GDAL MCP

# Educational Design Principle ## A Moment of Honesty When was the last time you chose a resampling method not because it was the best choice, but because it was the one you knew? When did you last pick a compression algorithm because "that's what everyone uses" rather than understanding the tradeoffs? How many times have you scrolled past an option in a dropdown menu—kriging, spline, inverse distance weighting—and thought "I should learn what that does someday," before clicking your usual choice and moving on? This isn't a criticism. It's the reality of working in a complex field with limited time and endless deadlines. We develop habits, we find approaches that work well enough, and we stick with them. There's wisdom in that—consistent methods lead to predictable results. But there's also opportunity cost: a vast landscape of knowledge that could improve our work, if only we had time to explore it. We are all standing in front of an enormous library, but we keep reading the same handful of books. ## The Discovery In the early days of ChatGPT, something remarkable happened. With careful prompting—asking not just for answers but for *reasoning*, for *tradeoffs*, for *alternatives*—geospatial practitioners discovered they could engage with what felt like PhD-level domain expertise. Not to replace their own knowledge, but to fill in the gaps. To explain that mysterious parameter. To suggest the method they'd never heard of. To ask, gently, "Have you considered why this approach might not preserve the property you care about?" It wasn't magic. The knowledge was always there, scattered across textbooks, buried in documentation, hidden in the heads of specialists we'd never meet. But suddenly it was *accessible*. Conversational. Patient. Available at 2 AM when a deadline loomed and a decision needed to be made. This wasn't about AI replacing geospatial scientists. It was about AI helping us become better ones. ## The Habit Trap We choose bilinear resampling because we saw someone else use it once. We always use LZW compression because it worked last time. We stick with Web Mercator because "everyone uses it for web maps." We never touch cubic spline because we're not entirely sure what it does differently from cubic. We avoid certain tools because they seem complicated, even when they might be exactly what we need. These habits aren't wrong—they're survival strategies in a field that demands we make dozens of methodological decisions every day while also meeting deadlines, managing stakeholders, and solving actual problems. But habits, left unexamined, can calcify into constraints. We stop asking "what's possible?" and start asking "what's familiar?" The opportunity isn't to abandon our habits. It's to understand them. To know *why* bilinear works for this task. To learn when LZW is suboptimal. To recognize the specific case where Web Mercator's distortion matters. To discover that cubic spline isn't scary—it's just a tool with specific tradeoffs. ## What Agents Can Offer Imagine working alongside someone who: - **Asks gentle questions**: "I see you specified nearest neighbor for elevation data. This preserves exact values but creates blocky artifacts that affect slope calculations. Would you like to explore cubic for smoother results, or proceed as specified?" - **Explains the unfamiliar**: "Kriging is a geostatistical interpolation method that estimates values at unsampled locations by modeling spatial autocorrelation. It's particularly useful when you have sparse measurements and need to quantify prediction uncertainty. For your regular grid, though, simpler methods like cubic spline might be more appropriate." - **Surfaces alternatives**: "DEFLATE with predictor=2 typically achieves better compression than LZW for continuous rasters like this DEM, at similar speeds. The predictor mode exploits horizontal correlation in the data. Would you like to compare options, or proceed with LZW?" - **Documents reasoning**: "Choosing EPSG:5070 (Albers Equal Area) for this national land cover analysis preserves area accuracy, which is critical for your 'acres per class' statistics. Web Mercator would distort area significantly at these latitudes." This isn't about challenging your expertise—it's about *augmenting* it. About being the colleague who remembers the tradeoffs when you're focused on the deadline. The patient teacher who explains without judgment. The methodological checklist that runs in the background while you focus on the science. ## Learning Through Doing Day after day, we're confronted with information that could help us if only someone were there to contextualize it: - That parameter in the documentation that seems important but unclear - The warning message that probably means something but we've learned to ignore - The dropdown full of options where we always pick the first one - The method we've heard mentioned at conferences but never tried - The tradeoff between speed and quality that we make without quite understanding the cost With the right feedback—*educational* feedback, not gatekeeping, not blocking—every decision becomes an opportunity to learn: - Choose a method → Understand why it fits - See an alternative → Learn when it's better - Encounter an unfamiliar term → Get it explained in context - Make a tradeoff → Understand what you're trading Not through lengthy tutorials or mandatory training, but through gentle, contextual guidance embedded in the work itself. The best time to learn about resampling methods is *when you're choosing one*. The best time to understand CRS distortion is *when you're picking a projection*. ## The Core Value **GDAL-MCP is designed with educational value as a first-class principle.