Extract Reusable Lessons From a Failed Run
The problem, from first principles
A failed run that just gets discarded teaches the system nothing. A cheap first step toward actually learning from it: recognize which category of failure this was, from a small, curated set of known patterns, and attach the lesson already written for that category — no model call needed, just pattern matching against categories a team has already identified and written guidance for.
From theory to code
You're given failed_step (the text describing what the failed step reported) and error_keywords, an ordered mapping from a substring pattern to the lesson text for that failure category. Implement extract_lesson(failed_step, error_keywords). Return the lesson for the first pattern (in error_keywords's iteration/insertion order) that appears as a substring anywhere in failed_step. If no pattern matches, return the literal string "No specific lesson identified -- investigate manually.".
Constraints
error_keywordshas 0 to 100 entries; matching is case-sensitive, plain substring.
Hints
Hint 1
dict iteration in modern Python walks keys in insertion order, so a plain for pattern, lesson in error_keywords.items(): loop already checks patterns in exactly the order the problem asks for — no need to sort or otherwise reorder anything.
Hint 2
Return the moment the first match is found. Don't collect every matching pattern's lesson and pick one afterward — the first match in dict order is the only one that matters.
Keywords
No results yet
Run your code to check it against the examples, or submit to run every test.