Question of how to recover DAG from prompts

[WangXinglin]

Hi LongCoT team,

First of all, congratulations on releasing this excellent benchmark and the dataset.

I have been looking through the released data and code in detail, especially the prompts under src/data. For some domains such as math and chemistry, it seems that the prompts already contain fairly explicit subproblem structure (e.g., Problem node_i / Subproblem i) and dependency relations.

I was wondering whether you have already implemented any internal script or tool to parse these prompts into an explicit DAG / graph representation (for example, nodes, edges, and target output nodes), even if it has not been released yet.

More specifically, I would be very grateful to know:

1. Whether there is any existing prompt-to-DAG parsing implementation used internally;
2. Whether any structured graph annotations (e.g., node list / edge list / intermediate answers) may be released in the future;
3. Whether you have recommendations on which domains/templates are the most suitable for recovering explicit DAG structure from the released data.

I am very interested in using LongCoT for research on structured task decomposition and DAG-based reasoning, so any guidance would be extremely helpful.

Thank you again for the great work and for releasing the benchmark.

Best regards，
Xinglin Wang