Lots of dead code in TTD-DR plugin

[Nurgak]

I'm attempting to understand the TTD-DR plugin. In the process, I've noticed a lot of dead code: large parts of the code base are not used and the code does not follow the process flow described in the README.

For example:

• generate_structured_report(), synthesize_with_memory() are never called.
• The outputs of phase 1.5 are unused.
• The examples do not contain the metadata that is supposed to be added.
• It seems that the references are seldom used: not all the references at the end of the example outputs are cited in the output. Each iteration seems to add more references, but they are not cited in the output.

Additionally, this implementation does not include component-wise optimization, via self-evolutionary algorithm, as described in the research paper.

Not that there's anything wrong with it, but I'm curious: was this largely vibe coded?

[codelion]

Hi, thanks for checking this out. Yes, I did the implementation using Claude code but it was tested on the deep research benchmark to ensure that the reports generated were working. I will clean up some of the unused code in the plugin in a PR.

The examples were all done by hand manually via the in build chat interface that's why they are missing the metadata, I copied them to the final report to share the samples in the sample directory. The references are an issue because we tested mostly with small local llm like Qwen3-0.6B and it makes mistakes like this frequently. I can add a check at end to remove the references that are not cited.

The self-evolutionary algorithm and component-wise optimization are not included for simplicity, do you have a particular example query or use case where it would have helped to have them?