The curve is flat enough that estimating the State of Charge based on voltage is challenging. Basically, the open circuit voltage of a Lithium Ion battery is not a very precise indicator of the SoC.
I don't know, I do this for a living and for me, it has never been challenging, but works well and is simple.
Of course, the indication jumps around a bit with load changes.
Coulomb counting is more challenging to get right, but nowadays it usually works well (in laptops etc.) During the first decade or so, problems with severely jumping indications were usual.
What you say is interesting - the voltage can be "good enough for Australia" to estimate the charge level?
What happens as the battery ages and the cycle count begins to rack up, and it begins to lose capacity - surely you must lose a lot of accuracy as time goes?
Voltage-based SoC indication adapts automatically to the capacity loss without any action. OTOH, if due to aging internal resistance of the cell increases, indication jumping due to load variations also increases. Voltage-based indication works best for fairly low-current loads (i.e., when under-load voltage is not too far from open-circuit voltage.) Short peaks are not a problem - just filter the reading.
Voltage based measurement is notoriously bad on NiCd, NiMH and lead acid chemistries because of their flatness, and works quite well in Alkaline primary cells and most li-ion chemistries. It would work very poorly with the Silicon Lightworks example cell, but again, I suspect this is a made-up curve (or some special snowflake cell).
Coulomb counting needs to track actual capacity, which requires reset points, which are voltage based anyway; usually full and empty. If battery is not regularly fully charged, coulomb counter drifts, lacking 100% reset point. If battery is not regularly fully discharged, estimate of actual capacity is not updated. If you want to make truly useful coulomb counter (without requiring the user to do full charging/discharging cycles), you need to really incorporate voltage and current measurements all the time, along the curve, so it would be a fusion of integrating charge, and estimating from voltage. I would guess the best "fuel gauge" ICs do this quite decently nowadays.
OTOH, there are also ICs that incorporate "advanced" algorithms
without coulomb counting, but still utilize information about voltage, current, and the pre-programmed cell curve. This allows near the accuracy of coulomb counting, without the cumulative error, and without need to fully discharge the cell to estimate actual capacity.