Should memory retrieval aim for more context, or context that is easier to inspect?

ThoughtDAG indexes local conversations and canvases. An optional Jev layer identifies topics and ranks relevant excerpts, while answers are generated by the language model the user configures. A context panel shows the material and excerpts added to a request, letting users inspect sources or exclude items. The documentation distinguishes selection-stage timing from the time required for complete retrieval and answering, and provides a rule-based fallback when the decision model is disabled. This shifts the evaluation of a memory system from “Did it remember?” to “Why was this included, and should it have been?” Retrieving more may improve coverage, but can also bring outdated decisions or previously corrected content into a new task. The editors suggest separately recording retrieval candidates, the final selections, and users' reasons for excluding items, then assessing the benefit through the subsequent answers. Retaining sources consumes context, while compressed summaries may lose the scope in which information applies. The balance should be determined by actual tasks, rather than simply increasing the number of retrieved items. Outdated evidence should also be marked separately. Discussion questions 1. When an old decision conflicts with a recent correction, what should take precedence? 2. Do users need to see all retrieved evidence, or only expandable source references? 3. Should evaluation focus on recall, the rate of incorrectly included context, or improvements in the final answer? Source: ThoughtDAG original README https://github.com/chenxiachan/thoughtdag/blob/main/README.md Editorial note: The official S1 Hub editorial account used AI assistance to compile public sources. This was not submitted by the original author, and this site has not conducted independent hands-on testing. Additional evidence and differing views are welcome.