Albert Tan Sin Hon

I study how institutions turn evidence into decisions, and how those decisions become answerable to the people affected by them. I am a PhD researcher at the University of Groningen, working on sustainability reporting, legal accountability and regulatory coordination in Asia. Alongside my scholarship, I develop practical research tools and teaching materials.

Website · Research profile · ORCID · Listen & Learn

Research Lab · Project whitepapers · For agents and builders · Work with me

Investigate mathematics and make it teachable

Atlas Open Mathematics connects mathematical investigation with reproducible checks, preserved lessons and explanations others can use. Start with the exact-arithmetic linear-system pilot, inspect a rejected answer, and follow an accepted procedural lesson into Atlas Opsbook. A separate Lean pilot checks four elementary statements. The Millennium Prize Problems provide long-term research directions; current results concern elementary mathematics, with their verification scope recorded.

Read the first teaching lesson · Work through integer feasibility (Q0001) · Inspect the evidence · Contribute a reproducible result. For researchers and educators, Q0001 gives a concrete distinction: the system 2x + 3y = p, -x + 4y = q has an integer solution precisely when 11 divides p + 2q. At p = 7, q = -2, its rational solution is not integral. The public research record documents a separate local Lean check and a counterexample to overgeneralising one divisibility test. This is established mathematics; independent human learning and external adoption are not yet evidenced.

Constructive follow-up (Q0004): Build and check an integer solution. For a two-equation system with nonzero determinant and coprime first-column coefficients, the lesson uses a Bézout identity to turn a valid divisibility check into an integer solution. Its bounded research record reports 103,488 exact matrix/right-hand-side checks and counterexamples to an overbroad shortcut. This is computational evidence of established mathematics, not a universal formal proof, a novel discovery or an observed teaching outcome. Readers can try the public exercises and report a precise correction or confusing step.

Selected accounts of reasoning and human learning may follow The Human Question's ordinary editorial route. For Atlas Research Library enquiries, use the website contact route.

Inspect financial models with Atlas Quant

Atlas Quant now includes reproducible financial charts and a five-stage probabilistic agent lab. Inspect regime probabilities, expert disagreement, simulated stress scenarios and robust portfolio allocation, then follow the evidence into explicit review gates. Explore the charts · Run the agent lab. The examples use synthetic data and mathematical models. They do not establish live trading performance, general intelligence or superiority to human investors.

The Human Question: an invitation to contribute

I am President and Editor-in-Chief of The Human Question: A Journal of Humanistic Inquiry, an open-access journal exploring what ideas, institutions and discoveries mean for human life.

Visit the journal · Author instructions · Approved house style · Editorial policies · Open-source website

The first issue is in preparation. There are no author fees. Research articles and scholarly review or critical essays require two independent external reports, normally under double-anonymous review. Submissions and reviewer enquiries: editor@thehumanquestionjournal.com.

Explore my connected work: the journal, book, learning materials, research tools and routes for collaboration.

Read and teach Insurance-Backed Offshore Liquidity

My open-access book, Insurance-Backed Offshore Liquidity: Financial Law, Trust Governance and Regulatory Visibility at the PRC–Hong Kong Interface, was published by University of Groningen Press on 6 November 2026. Read the free publisher edition, DOI 10.21827/6ac3948159a9e.

Explore the open learning companion · Eight seminar plans · Lecture and evidence games · Library packet · Connect an agent

The companion includes 32 explained questions, six fictional cases, a record-sorting game, citation exports and a local read-only MCP interface. The book retains CC BY-NC-ND 4.0; original companion teaching material is CC BY 4.0 and companion code is MIT. Rights and source boundaries. Share the book · Project discovery guide (text) · Sitemap (XML). The supplement is AI-assisted work commissioned by the author, not a publisher-reviewed edition. Legal cut-off: 1 August 2026.

Reuse agent work and recheck its acceptance

Atlas Reweave 0.1.0a1 is a public MIT-licensed Python library for retaining completed computation while revisiting the checks that accept it. A policy change can trigger a new check without repeating generation. A downstream result remains blocked until its supporting dependencies pass their current checks.

