Regarding the test flights at Toulouse, I'm going from Captain Bechet's interviewed statements on the ACI programme, and he at least seemed to be confident that it was close enough to the actual event (though obviously with some variation) to be useful. Other than that I make no claims whatsoever, other than HN39 (as usual) seems to be on the money regarding the effect of earlier vs. later application of full back-stick.
As for the software in the original simulators, it would have been standard practice at the time to provide complete and working hardware/software combinations as a unit - as long as the interface of the simulator and that of the aircraft was the same it should have been like-for-like. As such, the behaviour of the EFCS would have been one-and-the-same on the simulator and the real aircraft, but the simulator's flight model behaviour would still likely have been undergoing refinement. The "black box" (in engineering terms, i.e. meaning a 'sealed unit') was and still is for commercial considerations only - Airbus's competitors would have likely been very interested in the implementation of their systems, and - especially at that time - the fear of industrial espionage was a real one.
I know that the flight model was essentially fed with the data gleaned from the flight testing performed on the A320 prototypes, and I've been told that those prototypes were the most heavily instrumented and monitored machines of their type at the time. I don't know whether they were flight tested beyond the stall boundary, but wouldn't be surprised if they were at least taken as close as possible in order to gather data on airframe characteristics.
From my brief experience, I can say that modern A320 simulators do provide a plausible representation of behaviour going into stall and recovery, but can't vouch for how precise a representation that is.