A battery display showing 50 percent looks like a direct measurement. It is not. Electronics can measure current, terminal voltage and temperature, but the amount of charge still available inside a working cell must be estimated. That distinction matters when a vehicle predicts range or a storage operator promises energy for an evening peak. Coulomb counting is one of the central tools behind that estimate. It adds up charge entering or leaving the battery over time, then compares the result with an assumed capacity. The arithmetic is simple. Knowing the correct starting point, measuring every relevant current and keeping the capacity estimate current are the difficult parts. What state of charge actually describes State of charge, or SOC, expresses remaining charge relative to a defined full-charge capacity. It is usually presented as a percentage. A useful explanation must name the reference: a cell estimate, a pack-level estimate and the customer-facing usable window are not necessarily identical. Manufacturers may reserve margins outside the displayed operating range to protect the battery. SOC is not state of health. A worn battery can be fully charged relative to its present capacity and still hold less charge than it did when new. Nor is SOC an immediate power rating. Temperature, voltage limits, resistance and equipment constraints can restrict charge or discharge power even when the displayed percentage appears comfortable. Charge capacity in ampere-hours is also different from energy in watt-hours. Energy depends on voltage as charge moves through the circuit. Multiplying a nominal voltage by ampere-hours can give a useful approximation, but it is not a complete prediction of deliverable energy under every load and temperature. A precise grid-storage guarantee must specify its measurement boundary and conditions. This article focuses on how the percentage is estimated, rather than repeating a general battery-management-system overview. The BMS has many responsibilities, including monitoring and protection. SOC estimation is one input to those decisions, not a replacement for independent voltage, current and temperature limits. Coulomb counting starts with a known balance The basic calculation integrates current over time. For a constant discharge current, charge removed in ampere-hours equals amperes multiplied by hours. Divide that amount by the relevant capacity and subtract the resulting fraction from the initial SOC. Real implementations must define the current sign convention and may account for charge efficiency and other chemistry-dependent effects. Take a deliberately simplified 100 Ah battery starting at 80 percent. A 10 A discharge lasting two hours removes 20 Ah. Assuming the capacity is exactly 100 Ah, no other current paths exist, and losses or endpoint effects are neglected, the estimate falls to 60 percent. These are teaching numbers, not a test result or a recommended operating profile. The method measures change, not an absolute starting state. If the battery actually began at 70 percent while the software believed 80 percent, perfect current counting would preserve that ten-percentage-point mistake. Repeated integration cannot discover a wrong initial balance on its own. The estimator needs a trustworthy reference or another source of information to correct it. BatteryDesign.net describes this distinction explicitly: coulomb counting supplies relative change, while absolute SOC requires the starting value. Its technical overview lists initial SOC, current measurement, integration, capacity and timing as error sources. These are practical hardware and calibration questions, not merely a choice between competing software labels. Small measurement errors can become large drift A current sensor has a useful operating range, measurement uncertainty and potential offset. In a hypothetical example, a persistent 0.1 A measurement bias integrated for 24 hours creates a 2.4 Ah counting error. Against a 100 Ah reference, that is 2.4 percentage points of SOC error. The example isolates one effect; it is not a claim that commercial systems universally have that offset. Bias and random noise should not be treated as the same problem. A persistent offset can accumulate in one direction. Noise, sampling behavior and filtering affect the integral differently. A sensor selected to measure very large peak currents may also need careful validation at low currents, where the absolute error can be important relative to the signal. Sampling adds another source of uncertainty. Software approximates the current integral from discrete measurements. If current changes rapidly between samples, a coarse approximation can miss charge transfer. Clock accuracy matters because the integral includes elapsed time. A mathematically elegant algorithm cannot recover information that the acquisition system never captured. A research paper by Kiarash Movassagh and colleagues, available through arXiv, formally analyzes current measurement error, integration approximation, battery-capacity uncertainty and timing-oscillator error. It distinguishes errors that accumulate with time from those proportional to accumulated SOC change. That framework explains why an estimator should be assessed over realistic operating histories rather than a short, convenient demonstration. The denominator ages, and voltage needs context Suppose software still assumes a 100 Ah capacity when the relevant present capacity is actually 80 Ah. Removing 20 Ah is then a 25-percentage-point change, not 20. This simplified comparison shows why capacity estimation and SOC estimation interact. Using the original nameplate value forever can distort the interpretation even if measured current is accurate. Capacity itself depends on the conditions and endpoint definition used to measure it. Temperature, discharge rate, aging and cell variation all complicate the reference. An operator comparing two SOC values should ask whether both systems use the same usable window and current capacity assumptions. A precise-looking dashboard is not proof of equivalent definitions. Voltage provides another clue, but terminal voltage under load is not the same as rested open-circuit voltage. Current-dependent voltage drop and electrochemical relaxation affect the reading. BatteryDesign.net notes that open-circuit-voltage lookup methods require a rested cell and are affected by hysteresis as well as measurement accuracy. A quick voltage snapshot should not be mistaken for an exact fuel gauge. Model-based estimators can combine current integration with voltage and temperature measurements to correct predictions. Kalman-filter approaches are one family of methods, not a universal guarantee of accuracy. The model, parameter identification, sensor quality and tuning must suit the battery and use case. A poor model can produce a confident-looking estimate that is systematically wrong. What an SOC error means for real operations For a storage plant, an overly optimistic SOC estimate can make the schedule look feasible until the battery reaches a protective limit earlier than expected. An overly conservative estimate can leave useful capacity idle. The financial impact depends on the contract, power profile, reserve margins and actual available energy. No single SOC-error percentage translates into one universal revenue loss. For a vehicle, SOC contributes to the range prediction but does not determine it alone. Driving conditions and energy consumption also matter. A changing estimate after a rest period or a calibration event does not, by itself, prove that physical capacity suddenly disappeared. It may reflect the estimator correcting an earlier error. Persistent or unexplained behavior still warrants the manufacturer’s diagnostic process. Cell-level limits matter at pack level. A pack contains cells with some variation in capacity and condition. An average percentage can conceal a limiting cell that reaches a voltage