Context Engineering - Part 1

[Dalton_Ruer]

They say a picture is worth a thousand words. 

They are wrong. I recently spoke about 10,000 trying to explain the concept of Context Engineering to someone recently because I had no image. So, I went ahead and created this image to help others understand the concept of Context Engineering. If you use Qlik Answers or Qlik MCP on top of "as is" Qlik Sense applications you will get answers. But if you add "context" in the form of Master Items and Glossary terms the cost of getting answers with Qlik MCP will go down significantly for you, and with both Qlik Answers and Qlik MCP the accuracy and thoroughness of the answers will go up. 

 

Master Items

While that image describes the concepts of it, I know in the back of your mind you are probably asking "Why does Context Engineering help?" or "How does Context Engineering help?" So, let's "visualize" an example and back up a second.

Imagine you are using my ITIL Service Intelligence accelerator to read your ITSM data, move it through my pipeline and present the data to your end users.

If you are using my ITIL Pipeline Pulse application you are able to quickly get vital data on the health of the entire data pipeline. It's great to see the the Conform Success Rate and Parity Rate are both 100%. Right? If they are both 100% you probably don't care what those measures mean. 

But what happens if either of those numbers is less than 100%. My guess is that you are going to say "I've seen these numbers for 3 months and now suddenly one is showing 82%. Unlike my applications where all my users instinctively understand every metric perfectly so I don't both creating master items with good definitions, I sure hope that the Qlik Dork added good definitions so I can ask for the details. 

You are in luck my friend ... if you right clicked on the KPI and chose "Show details" you would in fact see a definition. In fact, notice that you see my description and the fact that the master measure behind the KPI is tied to a business glossary term "Conform Success Rate." 

You gotta love that. Right?

Great for you to know it's tied to a business glossary item as well. But it's not like you can just click right on it and jump to see that term in the business glossary or something. Just kidding. Of course you can just click right on that term and have the business glossary opened right up to that term. 

Business Glossary

Blah-blah-blah a nice definition including the expression needed. Yadda-yadda-yadda yes you can see that the business glossary term is tied to the master measure in the application.

But let's skip that boring stuff (for the time being) and jump right into the cool stuff. Those "related terms." 

Notice that Conform Success Rate is related to "Conform Health (Concept)" and "Conform Parity Rate." Conform Parity Rate is another way of looking at the health so I won't bore you showing it's master measure or the glossary term, but they are there. Another key aspect of this business glossary term is that it has a "steward." Which means that person has put their name behind the term. For now just keep that in mind, as we will get back to it. Instead of those things, lets jump right to the Conform Health (Concept). 

It explains that this is about a health signal and then describes both ways of tracking the "health" of the conform pipeline. 

Example

Seriously cool stuff and it's a really good thing for you that I went to these lengths to help you be prepared if either of those KPI's isn't 100%. Again, this is just about my application for you, and in now way implies that your end users need this kind of information about the applications you built. Because we know they absolutely understand every single thing on your screens. 🤔

The truth is all of that information about the master items and the business glossary wasn't even about the Service Intelligence accelerator at all. It's simply to lay the foundation for you to understand the opening Context Engineering image. 

Imagine that rather than using the application you simply want to ask like "How healthy is the conformance data in my ITIL data pipeline?" 

Good for you. That's a super reasonable question to ask. For a second, pretend that I didn't build out all of the context and that I simply put a bunch of expressions behind KPI's in charts. Or that I built master measures with expressions but without definitions. What do you think Qlik Answers would do? How do you think Qlik Answers would know how in an application with a lot of data it should use the 2 expressions behind those KPI's shown? 

