You're right in what you say GtE, but my interest and concern comes from earlier in the process than that. By the time we get round to the safety reviews you're talking about the trial scope has usually been defined by Other Beings - e.g. design offices, aerodynamicists, project managers - so all we're doing in the flight test safety review is minimising residual risks in that narrow remaining task. The mis-applied “safety” argument hits us much earlier than that.
For example, let's pretend we're evaluating a high AOA trial on a sporty plastic fighter jet. Aero model away for years, playing with computers and jolly expensive metal models in windy tunnels: they make their best professional guesses at the results but are the first to admit they can't model with great precision and confidence at high AOA, particularly when they’re modelling that stuff well ahead of prototype first flight so that they can work out how to design the mega-smart flight control system (FCS) to keep the plastic pointing forwards all the time. Then we start flying and envelope expansion, they start finding out what’s wrong with the model, and massage it accordingly. Trouble is, the closer they get to their predicted AOA limit, the more uncertain they get. Is the prediction real? Don’t know, they say, don’t have the data, have to put such margins on our modelling and wind tunnel data because of “safety” concerns that we can’t rely on the data. That means that we can’t get flight clearances to go and get the data because the data’s not “safe”. The programme managers then assume that because the data’s not reliable, we’d best not fly there, pity about the performance spec but we’ll trade that against something else. Meanwhile, every test pilot in the programme is up in arms over the negative, over safe “add another 10 knots for safety” design and clearance mindset while pointing out that this is actually what flight testing should be about – we have the instrumentation, the cleared spin chute, the post-departure modelling and a load of experience on how to safely run such a programme – so why don’t we go fly a progressive, controlled workup programme to get Real Data – using (what the FCS project test pilot of the time described as) God’s Wind Tunnel, thereby resolving all the modelling uncertainties.
In that little story (fictional, of course) there are at least 20 examples of the use of a “safety” argument for not doing something that wasn’t balanced by an assessment of the benefits to be gained for doing that something – and they all occurred long before the flight testing empire got around to applying their own safety analysis to the undemanding and simple test programme that was all that would remain. That’s what I was referring to in my original note.