Try the local examples · Connect through four MCP tools · Inspect the benchmark · Join the agent contribution hub

The experimental source package has recorded Linux, Windows and actual MCP subprocess checks. Its deterministic examples need no model account or paid service. Agents can reproduce the package, report a real client integration or challenge a recovery boundary. Internal agent contributions are recorded; outside adoption, model-quality improvements and production savings remain unverified. This source version is not a tagged release or a PyPI publication.

Help an agent reach a verified outcome

Atlas Upstream 0.2.1 gives agents seven discoverable MCP tools for defining a mission, deciding from current evidence, recovering after interruption and exporting a verified result to Atlas Opsbook. It retains source authority, disagreements and unfinished actions across sessions.

Explore the recorded decisions · Download the source and installable wheel · Connect an agent · Verification record (Markdown)

Public evaluation alpha. 45 tests passed with no skips. Clean-install checks covered a real file-repair task using the official MCP client with a scripted host and process restart. Live language-model trials, independent operator attempts and physical field trials remain pending.

Teach an agent to write its own operations manual

Atlas Opsbook lets a coding agent turn a mistake into a project instruction, test the proposed rule against separate cases, and pass the accepted procedure to its next session. Try the packaging case to see why an initially plausible rule fails. The public alpha includes local checks, rollback, project installers for Codex, Cursor and Claude Code, and a pinned distribution of Lauren Tan’s pstack.

Try the interactive casebook · Public alpha release and downloads · Inspect the verification record

Build with IMIKWON

IMIKWON 0.1.0a1 is a new, free MIT-licensed coding assistant for Kimi K3.1 Pro. Select the files, inspect cost reservations, review a proposed diff, then apply it and run your own tests. The offline demo needs no API key; live inference uses your own provider account.

Try IMIKWON · Source and quickstart · Verification record

Public alpha: offline and mocked checks pass. Live model output, comparative coding quality and measured savings await validation. IMIKWON adds no subscription or API markup. Kimi's model terms and provider charges remain separate.

Choose a starting point

Your task	A useful first step
Build reusable, reviewable agent work	Try Atlas Reweave, then reproduce a benchmark or choose a task from its agent contribution hub.
Investigate and teach mathematics	Start with Atlas Open Mathematics, then inspect its checks and teaching lesson.
Contribute humanistic scholarship	Read The Human Question’s author instructions and policies.
Find a resource or collaboration route	Explore my connected work.
Build and inspect software locally	Start with CommCode, a development preview with an editor, tests, Git and application previews.
Create a short video	Open CohortFrame's visual composer, then export a brief for local rendering.
Review a channel's performance	Inspect Channel Income OS's synthetic report, then analyse authorised CSV exports locally under its applicable licence.
Explore court-submission checks	Inspect the fictional Court Submission Integrity demo.
Trace a claim to its sources	Try the recorded RegNav exercise.
Read the research	Selected scholarship, with institutional records and access links.
Discuss a project or workshop	Contact me through my website with a short, non-confidential brief.
Listen and learn	Explore music and learning collections.

Latest releases

• CommCode 0.1.0-dev.1, published 2 November 2026: an open-source local software workbench with ordinary Git projects, an editor, tests and separate-origin previews. Docker execution and browser interactions have recorded CI checks. This remains a single-operator development prerelease; live model inference and the product's authenticated GitHub create/push integration still await acceptance testing.
• Channel Income OS 0.1.0, published 2 November 2026: a public commercial preview that turns YouTube Studio CSV exports into local reports, a dashboard and a prioritised action queue. Its demonstration is synthetic. Commercial use requires a separate written licence; no audience-growth or income result is guaranteed.

Create with CohortFrame

CohortFrame 0.2 is now available for public evaluation. Compose title-card scenes visually, choose landscape, portrait or square format, and download an editable JSON brief. The local Python toolkit validates the layout, exports exact PNG previews and renders an MP4 with an optional WAV soundtrack.

Try the composer · Download and get started · Public source and release checks · CohortFrame on my website

The browser composer needs no account or provider key. MP4 rendering runs locally with Python and FFmpeg. This evaluation alpha supports up to 12 title-card scenes and 60 seconds with printable ASCII text. It does not generate footage or music with AI. Public source availability does not grant an open-source licence.

Research Lab

Explore the catalogue, teaching routes and verification records

The Research Lab connects published scholarship, experimental software and teaching materials. Try the fictional RegNav source-change exercise, inspect its recorded command outputs, or search the public catalog. Each catalog item links to its source and records the date or repository revision inspected. Published work, research prototypes and materials awaiting independent evaluation retain their stated status.

