Show HN: Multi-attribute decision frameworks for tech purchases
4 comments
The 'without SEO bias' point is the most interesting part here. We're already seeing users trust LLM synthesis more than direct search precisely because it (theoretically) filters out the affiliate spam.
But aren't we just moving the problem up a layer? Brands are aggressively optimizing for LLM visibility now ('Generative Engine Optimization'). If your framework relies on LLM training data or RAG, it's still downstream of whatever content dominates the web.
Curious if you've seen any 'hallucinated brands' or persistent bias towards certain vendors in your tests across models? e.g. does Gemini favor Google products in your laptop comparison?
But aren't we just moving the problem up a layer? Brands are aggressively optimizing for LLM visibility now ('Generative Engine Optimization'). If your framework relies on LLM training data or RAG, it's still downstream of whatever content dominates the web.
Curious if you've seen any 'hallucinated brands' or persistent bias towards certain vendors in your tests across models? e.g. does Gemini favor Google products in your laptop comparison?
I have not seen any brands hallucinated. The biggest issue was the math as you pointed out. I created an if-then-else if statements in the prompt to force the use of code/Python for calculations.
The free version of Gemini functions less well than the others because it executes my instructions less rigorously, but I have not seen product prevalence over others. I'm also not sure how I could track though without running the same case study several dozen times to see if anything statistically significant comes ups or changes.
At the end of the day I would say my idea is best for getting you from 5% knowledge on a topic to an 80% level and a much more advanced and objective place to make a decision and finish out with your own "eyes-on" research.
The free version of Gemini functions less well than the others because it executes my instructions less rigorously, but I have not seen product prevalence over others. I'm also not sure how I could track though without running the same case study several dozen times to see if anything statistically significant comes ups or changes.
At the end of the day I would say my idea is best for getting you from 5% knowledge on a topic to an 80% level and a much more advanced and objective place to make a decision and finish out with your own "eyes-on" research.
This is a really interesting application of LLMs. The lack of "repeatable, traceable results" is indeed a huge issue for any serious use case (we see this constantly in enterprise adoption).
Have you found that forcing the LLM into a structured scoring framework reduces its tendency to hallucinate specs? Or does it just hallucinate the scores with more confidence?
Also, curious if you've tried different models for the "scoring" vs "reasoning" steps. We've found Claude is much better at adhering to complex constraints than GPT-4o for tasks like this.
Have you found that forcing the LLM into a structured scoring framework reduces its tendency to hallucinate specs? Or does it just hallucinate the scores with more confidence?
Also, curious if you've tried different models for the "scoring" vs "reasoning" steps. We've found Claude is much better at adhering to complex constraints than GPT-4o for tasks like this.
Most of the language I use in my prompting is structured around weeding out hallucinations. At each step along the way I ensure I'm asked to confirm the previous step's output.
The balance I have been trying to find is between "show me all you work" so you know the math is all there and correct (and perfect for enterprise) and then trying to tell friends or my parents who don't care to see the math (the background) that it needs to be there for this to work right. I have seen scores "ballparked" by ChatGPT. The rankings didn't change in the end, but the scores were a couple tenths off.
I've used ChatGPT 5.2, Claude and Gemini, but have never switched between steps which sounds interesting! I have found the same as you with Claude. ChatGPT is a close second and Gemini doesn't give me the type of response I'd prefer to keep things smooth and traceable. I'm looking to buy a new car right now and each of the 3 models has given me the same top 3 each time so I find that as re-assurance.
The balance I have been trying to find is between "show me all you work" so you know the math is all there and correct (and perfect for enterprise) and then trying to tell friends or my parents who don't care to see the math (the background) that it needs to be there for this to work right. I have seen scores "ballparked" by ChatGPT. The rankings didn't change in the end, but the scores were a couple tenths off.
I've used ChatGPT 5.2, Claude and Gemini, but have never switched between steps which sounds interesting! I have found the same as you with Claude. ChatGPT is a close second and Gemini doesn't give me the type of response I'd prefer to keep things smooth and traceable. I'm looking to buy a new car right now and each of the 3 models has given me the same top 3 each time so I find that as re-assurance.
Copy-paste LLM prompts that turn ChatGPT or Claude into a structured decision analyst for laptops, monitors, tablets, phones, and SaaS subscriptions. You define constraints, weight what matters to your workflow, and get scored recommendations with sensitivity analysis. Why I built this:
With the rise of LLMs (AI), I wanted to find a way to harness the computing power and ease of use the chat interface provides. The major problem: LLMs don’t always provide repeatable, traceable results if you ask the same question twice or even against 2 competing products. That is the dilemma this product aims to solve. Is this a PDF, yes, but it harnesses my systems analysis experience to help hard-code a framework for a person off the street to turn their AI chat box into an objective decision-helper in just 15 to 20 minutes of use.
I spent 10+ years applying decision science in defense and systems analysis—graduate work at Naval Postgraduate School, leading teams through decisions where the cost of choosing wrong wasn't just money; it was mission failure or lives at risk.
The method used uses multi-attribute utility theory: define hard constraints (binary gates), eliminate non-viable options, score remaining candidates on mission-critical attributes with explicit weights, then run sensitivity analysis to see what changes the outcome.
I use this myself all the time. The most recent was trying to upgrade my own laptop (Surface Pro stuck at Windows 10).
BLUF benefits:
• Helps prevent over-obsessing over specs (32GB RAM! RTX 4080!) while ignoring mission fit (do I really game that often?)
• Fleshes out hard constraints that sometimes come up until after purchase (bought Windows laptop, needs a way to support a MacOS app)
• Future-proofing: ensuring I won’t pay feature I'll statistically never use
• Aims to parse through the noise (SEO type posts) and get you a great first-pass research report of what you should value and why.
Consumer purchases don't need full enterprise rigor, but they deserve better than "Top 10 Laptops 2026" affiliate listicles or chatbots hallucinating specs.
How the framework works: 1. Mission definition: What must work reliably? (Video editing vs office work vs travel)
2. Hard constraints: Binary gates (budget ceiling, OS requirements, battery minimums)
3. Candidate generation: AI searches current market without SEO or affiliate bias
4. Weighted scoring: Performance, battery, reliability, portability—you control the weights
5. Efficient frontier: Which options dominate? Which are just expensive?
6. Sensitivity analysis: "If battery life matters 25% instead of 15%, MacBook Air wins. If reliability matters more, ThinkPad wins."
The PDFs include example case studies I’ve developed: policy analyst choosing ThinkPad X1 Carbon over MacBook Air (why reliability and docking beat battery life for enterprise work), freelance designer choosing Figma over Affinity Designer (why collaboration features justified 6x higher cost), consultant choosing Obsidian over Notion (why offline-first beat ease-of-use).
No barriers: No sign-up. No account. You get a PDF with prompts and case studies. Open ChatGPT or Claude (free version works), paste the prompt, answer questions. That's it.
I built this because I was tired of seeing people (and myself) wasting money on impressive-sounding specs that don't match their actual workflow. If you've ever regretted a tech purchase 3 weeks later, this might help.
Try it (I'd offer it free but then I loose my IP): • Tech & Electronics: https://decisioncontrolworks.gumroad.com/l/auzhsa • Software & Subscriptions: https://decisioncontrolworks.gumroad.com/l/zaucxt
Curious what HN thinks—especially if anyone's tried applying formal decision methods to everyday purchases.