Layering the path — a low-capacitance array at the FPC or connector entry, a short wide TVS return to a solid plane, and a series resistor as a second current limit — is the right backbone. I would not treat “10 pF”, “100 Ω to 1 kΩ”, or the ±8 kV / ±15 kV figures already given as universal budgets; those have to come from the controller’s pin capacitance and absolute maximum, the scan rate, and the IEC test level you actually owe. Three checks still decide whether a hit becomes a reset, a false touch, or a hung I²C session.
The first is the mechanical and charger return, not only the TVS footprint. A metal bezel, ITO shield or cover-lens ground that bonds to a different point than the TVS return will let the discharge current share the touch-sensor reference. The same split happens when the product is tested or used while a USB charger is plugged in: the hit can enter through the cable shield and leave through the panel, or the other way around. After a cover-glass or FPC change, re-check the bond of the shield and the charger earth path, not only whether the array is still in the BOM.
The second is coupling onto reset and the supplies. A clean clamp on the sensor lines can still inject current into the controller’s reset pin, DVDD or AVDD through a shared via or a long slot in the plane. Failures that look like “the firmware died” are often a narrow droop or a reset glitch; log the reset line and the local rails during the hit, and keep the bypass and any ferrite of those rails close to the IC. Residual energy after the connector clamp may still need a small local clamp at the IC pins if the FPC is long.
The third is recovery on the host. An industrial HMI or a Linux-based panel often talks to the touch controller over I²C. After a hit the bus can be left mid-byte even though the screen looks alive, which is then written down as a frozen UI. The driver should time out, recover the bus, and re-initialize the controller rather than waiting for a watchdog. Stamp the ESD event, the bus-recovery, and any false-touch burst with CLOCK_MONOTONIC so a later field log can be aligned with the test report. Acceptance should still record contact and air discharges separately, and should keep reset, false touch and hang as different failures, not a single “ESD fail” bit.