Automatic production testing and validation

A method of testing code comprising:

• receiving, at a processor, record data describing at least one interaction between a client and a production environment providing a first version of a software service;
• receiving, at the processor, company dump data describing a context of the production environment during the at least one interaction;
• constructing, by the processor, at least one simulation of at least one version of the software service including the context, the at least one version being different from the first version;
• replaying, by the processor, the at least one interaction within the at least one simulation to generate at least one replay result;
• generating, by the processor, a statistical model of responses by applying a machine learning algorithm to responses observed by the software service, wherein the statistical model defines a first subset of the responses that are required responses and a second subset of the responses that are optional responses;
• identifying, by the processor, at least one problem with the at least one version of the software service by comparing the at least one replay result with the statistical model of responses;
• and
• sending feedback from the processor to a development device that develops the code, wherein the feedback includes the problem and the feedback is to be implemented to perform automatic corrections to the code.