Stringer,
as a humble addendum to what the others said.
This phenomenon you describe is rather well-known, what can be recognised by the already given replies.
Often it may be found that the electrode composition plays a major part with this. This again is a variable dealing with the electrode materials' work function.
Professor John Norrish of the University of Wollongong in Australia - one of the most outstanding welding experts in the world - has conducted some examinations comparing the power supply's open circuit voltage vs. the HF arc start rate at relatively low currents.
At an OCV of 30 V and a current of 20... 30 A (electrode vertex angle was 45°) he tested thoriated, lanthanated, ceriated and Ytterbium-oxide (Y2O3) doped tungsten electrodes and achieved relatively similar arc start rates.
At 22 V and 24 V OCV however, the lanthanum-oxide doped electrodes (1% La2O3) showed definitely the best results in arc start induced by HF.
Anyway, often the arc start induced by HF becomes unstable due to both a combination of electrode material and an "...adverse phase relationship..:" (John Norrish, 1992) what means that a lack of synchronisation between the power supply and the arc starting device does occur.
Of course also the work function of the base material can play a part with an improper arc ignition. Often it's rather a tricky combination of all the little peripheral factors (electrode condition, variations in the vertex angle, surface contaminations,...,...) what leads to inconsistent HF arc ignitions.
We have helped ourselves in semi-automatic GTAW applications e.g. when welding thin stainless steel sheets at low current values, by wrapping the ceramic gas nozzle into a conductive foil (aluminium foil). The physical effect is to create a negative space charge, and to hereby improve the ignition. It worked quite well!
Perhaps it's worth a try for you.
Stephan