By the 18th century, the average mechanical clock was more than sufficiently accurate to meet the general public's civil and personal timekeeping needs. The scientific community, however, required a much higher level of precision in their laboratories and observatories. In response to this demand, clockmakers sought to make ever more precise clocks.
By the early 18th century, it was already widely know that temperature affects clock metals, diminishing timekeeping accuracy. In 1715, George Graham did a great deal of research into the thermal reactions of different metals, hoping to find two metals whose rates of expansion would "compensate," or cancel one another out. He found such a relationship between steel and the viscous metal mercury, and went on to introduce a mercury-compensated pendulum in 1722. Graham's work helped usher in a new era of precision timekeeping.
As clockmakers coupled a growing understanding of metallurgy with new technological designs, timekeeping accuracy continued to improve throughout the 18th century. Thomas Tompion's first precision clock, made in 1676 for the Royal Observatory in Greenwich, England, purportedly kept time to within two or three seconds per day. Approximately fifty years later, John Harrison built a regulator which incorporated his new temperature-compensated gridiron pendulum and was allegedly accurate to within an astonishing one second per month.
By 1900, the Riefler Company of Germany was making timekeepers accurate to within .01 seconds per day. This remarkable performance was due not only to improved mechanics and materials, but also to new case designs. In order to keep pressure, temperature, and humidity as constant as possible, manufactures put precision clock movements in airtight cases called tanks. These tanks were then often placed in designated clock vaults. The incredible accuracy of these clocks was not fully appreciated right away, as scientists had not yet defined sufficiently small increments of time. For instance, precision clocks produced by W.H. Shortt in the 1920s were believed to keep time within two milliseconds (.002) a day, but when one of these was tested in 1984, it was found to keep time to within 200 microseconds a day, or ten times more accurate than previously believed.