Overview
Simulation Sandbox lets users create hypothetical scenarios by injecting synthetic documents into a sandboxed copy of their intelligence space, then observe how the entire intelligence landscape changes. It answers the question: "What would our knowledge base look like if...?"
While Scenario Simulator reasons over existing data, Simulation Sandbox lets you change the data landscape and see second-order effects across all intelligence features.
The Problem
- Strategic planning (M&A, market entry, regulatory change) requires understanding how new information would affect the organization's knowledge landscape
- There's no way to test hypotheses about information without actually modifying the production knowledge base
- Due diligence teams want to preview how an acquisition target's documents would integrate with existing knowledge
- Regulatory impact assessments need to model how new requirements interact with existing compliance coverage
- "What if" analysis on structured data exists (financial models, BI tools) but not on unstructured knowledge
How It Works
- Sandbox creation — User creates an ephemeral fork of their space's intelligence state (embeddings, feeds, clusters, compliance matrices)
- Document injection — Upload or generate synthetic documents into the sandbox
- Full pipeline processing — Run the complete pipeline on synthetic docs: chunking → embedding → metadata extraction → feed assignment
- Intelligence re-evaluation — Re-run intelligence features against the augmented corpus: exploration prep, blindspot detection, compliance mapping, wiki generation
- Diff view — Show side-by-side comparison of sandbox vs. baseline: new clusters, shifted topics, changed coverage, new connections
- Multi-sandbox comparison — Compare multiple sandboxes side by side (e.g., Acquisition A vs. Acquisition B)
- Ephemeral lifecycle — Sandboxes auto-delete after a configurable period or manual dismissal
User Story
A corporate development team is evaluating acquiring Company X. They inject Company X's public filings, product documentation, and Glassdoor reviews into a sandbox. The system re-runs exploration prep and shows: "With Company X's docs added, 3 new clusters emerged (product overlap, cultural tension, IP synergy). Your Blindspot Detector now shows 2 fewer gaps in the Asia-Pacific region. Compliance Mapper shows 4 new regulatory requirements from Company X's industry." They repeat for Company Y and compare the two scenarios side by side — the diff view clearly shows Company X is a better strategic fit but brings more regulatory complexity.
Complexity & Timeline
| Aspect | Detail |
|---|---|
| Complexity | Complex |
| Estimated Build | 6–8 weeks |
| Platform Dependencies | Full ingestion pipeline, Explore (prep agent), Blindspot Detector, Compliance Mapper |
| New Infrastructure | Sandbox isolation (forked collections + tagged DB records), diff engine, multi-sandbox comparison UI |
Target Clients
- Personas: Corporate Development, M&A Teams, Strategy Directors, Board Advisors, Regulatory Affairs
- Verticals: Private Equity, Corporate Strategy, Financial Services, Pharmaceutical, Defense
- Pitch: "War-game your knowledge landscape — see how new information changes everything."
Revenue Potential
Executive-facing strategic planning tool that speaks the language of board rooms and investment committees. Unique capability: no knowledge management tool offers "what-if analysis on your knowledge base." Supports premium per-sandbox pricing given the compute intensity. Natural bundling with Scenario Simulator for a comprehensive "strategic planning" package. High-value for M&A due diligence, regulatory impact assessment, and market entry planning.
Feature Synergies
- Scenario Simulator — Simulator reasons over existing data; Sandbox changes the data landscape — complementary approaches to strategic analysis
- Blindspot Detector — Run gap analysis in the sandbox to see how new documents would fill existing blindspots
- Compliance Mapper — Preview how an acquisition or new product line would affect compliance coverage
Risks & Open Questions
- Sandbox isolation requires forked Qdrant collections and temporary DB records — significant infrastructure complexity
- Re-running exploration prep per sandbox is computationally expensive — may need a lightweight "quick scan" mode
- Storage management: sandboxes must have TTL to avoid unbounded growth
- Diff quality depends on deterministic pipeline outputs — stochastic LLM steps may produce noisy comparisons