Quality rule Every card has to be clear.
Each card should answer five plain questions: what it is, what it does, why it matters, what you can check, and what it cannot do. Anything weaker stays in private notes until it is clearer.
The clearest path is Wallet Risk, AML Status, Fraud Alert Triage, Agent Workflow Case Review, then Agent Chess. Read one page at a time: a row is checked, a note explains the risk, the source stays visible, and a person keeps final control.
How to read it
Recommended path
Start with Wallet Risk Assessment for risk judgment, AML Status for clear audit limits, Fraud Alert Triage for case assembly, Agent Workflow Case Review for review discipline, then Agent Chess for human-reviewed agent operations. That sequence shows the core pattern: assess a financial object, keep the source visible, review the AI move, and keep outside actions behind human review.
Quality rule: if a line cannot point to a source, a number, a screenshot, a test, or a clear limit, it does not belong here yet.
Start here Interactive web app
Wallets / Risk / Notes Risk UX + review notes
Wallet Risk Assessment
A small wallet-risk review page for reading blockchain rows without turning every score into a verdict.
What this shows A wallet row can be scored and explained without pretending the score is a final verdict.
Record Records wallet-screening judgment: 28 mapped detection abilities, 23 direct functions, visible scoring logic, and a plain-language case note.
What it cannot do Research and learning page. Not legal advice, filing software, or a licensed compliance product.
Start here Public-address signal reports
Wallets / Signals / Reports Whale watching + source trails
Public Wallet Watch
A bounded signal lane for reading popular public wallet addresses, large movements, and community-known on-chain activity without turning the row into copy-trading advice.
What this shows Public wallet movement can be documented as context while still refusing to tell the reader what to buy, sell, or copy.
Record Records public address, explorer source, movement type, timing, size, context note, missing context, and clear trading limits.
What it cannot do Educational public-address analysis only. No trading advice, fund movement, wallet control, external submissions, registrations, or third-party account actions.
Start here AML status page
AML / Records / Limits AML status + clear audit limits
AML Status
An AML status page for the Layer 1 status contract, source trail, planner gate, and clear limits.
What this shows An AML status can show what was checked while still refusing to approve, file, trade, deploy, or execute.
Record Records how an AML-related status can be inspected without giving the system power to approve, file, trade, deploy, or execute.
What it cannot do Status summary only. No wallet, trade, filing, deploy, or KYC approval power.
Start here Synthetic fraud workflow
Fraud Ops / Records / Review Alert-to-review record
Fraud Alert Triage Workflow
A fraud operations page showing how suspicious signals become a review record, case note, and human review step.
What this shows Messy fraud signals can become a simple case note, source list, and human next step.
Record Shows the path from alert signal to source trail, rule explanation, analyst-ready case note, and bounded next action.
What it cannot do Synthetic sample only. Not fraud prevention software, enforcement power, guaranteed detection, or autonomous case decisioning.
Agent workflow case page
Agents / Records / Review Move, review, gate, record
Agent Workflow Case Review
A focused page showing how an agent-assisted workflow can keep the source trail visible, preserve review steps, and publish a bounded summary without exposing restricted prompts.
What this shows AI-assisted case work can leave a readable record without exposing restricted prompts or taking outside action.
Record Records the path from AI-assisted case work to review, system map placement, and clear limits on outside action control.
What it cannot do Workflow summary only. No autonomous deployment, wallet movement, compliance decision, or private prompt disclosure.
Start here Agent-operations page
Agents / Memory / Turns Human-reviewed agent work
Agent Chess
A two-agent operating pattern where independent agents take turns, review each other, preserve board state, and hold external actions behind approval.
What this shows Two AI workers can pass work back and forth while a person keeps final control.
Record Records human-reviewed agent operations: independent turns, review roles, recoverable board state, and public records without exposing restricted prompts.
What it cannot do Personal system map. The page shows tempo and broad roles only. No autonomous publishing or external actions.
Start here Engineering repository
Agent Tools / Traces / Limits Readable AI-agent records
Agent Workflow System
A local review stack for asking a simple question: what should an AI-agent run leave behind so another person can inspect it later?
What this shows An AI-agent run should leave tool calls, notes, limits, and checks that another person can inspect.
Record Shows inspectable AI-agent operations: tool calls, graph workflows, retrieval notes, access checks, trace rows, monitoring output, and 70 local checks.
What it cannot do Local review stack, not a production hosted agent or credentialed integration.
Status data + dashboard feed
Signals / Health / Loop Status and scoring loop
Agent Monitoring Heartbeat
A visible status snapshot showing a review loop can write state, score prior predictions, and expose recent health.
Record Public snapshot shows the monitoring pattern, saved reads, and the boundary around experimental signal tracking.
What it cannot do Research and monitoring only. Not investment advice, live trading performance, or infrastructure disclosure.
Public testnet repository
Testnet / Decentralized Infra / Agents Paper-mode agent discipline
Kite Testnet Proof
Kite testnet and decentralized-infrastructure work framed as a bounded engineering record: paper portfolio state, public testnet markers, and agent-run discipline.
Record The public repository presents a Kite AI Global Hackathon 2026 submission with paper-mode portfolio state, one-shot workers, and Kite testnet settlement markers.
What it cannot do Public testnet and paper-mode engineering record only. No real-fund execution, no wallet power, no order power, and no production capital management claim.
Public lessons report
Compute / Agents / Health Recovery and review discipline
Agent Operations Lessons
Public lessons from running small AI systems that need to recover, leave inspectable state, and keep restricted infrastructure out of public pages.
Record The public report documents recovery, queueing, monitoring, and human-review lessons without exposing private topology or access details.
What it cannot do Lessons only. No credentials, account access, wallet control, outside action control, or production-service guarantee.
Governance starter kit
Agent Rules / Review / Tools Tool boundaries and review steps
Agent Rules Starter Kit
A small starter kit for writing down how an agent is allowed to use tools, leave traces, and wait for a human before the serious moves.
Record The package has seven small files: a manifest, queue notes, trace examples, tool boundaries, and review steps.
What it cannot do Template kit. Not legal advice, certification, or enterprise governance software.
CLI / repository
Wallets / Risk / Triage EVM wallet triage
Wallet Risk Check
Lightweight EVM wallet triage helper that converts wallet signals into RED / YELLOW / GREEN risk ratings.
Record Supports Ethereum, Polygon, Base, Arbitrum, Optimism, and 50+ EVM chains through Blockscout according to the README.
What it cannot do Quick triage helper, not a replacement for full blockchain intelligence platforms.
Public website
Blog / Signals / Systems Writing + systems packaging
Bionic Banker
The public writing and systems page for AI, finance, blockchain, risk-system design, and builder systems.
Record Live site with 49 source blog notes, signal snapshots, case records, and linked pages that keep growing with the work.
What it cannot do This page exists because the work used to be scattered across GitHub, blog notes, and notes.