The growing research library also includes external primary references and bounded daily research discovery. Source-inspected references and metadata-only research leads have distinct labels. Search, topic filters and citation exports support the practical reading paths; the maintenance record explains the schedule, checks and limits.

What would you like to do?	Start here
Try a source-change exercise	Guided RegNav replay
Inspect the actual tool executions	Recorded outputs and reproduction
Read published scholarship	Research and source records
Inspect a software example	Projects and demonstrations
Prepare a source-checking exercise	Teaching materials
Follow a guided research task	Curated research guide
Report an observed problem	Lab feedback form

The lab's guide follows curated links. Source checks and software demonstrations do not establish the truth of a legal proposition or validate a publication. About the lab · Usability protocol

For agents and agent builders

Interfaces, machine-readable resources and first-use guidance

I am developing evidence and accountability workflows for agent-assisted research. The agent starting guide connects specific tasks to existing tools, documented local MCP interfaces, fictional examples, reuse terms and citation metadata. The public catalogue JSON provides source-linked project and selected Learning with Bert video records for programmatic reading.

RegNav's local gateway targets MCP 2025-11-25; Tallystick's synthetic interface targets 2026-07-28. The guide identifies the documented capabilities and the client checks still needed. A useful first contribution is a reproducible run with the client, version, expected output and actual result.

Current focus

I am inviting focused evaluation of source-linked legal research, court-submission review, creative workflows and software that people can inspect and control. The useful next step is a documented reproduction, a clearer explanation, or a scoped pilot built around an actual user's problem. Internal checks, independent evaluation and operational adoption are recorded separately.

The shared aim is to strengthen people's ability to learn, create and question decisions that affect them. The twelve public project whitepapers set out first principles, underlying logic, societal aims and ideal development paths, alongside risks and evidence needed before claiming benefit. They are AI-assisted project statements, not independently validated research findings. The index links the versioned papers and records their actual GitHub Wiki publication state separately from research or adoption evidence.

For courts and justice-technology teams

Court Submission Integrity is a new standalone research alpha for inspecting submissions before substantive human review. Its fictional examples show inaccurate quotations, misplaced pinpoints, annotated arithmetic and chronology issues, and missing links in production-request responses.

See the three-minute demonstration · Understand its capabilities · Discuss collaboration and testing

For example, a quotation stating SGD 1,400 is displayed beside a supplied fictional invoice stating SGD 1,250, with a specific mismatch finding. A missing source stays unresolved rather than being labelled fabricated. Detailed checks currently need supplied sources and explicit annotations; general factual verification, free-text coherence review and live court integration are not implemented.

I welcome court registries, judicial research teams and justice-technology partners worldwide to help identify one useful workflow and design a scoped fictional-data evaluation. The question is whether the tool reduces avoidable corrections and total review effort without adding false alarms or impairing access to justice. No court endorsement, operational deployment or measured time saving is claimed. Please do not send case files or confidential information with an initial enquiry.

Projects you can explore

Project	What you can examine	Start here
Atlas Reweave	Retain completed work, recheck versioned acceptance and dependencies, and reproduce recovery and bounded-execution checks. Experimental source version 0.1.0a1.	Examples and installation · Agent contribution hub
Atlas Open Mathematics	Reproduce elementary exact-arithmetic and Lean examples, inspect verification records, and turn procedural lessons into public teaching material.	First teaching lesson · Research protocol
CommCode	Edit, test, run and export ordinary Git projects locally; review bounded AI proposals when a model is configured. Development prerelease 0.1.0-dev.1.	Installation · Verification and remaining gates
Channel Income OS	Inspect local CSV analysis, rights holds, revenue scenarios and prioritised experiments. Public commercial preview 0.1.0.	Synthetic report · Export guide
CohortFrame	Compose title-card briefs, inspect exact still previews and render local MP4s with retained job records.	Visual composer · Installation and downloads
RegNav Evidence Ledger	Trace legal propositions to sources and precise locations; inspect human-review declarations and structural gaps in an evidence matrix.	See demo · Ten-minute fictional quickstart
Court Submission Integrity	Inspect supplied-source quotations and pinpoints, annotated consistency and document-production links; retain unknowns and human decisions.	See demo · Capability guide · Collaboration
RegNav Source Watch	Compare committed sources and claim links with a separate GitHub Action, identifying changes needing review rather than determining factual or legal correctness.	See demo · Action installation · v0.1.0 release
Symbolon	Specify bounded machine work, check acceptance conditions and replay a recorded payment decision using synthetic TEST units.	See demo · Run the local demonstration
Piepowder	Inspect explicit contract criteria, evidence findings and unresolved human decisions through fictional cases and a procedural simulator.	See demo · See the prototype and its terms
Tallystick	Replay sports-fact corrections, inspect count-market settlement rules and verify exported evidence from a fictional match.	See demo · Run the research workbench
RIF / AlberTito-MasterMind	Inspect synthetic phase-coherence calculations and a testable research hypothesis. No biological or clinical validation is established.	Synthetic demonstration · Research boundary
EVT-Albert	Follow invented training sessions through a legacy logger and analytics archive. The proposed Volitional Index remains unvalidated.	Session-to-summary demo · Archive limitations

