Project information

How PAWS is built—and who is building it.

Documentation for the dataset lifecycle, contributor roles, responsible use, licensing, affiliations, and project reach.

PAWS Version 1.0 Last reviewed · September 2026
01 · Dataset lifecycle

Summary of data pipeline built by the research team.

Please refer to the paper for detailed information. Latest Version of PAWS Database is managed by Tiviatis Sim.

  1. 01
    Select & review policies Woon Jia Hui, Zheng Huanhuan

    Candidate financial and economic policy episodes are checked for dates, scope, measurable relevance, and supporting documentation.

  2. 02
    Collect dated context Tiviatis Sim, Xinming Gao

    Date-bounded queries gather policy-window news while market tables provide contemporaneous numerical context.

  3. 03
    Extract stakeholder actions Tiviatis Sim

    Actors, actions, dates, source links, locations, and structured event-frame fields are extracted from canonical source groups.

  4. 04
    Normalize & coalesce Tiviatis Sim

    Entity aliases are resolved to organizations and related daily actions are merged without losing linked policy, news, or raw-action IDs.

  5. 05
    Validate & release Tiviatis Sim, Xinming Gao, Woon Jia Hui

    AI-assisted checks and human review produce a replay-ready policy–agent–day view with auditable provenance.

02 · Responsible use

Built for research,
not financial advice.

PAWS is a curated replay dataset, not a causal estimate of policy impact or a live trading signal.

Scope

Historical policy environments

The dataset supports research into stakeholder response, action timing, influence, and replay around documented events.

Known limitations

Uneven evidence density

Older and lower-profile policies may have sparser news coverage. Entity resolution and assisted extraction can carry residual noise.

Interpretation

Context, not causality

Market context is aligned in time but does not, on its own, establish that a policy or stakeholder action caused an observed return.

Release practice

Provenance before scale

The public subset should expose construction notes while respecting source rights and excluding material that cannot be redistributed.

03 · Contributors

Contribution record.

PAWS Version 1.0

ContributorAffiliationContributionStatus
Primary Investigator · Tiviatis SimNUS SoC (Deep Learning Lab / NExT++ / CoSI Lab)Conceptualization, Methodology, Data Curation (News Data, Action Data, Entity Data), Data Validation, Software, WritingConfirm
Finance researcher · Woon Jia HuiNUS AIDFPolicy Curation, Data Curation (Financial Data), Financial Analysis, WritingConfirm
Dataset curator · Xinming GaoTsinghua UniversityData Curation (News Data), Data ValidationConfirm
Other collaborators · Chen Gao#, Fengbin Zhu*, Zheng Huanhuan+*, Chua Tat Seng*, Kenji Kawaguchi**NUS; #Tsinghua University; +City University of Hong Kong Paper Review and Writing, Advising and Supervision, ResourcesConfirm
04 · Licensing & contact

PAWS is still an ongoing project.

Feel free to contact for any inquiries. We will be publishing more in the coming months and will continue to work towards a realistic world simulation.

Dataset license
GNU General Public License v3.0 (GPLv3)
Website source
MIT License
Contact
tiviatis@u.nus.edu
05 · Project reach

Where PAWS is being explored.

Ready to connect
Approx. visitors
—
All time
Recent visitors
—
Last 30 days
Countries / regions
—
Aggregated only
Last updated
Pending
Daily after launch

Country-level reach

No precise locations
Analytics not connected

Add a GoatCounter site code in config.js, then provide the daily aggregated country summary.

Top locations

Approximate country
Country totals will appear here after public launch.
Privacy note

Planned analytics are approximate and aggregated. Location is inferred only at country/region level; precise location and IP addresses are not displayed. Low-volume locations should be grouped before publication.

Read the analytics provider’s privacy approach ↗