The oil balance is an accounting identity. Everything produced must be consumed, stored, or in transit, so supply minus demand equals the change in inventory. That is not a theory; it is arithmetic, and it is the frame every fundamental analysis is eventually expressed in.
The identity is exact. The measurements are not. Production is estimated country by country with varying quality, consumption is estimated even more loosely, inventories are measured well in a handful of OECD countries and barely at all elsewhere, and barrels on water take weeks to arrive. So when the numbers are added up they do not balance, and the gap is real and large.
What agencies do with that gap is publish it. The IEA reports a 'miscellaneous to balance' term; the EIA has an adjustment; both are the residual required to make the identity hold given the measurements. That residual is one of the more honest numbers in the whole subject, because it is an explicit statement of how much the publisher does not know.
Two structural facts make the demand side harder than the supply side. Demand is measured mostly as a residual from refinery output and trade rather than observed directly at the point of consumption, so demand estimates inherit every error in those inputs. And demand is genuinely seasonal at a scale comparable to the annual growth rate, which means a year-on-year comparison and a month-on-month comparison can point in opposite directions and both be right. See demand seasonality.
On the supply side, the market's structure means a handful of decisions matter more than the rest of the distribution: OPEC and its partners manage output deliberately, spare capacity determines how much shock the system can absorb, and US shale responds to price faster than any other large source.