> ## Documentation Index > Fetch the complete documentation index at: https://docs.semilattice.ai/llms.txt > Use this file to discover all available pages before exploring further. # Changelog > What moved, what shipped, what we're thinking about. ## Logged-in user journey simulations. Notifications. You can now run user journey simulations on logged-in user flows. Add test account credentials and watch simulated users navigate dashboards, account settings, private content feeds, and anything behind a login wall. We've also added in-app and email notifications, so you can trigger simulations, get on with something else, and be notified to return when they're done. ## MCP v2: agentic validation with simulated users. The Semilattice MCP now supports full User Research and User Journey simulations, enabling rapid validation at any point in your product or marketing workflow. Use it for discovery, concept validation, feature prioritisation, message testing, conversion flow optimisation, UX comprehension, and usability testing. See the [MCP introduction](/mcp/introduction) to get started. ## Automatic simulation titles. Simulation titles are now set automatically when the simulation plan is generated, based on the simulation's parameters. Works for both user research and user journey simulations. You can rename a simulation from the simulations list or from its results page. ## Multi-segment simulations. You can now run a single simulation across multiple user models. Toggle **Compare multiple** on the user model selection step, add the user models you want to test, and run. Results come back segmented by user model: paired bars in the predictions pane so cross-segment differences read at a glance. Use it to understand your users' pain points or differences; or compare how messages or ideas resonate with different user segments in one simulation. ## Compare options in user journey simulation. You can now paste two URLs instead of one. Pick a user model, and simulated users navigate both versions. Results come back paired: completion rates, page flow, time on task, and per-user observations for each variant. Because the data comes from the user model rather than real traffic, neither URL needs to be your own. Test a draft against what's in production. Test your page against a competitor's. Test today's site against last year's from the Internet Archive. Any two live URLs work. The rest of the flow is unchanged. Pick a test type, paste the URLs, pick a user model, review the generated task and success criteria, and run. The "Compare two options" toggle sits on the Define step. ## User journey simulation. Open sign-up. Simulated users can now navigate your product. Give Semilattice a URL, pick a user model, and watch them click through your product in minutes. See where they get stuck, what they find and what they miss. Five modes. Findability, conversion flow, comprehension, discovery, or define your own. The system writes the test scenario for you. Describe what you want to learn, pick a user model, and it generates a realistic user task and success criteria. Review, edit if you want, then run. Results show completion rates, page flow, per-user session replays with screenshots, and written observations describing exactly what happened. Key findings and recommendations extracted automatically. Export as PDF or share via URL. Sign-up is open at semilattice.ai. ## Results you can act on. Export and share. Simulation results are now organised for decisions. We extract findings that are well-supported by the evidence, highlight noteworthy things you might otherwise miss, and list specific recommendations for next steps. Charts sit in a side pane so you can read the synthesis and check the data at the same time. The most important predictions come up first. Every report shows the user model's prediction accuracy, so you can calibrate your confidence in the results. New export options make it easier to take Semilattice findings into your process. Download a PDF to share with execs, a CSV for your own analysis, a Markdown file for Claude or Notion, or chart images for embedding anywhere. And you can now share results with colleagues via a URL. Context extraction is richer. The product pulls more from your URL and understands more about your product before you ask your first question. ## User models, redesigned. Product testing in alpha. Complete redesign of the user model experience. Every user model now has a detail page with accuracy scores and accuracy badges so you can gauge reliability at a glance. Drill into individual questions to compare ground truth against predictions with paired bar charts. Search and filter across everything. [Read more on the blog →](https://semilattice.ai/blog/audiences-redesign) User journey simulation (alpha) gives you simulated users that interact with live websites as specific user segments would. They navigate, convert, get confused, miss