window specifications, and adjusting roof insulation levels. Each scenario requires running a physics-based thermal simulation that takes 1–2 seconds. Evaluating even a subset of 500 buildings with 15 scenarios each would require over 7,500 simulation runs, taking over 3 hours to complete. This is far too slow for the interactive, exploratory workflow that analysts need. This performance bottleneck is not unique to building energy; similar challenges arise whenever digital twins rely on computationally expensive simulations. Current digital twin platforms and time-series databases treat each what-if query independently [4, 5], running a full simulation for every parameter variation. This approach misses a critical opportunity: scenario exploration is inherently incremental. Analysts do not jump randomly between distant parameter values; instead, they explore nearby configurations in a structured manner, for example, testing wall U-value 0.18, then 0.19, and then 0.20, gradually refining their understanding of the solution space. This incremental exploration pattern creates an opportunity for dramatic performance improvements if we can exploit the similarity between consecutive queries. We present TwinDB, a scenario-aware database system that introduces Incremental Scenario Maintenance (ISM), a technique that incrementally updates cached simulation results when parameters change slightly. The key insight enabling ISM is that physics-based simulations often exhibit local linearity: small parameter changes produce proportional output changes. For building energy simulation, heat loss is linear in thermal transmittance according to the fundamental heat transfer equation 𝑄 = 𝑈 × 𝐴 × Δ𝑇 × 𝑡, where 𝑄 is heat energy, 𝑈 is thermal transmittance, 𝐴 is surface area, Δ𝑇 is temperature difference, and 𝑡 is time. By learning sensitivity coefficients (e.g., “energy changes by 14.4 MWh per W/(m2 ·K) of wall U-value”), TwinDB can predict outcomes for nearby scenarios in 𝑂 (1) time rather than running the full 𝑂 (𝑇 ) simulation for 𝑇 timesteps. This demonstration showcases TwinDB using real-world data from the Danish Smart Heat Meter dataset [6]. Conference attendees can interact with a visual workspace (Figure 1) that enables them to explore retrofit scenarios and observe ISM speedups in real-time, verify accuracy against full simulation, compare historical versus simulated energy consumption side-by-side, write TwinQL queries with syntax highlighting and instant results, and upload their own CSV time-series data to experience the system with personalized data. Digital twins enable what-if analysis by simulating physical systems under hypothetical conditions, but simulation is computationally expensive. In this context, a scenario represents a configuration of input parameters (e.g., wall insulation, window properties, heating setpoints) defining a hypothetical condition to evaluate. A building energy analyst exploring retrofit options for 500 buildings faces hours of waiting time for a typical session. We demonstrate TwinDB, a scenario-aware database system that achieves significant speedup through Incremental Scenario Maintenance (ISM), which incrementally updates cached simulation results when parameters change slightly rather than re-running full simulations. Our interactive visual workspace lets conference attendees (1) explore retrofit scenarios with instant feedback, (2) compare scenarios side-by-side with confidence intervals, (3) verify ISM accuracy against full simulation, (4) write declarative TwinQL queries (a domain-specific language extending SQL for scenario-based analytics), and (5) upload custom CSV data. Built on PostgreSQL with TimescaleDB and fully open-source, TwinDB transforms digital twin analytics from batch processing to interactive exploration.