I'm confident that you can use either a four-electrodes or an induction setup over a wide range of conductivity. The limits of the meters you tested must rather result from their intended range, which the designer didn't bother to widen, or from the skills of the designer, since this needs both electromagnetism and analog electronics.
For the inductive measure
Observe only the losses created by water conductivity. At 1000µS/m, 1% accuracy accepts j100µS/m from permittivity, and this effect is easily computed away afterwards. εr
=80 allows F=22kHz, for which ferrite pots are excellent and let make coils with Q>100.
Build a resonant LC circuit at 22kHz, excite it and measure the decay time, or measure the voltage amplification when injecting a permanent excitation through a small capacitor. I take a water toroid of arbitrary h=20mm Ri
=15mm: at 1000µS/m its resistance is 785kΩ. If loading with the water sample shall drop its Q to 50, the necessary reactance is j16kΩ obtained at 22kHz by 120mH and 470pF, and the coil's equivalent series loss resistance must be << 320Ω while the parallel excitation and measurement circuit must apply >> 785kΩ.
Using an RM14 gapped for 1µH/turn2
(warning, this is not correct nor consistent, just for a quick check)https://en.tdk.eu/download/519704/069c210d0363d7b4682d9ff22c2ba503/ferrites-and-accessories-db-130501.pdf
it takes 346 turns which, if Litz wire occupies 30% of 20mm2
, have 28Ω resistance << 320Ω.
The N41 material offers Q=100 at 22kHz without gap, so the gap dividing the inductance by 6.8 brings Q=680 which is also >> 50. For accuracy better than 10% you must compare the Q loaded with water and unloaded.
The Q margin should allow to iterate the dimension design into something consistent.
For the resistive four-electrode measure
, which makes simpler electronics:
Platinum isn't a must for water... Thin gold can be deposited on the copper of a printed circuit but is weak against abrasion. I guess graphite would do it too, and nickel as well - looks better than graphite, easy to clean, can be deposited by any subcontractor. Nickel is non-magnetic if its phosphorus content is >11% (or 15%?), which can be better at high frequencies due to the skin effect.
You could switch among a set of constant currents, or build a feedback: inject and measure the current needed to develop 100mV between the central electrodes.
I don't see a fundamental limit to the resistivity range here. It's more a question of design choices.
You might consider a capacitive measure
Build a capacitor whose dielectric is the water. A coaxial shape is less sensitive to EM perturbations and geometric tolerances. Operate at a frequency where the expected water is capacitive and the conductivity introduces losses, for instance 2MHz.
Build a resonant LC circuit for 2MHz, feed it through a small capacitor, measure the voltage amplification, deduce the losses and the conductivity.
I believe to have data about how constant water's conductivity is up to a few GHz, but didn't check it.
The capacitive design seems much more convenient than the inductive one: drop the capacitor in the water, instead of filling a toroid.