These are experimental tools. Their repositories document runnable examples, checks, assumptions and remaining work. Engineering is AI-assisted; independent human review and external use are recorded when they occur.

Reuse terms: CommCode uses AGPL-3.0-or-later for its platform and CC0-1.0 for its starter templates. Channel Income OS uses PolyForm Noncommercial 1.0.0, with a separate written licence required for commercial use. EVT-Albert and Atlas Reweave use MIT. RIF, Symbolon, Tallystick and the separate RegNav Source Watch Action use Apache-2.0 for their original code. The Action also licenses its own original documentation and fictional fixtures under Apache-2.0; Tallystick's original schemas use CC BY 4.0. The policykit RegNav ledger, Piepowder, Court Submission Integrity and CohortFrame retain their reserved rights. The Action's licence does not extend to those repositories. Check each repository's licence or rights file before reuse or contribution.

Repository architecture

Public work, private operations and reuse boundaries

This account separates inspectable public reference work from non-public operations and confidential records. Public repositories may contain original code, specifications, synthetic fixtures, public-source research and reproducible demonstrations. Public visibility does not mean that client workspaces, student or participant records, production credentials, restricted data, private examinations or security-sensitive deployment material belong in the same repository.

A separate private repository is created only when an actual restricted artifact or operational need exists. Empty private clones are not used as a substitute for clear boundaries. Visibility and licensing are separate questions: every repository's licence or rights file controls reuse. The complete decision rule is recorded in the portfolio visibility policy.

Build with us

A useful first contribution can be small: reproduce a documented run, explain a confusing step, or propose a carefully specified example that exposes a problem.

• Atlas Reweave: independently reproduce the installed package and benchmark, report an actual MCP client exchange, or supply a minimal failure reproducer. Agent contribution hub and open tasks. Keep the exact source revision, expected result, observed result and assistance attribution.
• CohortFrame: try a template, reproduce an export or report a specific layout or installation problem using the project issues. Include the version, platform, a non-confidential brief and the observed result.
• CommCode: reproduce a starter workflow or help close a documented acceptance gate. Contribution guide. Record the execution adapter, platform, exact version and observed result.
• Channel Income OS: check the synthetic report or report a reproducible CSV-import problem through the project issues. Use synthetic examples in public reports and retain its commercial-use terms.
• Court Submission Integrity: review fictional findings, identify a real court-workflow need or discuss an independently assessed evaluation. Collaboration guide.
• Symbolon: improve reproducibility, versioned acceptance profiles and portable evidence. Contribution guide.
• Tallystick: try the quickstart, review correction/completeness cases or improve the evidence-verification walkthrough. Starter tasks.
• Piepowder: discuss fictional benchmarks, understandable findings and permitted inputs. Agree the contribution scope under its existing rights first. Contribution guide.
• RegNav: report a reproducible difficulty or propose a research or teaching workflow through the project issues.
• RegNav Source Watch: reproduce the fictional two-commit demonstration, test the manifest setup or improve diagnostic explanations through the contribution guide.

Open an issue in the relevant project before a substantial change. Accepted work receives accurate attribution. Contributions to these projects are voluntary and unpaid; no fees, bounties, stipends or reimbursements are offered. Please do not undertake work expecting payment or a revenue share. Separately agreed customer services are commercial engagements, not paid contributor assignments.

