that agrees across independent sources is more robust than one that appears once.

What it means for publishers: The cost of a correction is higher than the cost of a second check.

A note on tone: Listing discrepancies is not an accusation. Companies, ministries and consultancies publish quickly, and a table that does not sum is usually a drafting slip. The remedy is a note from the issuer, not criticism. Where an issuer corrects a figure, we will update the relevant piece and say so.

What we do not know: We do not know the resolution of any of the ten, and we have not contacted the issuers. Where a primary source later clarifies, we will update the piece.