Traceroute devlog #6

This week marked a significant milestone for the project, as an automated pipeline successfully generated a new puzzle and went through a review process before receiving final human approval. This decision, made early in the project's development, felt like closing a loop that had been open since the beginning. While the initial implementation faced several challenges, each failure provided valuable lessons.

The first failure occurred due to the reviewer's explanation being cut off mid-sentence, leading to a crash when the code attempted to read the incomplete data. This was followed by another failure where the loop responsible for generating a good puzzle had no limit, causing it to run indefinitely. The third failure was caused by the automation not having permission to write to the repository, while the final failure was due to an unrelated setting in the repository itself.

Ultimately, the pipeline successfully created a pull request, marking a major achievement for the project. However, the journey was not without its challenges. A level that had never been cleared started appearing in the level list with a full score attached, leading to a deeper investigation. The issue was caused by a gap between three correctly functioning components, resulting in a delayed check that could be triggered by a player leaving and reloading the game.

Another bug was due to the save system quietly discarding wrongly-attributed results, requiring manual intervention to fix. Additionally, the reporter noticed a drift towards editing files directly instead of using the project's prompt-based approach, which was promptly corrected. Throughout this process, the reporter found that many seemingly finished features needed further refinement when exposed to real-world conditions, such as permissions, API limitations, and human interaction.

The reporter reflected on the various bugs encountered, including the undo rule, two-way path feature, and hint system, which all required multiple iterations to achieve the desired functionality. Despite these challenges, the reporter believes they have built a functional system, from end to end, that can hold up under scrutiny.

While there is still work to be done in terms of polish and documentation, the reporter is confident in their ability to explain the core concepts of the project, such as pruning, retry budgeting, and difficulty measurement, to an interviewer. The journey from a blank document to a running system has proven to be a valuable learning experience, demonstrating the reporter's capability to turn abstract ideas into tangible results.