The state-of-the-art timekeeping technology a century ago was comprised of pendulum clocks. Refinements were made in the areas of obvious problems, such as the mechanical escapement which robs the system of energy, the vulnerability to changes in length from temperature and humidity, and vibrations. The culmination of this was the clock of W. H. Shortt, which had two pendulums, a master and a slave. The master oscillator was a free pendulum, and as it did no work to drive any mechanism, it was able to keep very precise time. The pendulum was made of invar, a material that had a very low thermal coefficient of expansion, and was encased in a chamber that was evacuated to several millitorr of pressure. The chamber was bolted to a wall that typically rested on a massive platform of the type used for telescopes, which minimized effects from vibrations. The pendulum was given an occasional boost to keep its amplitude roughly constant. The slave pendulum, which did the mechanical work of the system, received periodic electronic impulses from the master clock to correct its motion. This type of clock could keep time to better than a millisecond a day. A shortcoming (as it were) was in the measurement of the time; as Loomis notes
This remarkable result is accomplished through the possibility of averaging a large number of observations. A single impulse from a master Shortt clock has an uncertainty of 1 or 2 milli-seeonds. The master pendulum carries a small wheel. The impulse arm rests on this wheel, and as the pendulum swings out the pallet on this arm travels down the edge of the wheel, finally falling clear . It then trips an arm which falls, making the electric contact . If the small wheel is not exactly circular the arm will fall at slightly different times as the wheel is given a small turn with each fall. These variations are entirely smoothed out when a series of sparks are averaged.
So while the clock is precise in the long-term, the system of measuring it (described below) is limited at shorter durations.