** This isn't about: - ❌ Replacing professional expertise with automation - ❌ Creating barriers that slow down experts who know what they're doing - ❌ Forcing justification for every decision - ❌ Teaching you things you already know - ❌ Making you feel inadequate for not knowing something This is about: - ✅ **Learning in context**: Explanations appear when relevant, not as prerequisites - ✅ **Respecting expertise**: Explicit choices are documented, not questioned - ✅ **Gentle guidance**: Concerns are raised conversationally, not as blockers - ✅ **Building knowledge**: Each interaction leaves you slightly more informed - ✅ **Growing the field**: Helping everyone access the collective wisdom of geospatial science ## The Advisory Pattern Our implementation reflects this philosophy through **advisory rather than prescriptive guidance** (see [ADR-0026 Amendment](../ADR/0026-epistemic-governance.md)): **When you specify a method explicitly**: The system documents why it's appropriate for your stated goal. If there's a concern you might not be aware of, it asks conversationally. It never blocks. It never lectures. It partners. **When the AI chooses autonomously**: It explains its reasoning so you can learn from the choice. Not "I picked bilinear," but "I chose bilinear because this continuous data benefits from smooth interpolation, balancing quality and performance." **When concerns arise**: Educational intervention happens through conversation, not gatekeeping: ``` Instead of: "ERROR: You cannot use nearest neighbor for continuous data" We offer: "I notice you specified nearest for elevation data. This creates blocky artifacts that affect slope calculations. Would you like cubic for smoother results, or proceed with nearest as specified?" ``` The difference is respect. Trust. Partnership. ## Resources as Teachers Beyond prompts, we provide [comprehensive reference resources](../ADR/0023-resources-manifest.md): - **Compression methods** with guidance on when to use each - **Resampling methods** with explanations of categorical vs continuous data - **CRS/datum information** for understanding projection choices - **Format characteristics** for choosing appropriate outputs These aren't just lookup tables. They're designed to be educational: clear descriptions, use case guidance, tradeoff explanations. The AI consults them naturally, learns from them, and shares that knowledge contextually. You're never forced to read documentation. But when you need it, it's there—clear, accessible, and embedded in the workflow. ## The Vision Imagine a geospatial science where: - Junior practitioners learn best practices naturally, through contextual guidance rather than trial and error - Mid-career professionals discover techniques they'd never encountered, expanding their toolkit - Experts are freed from teaching the fundamentals over and over, with AI handling the "why do we use this projection?" questions - Methodological knowledge isn't locked in the heads of specialists but accessible to anyone doing the work - The field grows not by replacing human expertise but by making it more accessible, more shareable, more democratic This is the opportunity that epistemic AI offers: not to automate away geospatial science, but to make it more learnable, more transparent, more available to those who want to do it well. ## Not a Classroom, a Workshop We're not building a training simulator. We're building a *working environment* where learning happens naturally: - No mandatory tutorials - No certification requirements - No knowledge gates - No "you must learn this before proceeding" Just: work alongside an assistant that explains when you're curious, checks when you might be making a mistake, documents so you can trace your reasoning later, and quietly teaches you something new with each interaction. The best education is the kind you don't notice happening—until one day you realize you understand kriging, you've internalized the difference between bilinear and cubic, you can explain why that CRS matters, and you're making better decisions without even thinking about it. ## An Invitation This document isn't a requirement. It's an invitation. An invitation to be curious about the dropdown options you've always ignored. To ask "why?" when choosing a method. To learn from the alternatives you didn't pick. To understand the tradeoffs you're making. To grow as a geospatial scientist, one decision at a time. The tools are here. The knowledge is accessible. The AI is patient. The only question is: are you ready to keep learning? --- *"The beautiful thing about learning is that no one can take it away from you."* — B.B. King *"Tell me and I forget. Teach me and I may remember. Involve me and I learn."* — Benjamin Franklin (attributed)