Context engineering · 7 min read
Same symptom, different cause: why context beats keywords
How multi-variable case matching turns old incident records into useful evidence instead of noisy search results.
Every shift creates knowledge. The challenge is making it available to the next person, at the exact moment they need it.
A symptom is only the entrance
Two machines can report the same symptom while failing for entirely different reasons. Like two cars that will not start—one may have an empty tank while the other has a weak battery. Searching only for the symptom treats both as the same problem.
Similarity lives in the surrounding variables
Sound, temperature, recent maintenance, product variant, shift conditions and attempted actions all change the meaning of an event. Extracting these variables makes it possible to retrieve cases with genuinely useful overlap, even when their wording differs.
Retrieval gives the model evidence
A language model should not invent the plant context. Its role is to assess the relevance of context-rich cases that the system has already retrieved. This is how a five-year-old incident can become practical evidence for today’s decision.
The practical test
Could a colleague on another shift use this record without calling its author? If not, the why is still missing.