TRAJEXA: Keeping Self-Reports, Task Responses and Recalled Episodes Distinct in Personal Exploration

Personal exploration can draw on current self-descriptions, activity interests, responses to short tasks and recollections of earlier activities. These sources can inform one another without supporting the same claims. We present TRAJEXA, a research system that retains these distinctions in a local, rule-based workflow. Fifteen personality-oriented statements, twelve interest statements, five branching task families and up to three recalled episodes produce separate profiles, a source-coverage view and a graph of retrospective order. Earlier self-reports remain separate from current scores; optional dream reports do not enter those scores. We formalise the implemented rules and audit the gap between their intended meaning and their actual behaviour. Fourteen authored counterexamples expose eight failures in the baseline implementation, including treating missing current ratings as contradictions, accepting responses from inactive task branches and presenting recalled order without an explicit observation boundary. The revision satisfies all fourteen declared checks. These are regression results on cases that informed the repair, not estimates of diagnostic accuracy. An exact enumeration of eighty task-response paths also shows that, under independent uniform guessing, at least one answer is correct on 7/16 paths within each family. The manual task-to-interest mapping has no task coverage for one direction and identical coverage for two others. We therefore interpret the outputs as inspectable exploration prompts, not validated personality, intelligence or vocational diagnoses. The contribution is an executable account of evidence boundaries and their failure modes, accompanied by a reproducible audit and a prospective evaluation protocol. Interactive demonstration: https://trajexa.net/try.html