TrueLanded

**TrueLanded — Evidence-to-Decision Engine** In international trade, a single documentary discrepancy can delay shipments, increase costs, and expose businesses to compliance risks. When a packing list declares 28 pieces and an invoice declares 31, extracting information is not enough. The real challenge is determining what the evidence supports, what remains uncertain, and what must be verified next. TrueLanded addresses this problem through an evidence-first AI architecture that separates probabilistic information extraction from deterministic documentary verification. Built for the Meta Global AI Developer Hackathon, our standalone demonstrator combines five capabilities: 1. **Evidence Spotlight:** Connects documentary claims to exact source passages, establishing transparent and traceable evidence. 2. **Next Best Proof:** Identifies missing or conflicting documentation and recommends the next verification step without taking unauthorized action. 3. **Counterfactual Cascade:** Demonstrates how changing a documentary variable affects the resulting analysis while preserving unresolved obligations. 4. **Tamper Detection:** Uses independently pinned evidence references to detect altered source documents and block unsupported recommendations. 5. **Adjudication Lab:** Demonstrates defensive behavior against 30 synthetic adversarial cases using reproducible, precomputed Python results. Historical testing with Meta Muse Spark supported candidate field extraction. Independent verification evaluated 18 exact source literals, while 10 separate Muse Code High evaluations examined synthetic cases. The browser demonstration operates offline without live model inference. All demonstrated commercial documents are fictional, and human judgment remains essential. TrueLanded's fundamental innovation is moving beyond asking AI, "What does this document say?" toward a more consequential question: **"What can the available evidence actually prove, and what should a human verify next?"** The competition submission is an isolated demonstrator developed for this hackathon, separate from the pre-existing commercial TrueLanded project.