A reliable financial dataset comes from an authoritative source, states how it was adjusted and when it was revised, and covers the full universe without silently dropping companies. An unreliable one hides its methodology, changes past numbers without a trail, and shows gaps you notice only when the figures stop making sense. Judge data less by how polished it looks and more by whether you can trace each number to its origin.
Key takeaways
- Source authority matters first: prefer exchanges, regulators, and registered filings over aggregators that copy from unnamed places.
- A trustworthy dataset documents its adjustment methodology, such as how it handles splits, bonuses, and dividends.
- Revision history tells you whether past numbers have quietly changed, which is a warning sign if it is undocumented.
- Coverage and gaps decide whether an average is honest or distorted by the companies that were left out.
- Consistency checks across independent sources catch errors that no single dataset will admit to on its own.
Why does the source of the data matter more than its polish?
Financial data is only as trustworthy as the place it originally came from. The most authoritative sources are the ones that generate the number in the first place: the exchange that recorded the trade, the company that filed the statement, or the regulator that mandated the disclosure. In India that means the NSE and BSE for prices and corporate actions, company filings for financial statements, and SEBI disclosures for regulatory events.
Many websites present data with clean charts and confident figures, but a polished interface says nothing about where the numbers came from. An aggregator that copies figures from an unnamed upstream source inherits every error in that source and adds its own. The first question to ask of any dataset is not "does it look right" but "who created this number, and can I trace it back to them".
What does adjustment methodology tell you?
Raw prices and per-share figures are meaningless until you know how they were adjusted. When a company does a Stock Split, a Bonus Issue, or pays a Dividend, the share price and history must be adjusted so that the past is comparable to the present. A dataset that does this well documents exactly which corporate actions it adjusts for and how.
The problem is that two datasets can both be "correct" and still disagree, because one adjusts a price series for dividends and the other does not. This is a common reason that CAGR and long-run return figures differ between sources. A reliable dataset states its adjustment rules openly. An unreliable one leaves you guessing why its history does not match anyone else's, which connects directly to Why historical prices change.
Filter companies on Artha Terminal using exchange-sourced figures you can trace back to their origin.
Why does revision history reveal quality?
Good data providers revise numbers, and that is not a flaw in itself. Companies restate results, exchanges correct recorded trades, and preliminary figures get finalised. What separates a reliable dataset is that it records these revisions with a timestamp, so you can see that a number changed and when.
An unreliable dataset overwrites the past silently. A figure you noted last month is simply different today, with no note explaining why. This matters most when you are studying history, because a dataset that quietly rewrites old values cannot tell you what was actually known at the time. The disciplined version of this idea is Point-in-time data, which asks not just what a number is now but what it was on the date you would have acted on it.
How do coverage and gaps distort what you conclude?
A dataset can be accurate on every row it contains and still mislead you through what it leaves out. If a screener or a historical average only includes companies that exist today, it has quietly dropped every firm that was delisted, merged, or went bankrupt. The surviving set looks healthier than the real universe ever was.
This is why coverage is a quality dimension in its own right. Before trusting an average, a ranking, or a backtest, ask which companies are in the universe and which fell out. Gaps in coverage are rarely advertised, and the missing rows are exactly the ones that would have lowered the average. This failure has a name, and it is serious enough to deserve its own treatment in Survivorship bias in screeners.
How do consistency checks catch errors no source admits?
No single dataset will tell you it is wrong. The practical defence is triangulation: check the same figure across two or three independent sources and investigate any disagreement rather than picking the number you like. If a Market Cap or an EPS figure matches across the exchange, the company filing, and an aggregator, your confidence is earned. If it does not, the disagreement itself is the finding.
Internal consistency matters too. Does Book Value per share reconcile with total equity divided by shares outstanding. Does a reported growth rate match the underlying numbers. Errors usually show up as small contradictions before they show up as obviously wrong figures.
On Artha Terminal, market data is sourced from exchange and end-of-day feeds with the adjustment and revision behaviour documented, so a figure on a stock page can be traced rather than taken on faith. You can also ask the assistant, Ask Warren, to explain how a particular number was derived before you rely on it.