Solver Certificates for Faithful Mission Planning Explanations

🔑 Enhanced Key Takeaways

• •Paper submitted to arXiv on February 28, 2026, by sole author Najeeb Khan, marking its initial public release just days before the article date[2].
• •Addresses a specific gap in prior art: no existing methods for space mission scheduling provide solver-grounded explanations with formal faithfulness guarantees[1].
• •Certificate extraction latency is sub-second per certificate (solve 6 ms + extraction 440 ms), dominated by O(k·c) solver calls for MIS extraction where k is core size and c is candidate constraints[1].

🛠️ Technical Deep Dive

• •Explanations derived directly from optimization model: minimal infeasible subsets (MIS) for rejections via deletion algorithm, tight constraints and contrastive trade-offs for selections, inverse solves for what-if queries[1][2].
• •Faithfulness formalized via three properties to prevent failure modes: soundness (verifiable over solver model), counterfactual validity, and stability (e.g., Jaccard=1.0)[1].
• •MIS extraction independent per order, trivially parallelizable; pre-filtered orders (visibility, cloud cover) and optimality trade-offs skip full MIS computation[1].
• •Scalability: per-certificate cost 508 ms to 4.1 s for growing models, but operational batches (25-50 orders) complete under 15 s, up to 30 s for larger[1].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline