Project Title + Description
Mirrors Google Form fields 9 (descriptive title) and 10 (project description, 3000 character hard limit). Post-evaluator-hostile-review revision (2026-05-07).
Field 9 - Descriptive Title of Proposed Research
Title (paste-ready):
Interpretable Anomaly Detection for Tokenised Real Estate and RWA Markets in the UAE: An Open SupTech Toolkit Anchored to AUS
Field 10 - Project Description (3000-char limit)
EOI BODY (paste-ready)
The UAE Securities and Commodities Authority’s Decision 15/RM/2025 (June 2025) established the Gulf’s first dedicated regime for commodity-token contracts. It also created the UAE’s first comprehensive framework covering both security tokens and commodity tokens. The Dubai Land Department, VARA, and Prypco Mint launched the first live tokenised real-estate pilot in May 2025. UAE on-chain real-world assets crossed roughly US$17B by end-2025. No open SupTech tooling deployable to the UAE token taxonomy currently detects manipulation, oracle decoupling, or tail-risk events with regulator-readable explanations.
We propose rwa-suptech, an open MIT-licensed library for tokenised real-estate market supervision. Multimodal data fusion across on-chain transfers, oracle feeds, news arrivals, and order-book deltas feeds the detector. Anomaly detection runs through interpretable extreme-value-theory methods. Counterfactual explanations come from a diffusion-model layer drawn from AR-Pro (NeurIPS 2024). Verifiable alert provenance follows the Verifi methodology (TDSC 2024) led by Co-PI Youcheng Sun. An agentic AI layer orchestrates detection, explanation, and verification into a supervisor-readable workflow. Differential-privacy aggregation gates open dataset releases. FRG23 and FRG24 at AUS plus the joint Physica A 2024 paper anchor the methodological precedent. Foundation-model tools serve inference only.
Specific aims: (1) Develop the interpretable EVT detector with counterfactual explanations. (2) Release an open dataset of privacy-preserving derived statistics from tokenised real-estate activity, CC-BY year 1. (3) Deliver rwa-suptech under MIT by year 2. (4) Validate on at least two real or pilot data streams by month 24. (5) Train two UAE-based PhDs at AUS and MBZUAI and launch the AUS course Interpretable AI for Tokenised Markets.
Application-readiness arc. Months 0-6 deliver the Applied baseline. Months 6-12 produce a prototype on synthetic and public on-chain real-estate data plus the first AUS workshop. Months 12-18 reach Tech Development with real-data validation against the DLD and Prypco pilot subject to access agreement. Months 18-24 reach the first Validation step. A named user from UAE regulators or the COST 19130 advisor pool runs rwa-suptech with measured detection latency by month 24.
PI Stephen Chan is at AUS Mathematics & Statistics. Co-PI Joerg Osterrieder chairs COST 19130 and coordinates MSCA-DN Digital Finance. Co-PI Youcheng Sun at MBZUAI Computer Science leads verification and AI safety. UAE delivery anchors at AUS plus MBZUAI through Co-PI presence. One UAE-based PhD is recruited at AUS under Stephen, one at MBZUAI under Youcheng with Stephen as co-supervisor. UAE regulator outreach to SCA, ADGM RegLab, DFSA, and DLD/VARA runs during the full proposal phase. Tokenised commodities are a Year-2 extension contingent on SCA Phase-1 issuer availability. All code MIT, methods/datasets CC-BY.
END BODY
Notes on this draft
- Language does not use “causal” as a methodology label. Counterfactual explanations are described as descriptive interpretability, not treatment-effect identification.
- Substrate is reframed as FRG23 + FRG24 joint methodological precedent producing papers and smaller code modules; the new MIT codebase is built from scratch in the project.
- Domain is real estate primary (Prypco / DLD / VARA pilot live since May 2025); commodities are a Year-2 extension contingent on SCA Phase-1 issuer availability.
- Regulator user is honestly framed: UAE regulator outreach is “during full proposal phase”; the COST 19130 EU regulator participant pathway is the documented Year-1 advisor fallback.
- Capacity building explicitly names three UAE-anchored artefacts: AUS course, two workshops at AUS and MBZUAI, one UAE-based PhD.
- Honest limitations are acknowledged in the full plan (no causal-ID publications; no foundation-model training experience; multivariate-tail dependence staged across years).
How to use this page
The body between ### EOI BODY (paste-ready) and ### END BODY is the verbatim text to paste into Google Form field 10. The title above is verbatim for field 9.
Before submitting, run the dhdna-profiler skill on the EOI BODY block per the CLAUDE.md mandatory rule for substantial generated text; see drafting-notes/dhdna-pass1.md for the first pass.