Even worse ... imagine you asked that question inside a chat tool like Claude. Don't worry I want your head to hurt trying to imagine exactly what happens when questions are asked so I will bail you out. Here is the generate path if you have no glossary or master items in place:

• qlik_search → "conformance" — returns 6 apps, no way to tell which one actually tracks pipeline health
• qlik_describe_app (App 1: "ITIL Service Intelligence") → looks promising, but "conformance" isn't a field name anywhere in metadata
• qlik_list_sheets (App 1) → 22 sheets, none obviously labeled "Conformance" or "Pipeline Health"
• qlik_get_fields (App 1) → 280 fields returned, no glossary to explain what "conformance" maps to
• Scan field names manually — find 83 fields prefixed ALL:, 40 fields prefixed CFM:, no idea which prefix means "conformance layer" vs something else
• qlik_describe_app (App 2: "Load-Conform-Service Pipeline") → name sounds closer, dig in
• qlik_list_sheets (App 2) → find a sheet called "Data Quality," open it to look for anything conformance-related
• qlik_get_chart_info on 5 charts to check what each one is actually measuring
• Find a chart using [Record Count Variance] — sounds relevant, but no way to confirm it's the sanctioned "conformance health" metric
• qlik_get_field_values on 3 candidate fields ([Load Row Count], [Conform Row Count], [Variance Pct]) to sanity-check which represents "health"
• Wrong assumption — [Variance Pct] turns out to track schema drift, not row-count conformance
• qlik_get_chart_data on a different chart using [CFM: Match Rate] — closer, but still unverified whether this is the KPI or just one dev's ad hoc measure
• qlik_list_dimensions (App 2) → check if there's a master dimension confirming this is the governed metric — finds none, it's a sheet-level expression only
• No way to confirm confidence — no glossary entry, no certified master measure, no source of truth for what "conformance health" is supposed to mean

Now does the token cost and accuracy for the image make sense for the Without Context Engineering side make sense? Maybe it will make even more sense if we now look at the With Context Engineering side. 

• qlik_search_glossary_terms → "conformance" — 1 hit: 🧩 Conform Health (Concept) — conceptual umbrella term, drafted by Dalton Ruer
• Read definition — this term isn't itself a master item, it's a pointer explaining that health has two complementary signals that "should always be read together"
• Follow Related terms → Conform Success Rate (See also) and Conform Parity Rate (See also)
• qlik_search_glossary_terms → "Conform Success Rate" — resolves to certified master measure + confirms logic: FAILURE if conform_row_count_delta ≠ 0 OR conform_source_record_count = 0
• qlik_search_glossary_terms → "Conform Parity Rate" — resolves to its own certified master measure
• qlik_get_chart_data on both master measures — pull Success Rate and Parity Rate together, since the glossary explicitly warned against reading either in isolation
• Synthesize a single answer combining both signals — e.g. "Success Rate is 98% but Parity Rate flags a partial mismatch, so there's a silent under-count worth investigating"

You have to LOVE that difference. Instead of Claude seeing your entire tenant environment and hoping to find the right application it starts by using the glossary and finds the concept needed to respond to your question. A concept that is linked to the right application to answer the question. Not only is it going to find a single answer for you (either of the metrics) it's going to provide a thorough response that includes both of the measures. 

Context Engineering

I kind of teased about the fact that all your end users thoroughly understand every single measure in all of your applications and that it was ok if you didn't provide descriptions because they would never need to "show details" to understand them. Hopefully, you now see the bigger picture of the world we live in now. Those same master items, same descriptions it's allowed you to build for 12 years Qlik Sense has been around, now provide the backbone for Context Engineering. It's the meat that allows you and your end users to ask questions with Qlik Answers or Qlik MCP through chat clients and get accurate and thorough answers using less tokens. 

Remember when I showed the image and said that Dalton Ruer was identified as the "steward" and said ignore it for now. Well now I need you to recall it and consider this. With a simple skill in place for Claude (whatever client you use) you could enhance the returned answer even further by taking advantage of all of the bits in the business glossary.  

• Glossary term found, no steward/status → medium confidence — there's a governed definition, but no accountable owner backing it
• Glossary term found, drafted/owned by a named Steward (like "Draft by Dalton Ruer" on Conform Health) → higher confidence — someone is accountable for that definition being correct, not just documented
• Glossary term is "Published/Certified" status with a Steward → highest confidence tier — governed, owned, and formally approved

Summary

I started by saying I spent 10,000 words explaining something that I needed to create that image for. Probably seemed like an exaggeration but now I'm guessing you believe me. Even more, I hope this solidifies how important Master Measures have always been, and how vital the business glossary is for context engineering. I will follow up with a part 2 soon to explain the simple skills needed to add that TRUST SCORE information to responses. Because my fundamental belief and something I've said for 16 years is that "You can't ACT ON insights that you DON'T TRUST."