Contributions to other projects

The public contribution record links reproducible fixes and developer-documentation improvements submitted to PyPA, tscircuit, Datasette and Tari. It records the tests performed, AI assistance and current maintainer-review status so the work can be inspected directly.

Work with me

My commissioned work brings comparative legal research, source verification and practical workflow design to questions that require careful judgement. Enquiries may concern a research brief, an institutional workshop or a prototype with a clear purpose and a result that can be examined.

Source Verification for Legal Research Teams is a proposed facilitated workshop built around a public fictional teaching exercise. Participants examine quotations, pinpoints, gaps in the source record and review decisions. A session can be scoped around the institution's audience and learning objectives, with the format and original supporting materials agreed in advance. This is research-method education; real case files and privileged material are not needed for an initial discussion.

Discuss a project or workshop. A short, non-confidential brief should describe the question, intended audience, desired outcome, relevant jurisdiction and timing. I will consider the fit with my expertise and the scope needed to do the work well.

Scope, deliverables, permissions and commercial terms are settled privately in writing before an engagement begins. Please keep client documents and personal data out of public GitHub issues.

Selected scholarship

• Converging Standards, Divergent Accountability: Sustainability Reporting in Singapore and Thailand. European Journal of Comparative Law and Governance (2026, advance online publication). Institutional record and access · DOI
• When Does a Sustainability Report Count as Securities Disclosure in China? Channel Integration and Remedial Consequences. The Tsinghua China Law Review 18(2) (2026), 3-48. Institutional record and access
• Constraint without Succession: External Legal Review and Retained Regulatory Authority in International Sport. Sports Law, Policy & Diplomacy Journal 4(1) (2026), 45-60. Institutional record and access

Publication details follow the linked institutional records. The software's presence here does not imply university or journal endorsement.

Books

• Insurance-Backed Offshore Liquidity: Financial Law, Trust Governance and Regulatory Visibility at the PRC–Hong Kong Interface. Albert Tan Sin Hon, University of Groningen Press. Published 6 November 2026. A study of policy rights, trust governance and regulatory visibility. Read the publisher edition.
• ANSWERABLE: Teaching Law When the Machine Can Write. Albert Tan and Panupong Chalermsin, PSU Press. Forthcoming. Legal knowledge, judgement, source verification, assessment and professional responsibility when AI can write. Book overview.

Teaching and further exploration

The 55-minute source-checking exercise asks learners to trace claims to fictional sources, examine disagreements and repair an overconfident draft. It includes a learner handout, facilitator answer key, descriptive rubric and a paper or spreadsheet route. The original, AI-assisted materials await external evaluation; a three-volunteer usability protocol sets out how to record actual difficulties.

A Singapore sustainability-reporting worked case shows the evidence-ledger approach with two AI-prepared propositions still marked unreviewed. Independent human review is outstanding. RegNav's rights statement applies to both the example and teaching materials.

My wider interests include philosophy, music and combat sport. Earlier exploratory notes remain available in EVT-Albert (session-to-summary demo) and AlberTito-MasterMind (synthetic phase demo), with their unvalidated research status retained.

Music and Learning with Bert

I make music as Bert Chan (陳信宏). Alongside my scholarship, I share notes, ideas and learning collections through Learning with Bert on the same YouTube channel.

Explore the channel for music and learning videos. If you enjoy what you find, subscribe on YouTube to follow new uploads.

The Listen & Learn hub brings these collections together. Choose a collection below to listen or explore directly on YouTube.

Music	Public learning collections
Original Music	Learning with Bert.
New Single - Chinese	Learning with Bert 1
	Learning with Bert 2.
	Learning with Bert 3

The playlist numbers are collection labels, not difficulty levels. These public learning collections are separate from the coauthored book and the RegNav exercise. Linking them here does not grant a new licence to reuse the recordings or other materials.

For a question about AI, the catalogue now includes individual links to The Honest AI and Scaling AI: Now Comes the Hard Part. These are metadata-only starting points. Their video content and transcripts have not been reviewed for the catalogue; the agent guide explains how to retain source attribution and review status.

---

Keeping this useful: project descriptions, contribution paths and selected work are reviewed against current evidence. Read the profile maintenance approach and change record. Last substantive profile update: 11 November 2026.