Case type: Independent user
Evidence strength: Owner-approved user report with one public method reference Runtime / scale: More than 13 hours; one long-running agent reported
An independent user used LoopX on a complex C++ algorithm accuracy problem. The reported failure mode was not simply that the code was hard to read: a long execution could drift into local parameter tuning, lose the original accuracy goal, and repeatedly compress away useful context about call relationships.
LoopX kept the multi-stage task aligned to the declared vision for more than 13
hours. When the current route no longer appeared able to satisfy that vision,
replan prompted public research instead of more local thrashing. The run then
selected the public
codebase-memory-mcp project
to help inspect C++ call and coupling relationships.
The user reported no parameter micromanagement or repeated local redirection during the multi-stage run. This is not a claim of zero human involvement in the surrounding project: the original goal and final assessment still came from the user.
The user reported a clear improvement in final precision, less context compression, and enough retained experiment evidence to understand why the final approach worked. The private project and its measurements are not available for independent reproduction, so the performance result remains a user-reported outcome.
Source: an owner-approved message excerpt from the LoopX public Lark developer group. Runtime and outcome are user-reported.
Source: an owner-approved follow-up message excerpt from the same LoopX public Lark developer group, describing the public-research transition.
The screenshots are owner-approved message excerpts from the LoopX public Lark developer group and are stored as local repository assets. The project repository, raw run state, prompts, experiment data, and performance measurements remain private. The MCP repository is public and independently inspectable; it does not independently prove the reported precision gain.