Attach identity to every number
Revenue of ₹500 crore is incomplete. A usable fact needs the company, metric, unit, period, duration, scope, currency, source and restatement status.
The label is part of the data. Without it, a correct number can still support the wrong conclusion.
Do not mix durations
A three-month quarter should not be compared with a year-to-date period as if both covered the same duration. Indian filings often show current quarter, preceding quarter, corresponding quarter and year-to-date columns together.
Normalize the period first. Then calculate growth or margins from clearly identified inputs.
- Quarter versus year to date
- Standalone versus consolidated
- Audited versus unaudited
- Reported versus restated
- Rupees, lakhs, millions and crores
Keep reported and derived values distinct
If the company reports revenue and profit, those are observed facts. If you calculate a margin or growth rate, that result is derived. Store the formula and inputs so another researcher can reproduce it.
This distinction becomes essential when AI systems analyse thousands of filings. The model should know what the company said, what the platform computed and what still requires judgement.