things. ## 91% accurate against the FCA Financial Lives Survey We tested our UK Consumer Finance user model against the FCA's most comprehensive benchmark: 207 questions it had never seen, covering trust, financial literacy, product usage, and demographics. Overall accuracy: 91%. The interesting part is where it gets things wrong. The model knows the right answer to financial literacy questions, but real people often don't. It predicts what people *should* think, not always what they *do* think. That gap is the research question. [Read the full analysis →](https://semilattice.ai/blog/fca-comparison) ## New landing page, new colour language, in-app login The entire public face of Semilattice has been rebuilt. New landing page with an interactive decision tree hero, animated product walkthroughs, and a colour system built around indigo, pink, and teal. Login and signup now happen inside the app. ## Decision tool: from question to evidence in minutes New guided flow for product decisions. Describe what you're considering, choose a user model, and Semilattice designs the research, runs it against simulated user models, and returns a report with key insights, survey results, and recommendations. Minutes, not weeks. Built for the moment someone asks on a Sunday whether you should move that feature behind the paywall. [Read more →](https://semilattice.ai/blog/less-noise-in-early-decisions) ## 16 segments of financial life 16 UK Consumer Finance user models, from Growing Families to Just About Managing. Each segment built on real research data with distinct demographics, financial behaviours, and attitudes. Every product decision in consumer finance is a "what if," and the people that matter to you aren't one group. Now you can ask each segment separately. [Explore the segments →](https://semilattice.ai/blog/fintech-user-models) ## Semilattice MCP [Semilattice MCP](/mcp/introduction) lets AI assistants and agents predict how specific user models answer questions. Connect your IDE, your agent framework, or any MCP-compatible tool. Four tools: `list_population_models`, `create_prediction`, `get_prediction`, `wait_for_prediction`. Your AI can ask your user model a question and wait for the answer programmatically. [Connect yours →](/mcp/connect) ## Test batches: evaluate accuracy for your use case New evaluation feature. Upload ground truth data for your specific domain and test how well a population model performs on the questions that matter to you, not just the seed data it was built on. Cross-validation tells you the model's general accuracy. Test batches tell you whether it's accurate *for your problem*. [Learn how →](https://semilattice.ai/blog/test-batches) ## Batches, 3× faster predictions, simplified API `API v1.1.0` · `SDK v0.6.0` #### Batches Group predictions with a name and description. Fetch all predictions in a batch with `/predictions/batch/{batch_id}`. #### Speed Predictions now return in \~20 seconds, roughly 3x faster than before. #### New methods * `GET /populations` — list all available population models * `GET /predictions|tests/batch/{batch_id}` — fetch batch details #### Simplified naming We renamed methods and fields to make the API easier to read. Old names still work but are marked deprecated. * `POST /answers` → `POST /predictions` * `POST /answers/benchmark` → `POST /tests` * `simulated_answer_percentages` → `predicted_answer_percentages` * `kullback_leibler_divergence` → `information_loss` ## The Semilattice API: user insights as infrastructure `API v1.1.0` · `SDK v0.5.2` V1 release. Predict user behaviour like you make database queries. * Create population models with custom seed data * Test model accuracy with built-in cross-validation * Test against external ground truth * Predict new questions [Get started →](/introduction) ## Humans + Time Human systems are complex, emergent, unpredictable, and fundamentally resistant to the modelling techniques that work for physical and biological systems. The human element has held back modelling because we lacked both the data and the computational models to work with it. Large Language Models change this. They contain enough encoded human behaviour to simulate how specific groups of people would respond to questions they've never been asked. That's the foundation Semilattice is built on. [Read the full essay →](https://semilattice.ai/blog/humans-time) ## Introducing Semilattice Semilattice gives you an AI user model that answers questions like they do, in seconds. 15 ready-to-use public models covering tech industry segments, from designers and front-end developers to data scientists and engineering managers. The decision-making balance has tipped toward action and away from research. Computational research tips it back without delaying action. Custom models from your own data. Public models to start immediately. [Read the introduction →](https://semilattice.ai/blog/introducing-semilattice)