Our model features reputational dynamics related to those in Holmström (1999), with an agent of uncertain quality working for a market of homogenous short-lived principals. The agent knows its own quality, but the market does not and must infer quality through the agent’s work history, which generates a stochastic process that is publically observable. The agent’s work history thus provides a steady stream of information to the market, continuously updating the agent’s reputation. Our model imposes few assumptions on the stochastic process itself, or the agent’s initial quality distribution. In contrast, most papers in the reputation literature make much stronger assumptions such as binary quality types or Brownian motion signals. Conditional on its work history, the agent chooses when (or whether) to certify, revealing additional information to the market. Certification has a cost to the agent of k and verifies that true quality lies above some standard q, but does not reveal the exact quality level. This type of “imprecise” certification is one of the most common in practice, with examples such as board certification for doctors, pass/fail exams for accountants, and security seals for websites.