The report makes an important contribution by analyzing some reactor accidents that are worse than the DBAs. Specifically, the report treats:
(1) the LOCA-without-ECCS type of core meltdown;
(2) a class of accidents called "transients," which basically come under this author's category of power-cooling mismatch accidents (PCMAs) and heat exchange accidents; and
(3) the spontaneous reactor vessel rupture. The power excursion accidents (PEAs)and the worst forms of PCMAs are essentially excluded. More- over, on the basis of certain assumptions, the report treats the "transients" and the reactor vessel rupture as no worse than a LOCA without ECCS (slow core meltdown), which is not the worst course these accidents could take.
(The report's treatment of the LOCA-without-SCRAM-type accident was examined earlier; see pp. 34 -39.) In short, the report is grossly inadequate in scope.
Looking at the Three Mile Island incident - The NRC report points to a LOCA caused by a faulty valve - compounded by
Bad training
faulty human-machine interface
bad sensors
bad/untested emergency procedures
Subsequent visual examination of the rector core showed far more damage than was initially thought, but reactor breach did not ocurr.
A lot of lessons were learned, and changes were made. All good stuff.
That nobody was injured seems to smack of the "no danger from fallout" pronouncements back in the day of open air testing.
Most metallurgical studies I have seen on pressure vessel rupture indicate large-scale failure at a vanishingly small probability. For smaller-scale ruptures, the LOCA without EECS accident evolution is a reasonable model. Sure, you can postulate a large-scale rupture, but your accident analysis credibility would be challenged.
If anything, the TMI accident sequence indicated that accident analysis is probably overly conservative. That much core damage and energy release would have led many to believe that a slow-evolution vessel breach should have occurred. That it did not is probably evidence that our assumptions and accident models have a fair amount of conservatism built into them.