Berkeley pairs a shake table with VR to study how occupants perceive risk in an earthquake

The denominator is the first thing to ask for, and this account does not supply one: the nearly-40% figure arrives without the number of occupants surveyed ^[12], which makes it a strong direction rather than a measured rate ^[1].

Take it at face value anyway, because the arithmetic is the interesting part. If roughly 40% reported being incapacitated or frozen and very few executed the protocol, then something close to three in five occupants were mobile and doing something other than drop, cover and hold on ^[11]. Other work has catalogued some of that something: people running because others were running, people moving toward family rather than away from hazards ^[10]. Ceferino's argument is that those choices raise injury risk and appear nowhere in the guidelines ^[10].

The measurement problem is why this lab ended up in a museum. Perceived risk has been reconstructed from post-event surveys, which lean on memory and cannot be tied to the shaking intensity at one spot at one moment unless the building was already covered wall-to-wall in sensors ^[3]. The Shake House inverts that constraint. The accelerations are known because the researchers wrote them, drawn from the simulator's replays of the 1906 San Francisco and 1989 Loma Prieta shaking, then used to calibrate falling objects in the headset so that what participants see matches what they feel ^[5]. Gaze logging gives attention on a timeline instead of a recollection ^[8]. That is a real instrument, and it removes the two weakest links in the existing evidence.

What it does not remove is the absence of consequence. Fifty museum participants in a virtual one-story living room ^[6][7] are afraid of a simulation, and no ceiling above them is going to fail. The freeze rate inside that rig should not be read as an estimate of the freeze rate in an occupied apartment block; the usable output is the shape of the relationship between shaking intensity and behaviour, not its level. The account also reports no behavioural results from the run and describes no comparison condition ^[14], so for now there is nothing to check the field number against.

One further distinction. Ceferino says his own work and prior literature point to risk perception as key to what people do, and as critical to how injuries occur ^[15]. A recall survey cannot separate that from the plainer story in which the buildings that shook hardest produced both more freezing and more injuries. A calibrated simulator can vary intensity while holding the room constant, which is precisely the comparison the survey cannot make ^[9].

My read, with its condition stated: the guidance is sound advice measured against the wrong endpoint. Asking whether people know the rule tests knowledge ^[2]. The field data suggests a large minority never get to the point of applying it ^[1], and that gap is a design problem for the guidance, not a training failure in the occupant.