One plan; fresh runs can still diverge
Markdown inline parsing · A recorded rewrite case from the preceding RSR paper
Source · Turn 26
Try parser generation
Prepare a text parser
The successful source trajectory tries tree-sitter generate at turn 26. tree-sitter is a toolkit for generating and using parsers, which turn Markdown text into structures a program can process. The diagram shows key events selected in the paper.Source · Turn 51
Use a prebuilt parser
Load a compiled component
A prebuilt parser is an already compiled parsing component that can be loaded directly in a compatible environment. The source trajectory uses it at turn 51. The paper abbreviates the command as tree-sitter parse -l ...node --lang-name markdown /tmp/test.md.Source · Final check
64 turns, passed
Success informs the plan
The source run takes 64 turns and passes final verification. Its successful experience is distilled into a plan that guides the model through the task again in a fresh environment.The same runbook, distilled from success
Compatible parser→Match output summaries→Table-handling options
A runbook records key procedures, checks, and recovery steps. This plan distills the successful source run into finding a compatible parser, matching output summaries, and checking table-handling options. A and B below execute independently in separate fresh environments.Rerun A · Turn 9
Use a prebuilt parser
Reach a working parser early
Rerun A uses a prebuilt parser at turn 9. It shares the plan with B on the right, but each runs in its own fresh environment. The source excerpt at turn 51 records this operation too.Rerun B · Turn 26
Inspect parse output
Terminal: inspection tool not found
Rerun B tries xxd .../parse.txt | head -20; the terminal returns bash: xxd: command not found. xxd is a command-line tool for viewing or converting hexadecimal file contents. This event records a failed tool call.Rerun A · Turn 21
Check rebuilt output
Run the feedback script
Turn 21 checks the rebuilt output with python3 /app/tools/feedback; feedback is the result-checking script used here.Rerun B · Turn 36
Recheck after edits
Run feedback again
After making changes, rerun B runs the feedback result-checking script again at turn 36. This and turn 26 are two execution events selected in the paper.Rerun A · Final check
32 turns, passed
Rerun A takes 32 turns and passes final verification. The successful trajectory then undergoes quality and leakage screening before it can become a training demonstration.Rerun B · Final check
38 turns, failed
Rerun B takes 38 turns and fails final verification. It is filtered out at the final check; the excerpts show its tool-inspection and result-checking steps.Distill successful experience
A · Separate fresh environment
B · Separate fresh environment