1. Applicant & Project Overview
Project Name: AgentWitness
Team / Individual Name(s): Solo builder (GitHub: mrtarasyuks), built with AI assistance (Claude).
Links:
100-Word Summary: AgentWitness is a read-only trust explorer for the Intuition knowledge graph. Paste a wallet, an atom/triple id, or a name — see how much TRUST is staked on it, by whom, the FOR-vs-AGAINST split for a claim, the full signal history as a cumulative-stake chart, the top backers annotated with their on-chain activity, and a sybil heuristic that flags when backing looks like a cluster of fresh accounts. Everything is computed client-side straight from Intuition’s public GraphQL indexer — no wallet connect, no signing, no backend, no API keys. It turns “what score does this have” into “here is the evidence behind that score.”
Project Category: InfoFi / Knowledge Graph (overlaps with Identity/DID/Reputation, since it reads trust signals about accounts and agents).
Elevator pitch: Paste a wallet, an atom, a triple, or a name — and read the trust behind it before you trust it.
Origin story: Built to fill a gap noticed while using Intuition’s own portal: it shows a reputation number but not why — who backed it, when, and whether that backing looks organic.
Notable traction or achievements: None yet — the tool just shipped. No fake users or metrics will be claimed here or anywhere in the project.
Current users or early testers: None beyond the builder’s own testing against live mainnet data during development.
2. What You’re Building
Problem Statement: Intuition makes “who is trustworthy” into staked, on-chain data, but the portal only shows an aggregate number. Anyone staking on a claim, backing an agent’s reputation, or reviewing a project for a grant has no quick way to see who is behind a trust signal, when they staked, how much, and whether the backing looks like organic conviction or a cluster of fresh wallets.
Proposed Solution: AgentWitness reads the live Intuition graph and, for any atom / triple / account, shows: TRUST staked and by how many backers at what share price; for a claim, the FOR vs AGAINST split; the full signal history as a cumulative-stake timeline chart plus the latest deposits/redemptions; the top backers annotated with their own on-chain activity; a sybil heuristic flagging when a lot of backing comes from fresh, low-activity accounts; and for a wallet, its own reputation received and everything it backs. Every result links back to Blockscout for independent verification.
Stage of Development: Live — shipped, public repository, deployed on GitHub Pages, tested against real mainnet data (chain id 1155), with screenshots in the repo.
Technical architecture overview: Static HTML/CSS/JS, no framework or bundler, no backend. Chart.js (via CDN) renders the signal timeline. The browser calls Intuition’s public Hasura GraphQL indexer directly.
Integrations or dependencies: https://mainnet.intuition.sh/v1/graphql (Intuition’s public mainnet indexer — read-only, open CORS, no key) for all data; https://explorer.intuition.systems (Blockscout) for wallet/transaction links. No MCP/A2A/ERC-8004/x402 — deliberately dependency-light so it works with zero setup for any visitor.
Security considerations: Read-only by design — no wallet connect, no signing, no transactions, no API keys, no backend to leak or attack. The README states plainly that the sybil flag is a heuristic, not proof, and that the tool shows an error rather than invented numbers when the indexer doesn’t answer.
3. Team & Execution Ability
Team backgrounds: Solo builder with hands-on experience operating real wallets and reading on-chain data across multiple networks, working with an AI pair-programmer (Claude) for implementation.
Execution proof: The project itself is the proof — idea to a live, publicly deployed, tested tool on real mainnet data, with a public GitHub repository and an honest README, completed before this grant was even requested.
Commitment level: Part-time (the builder also works on other on-chain tooling), but this specific tool is complete, deployed, and will keep receiving incremental fixes regardless of grant outcome.
Team structure and roles: None — one person covers idea, data layer, UI, and QA end to end.
Advisors or collaborators: None.
Expected future hiring: None planned; the project is intentionally kept small and maintainable by one person.
Prior crypto / AI experience relevant to the project: Experience building and operating read-only on-chain tooling, and working daily with AI coding assistants to ship small, focused products.
4. Grant Request & Milestones
Amount requested: $2,500 USD (or TRUST equivalent at time of payout).
Budget breakdown by milestones:
- Milestone 0 — already delivered, $0 requested (proof of execution): Live read-only explorer — atom/triple/account profiles, signal timeline, top-backer list, sybil heuristic — shipped on real mainnet data, public repo + GitHub Pages site.
- Milestone 1 — $1,200: Replace the indexer’s hard page caps (250 signals / 100 positions) with cursor-based pagination so the busiest atoms show their full history, not just the first page; refine the sybil heuristic from a flat “fresh account share” into a stake-weighted, time-decayed score.
- Milestone 2 — $1,300: Side-by-side atom/triple comparison view; an embeddable “trust snapshot” badge/widget that other Intuition-ecosystem tools can link to or embed, so AgentWitness becomes a shared reference rather than a standalone page.
Expected Timeline: Milestone 1 within ~3 weeks of grant approval; Milestone 2 within a further ~3 weeks (Milestone 0 is already complete today).
Success criteria:
- M1: pagination verified against the atom with the most signals/positions on mainnet at the time; the refined sybil score visibly changes on a known clustered-backing example vs. the old heuristic.
- M2: comparison view live and shareable via URL; badge embeddable with a documented snippet; at least one other Intuition-ecosystem page links to or embeds an AgentWitness view.
Stretch goals: Support for non-base bonding curves; a thin read-only API wrapper so other projects can query a sybil score directly instead of scraping the page.
Key dependencies: Continued availability of mainnet.intuition.sh/v1/graphql with its current open-CORS, no-key access.
Risk assessment: Mainnet only launched in November 2025, so many atoms still have thin history — pagination and decay scoring matter more as activity grows; an indexer schema change would require query updates.
How this work compounds value for the network: Easier-to-read trust evidence should encourage more informed signal activity rather than blind staking, and a reusable badge lets other builders (e.g. AgentID, IntuRank) link to evidence instead of rebuilding it themselves.
5. Intuition Ecosystem Alignment
Why Intuition: The entire product exists only because Intuition exposes staked claims — atoms, triples, and signals — as a publicly queryable graph. No other chain publishes per-backer staking history on claims this way; AgentWitness is pure read-access to that primitive.
Which Intuition primitives are used: Atoms, triples (subject / predicate / object), signals (stake events), positions (who holds shares in which vault), and vaults (current_share_price, total_assets, total_shares, position_count on the base curve).
Why it must be built on Intuition: The dataset itself — who staked for or against a claim, when, and how much — is specific to Intuition’s MultiVault design; there is nothing to port to another chain.
Data structure plan: Purely read queries against the existing schema; the tool writes nothing to the graph, so it adds zero write-load while increasing legibility of data that already exists.
TRUST integration: None, by design — AgentWitness never holds, swaps, or requires TRUST; it stays strictly non-custodial and read-only.
New schemas, patterns, or agent types: None — it deliberately does not write to the graph.
How this increases network activity or knowledge density: Lowers the cost of inspecting a claim before backing it, which should encourage more informed signal activity rather than blind staking.
Long-term contribution (12–24 months): Becomes a default “show your evidence” link other Intuition tools point to — see the differentiation below.
How this differs from funded projects: Cohort 1 already funded AgentID (portable identity / credentials for autonomous agents) and IntuRank (a single aggregate reputation-intelligence score across the graph), plus IntuitionMCP (an MCP server so AI assistants can query the graph natively). AgentWitness is neither a registry nor a scorer: it is a read-only evidence layer for any atom, triple, or account — who staked, when, how much, for or against, the signal history over time, and a sybil heuristic on the backers. It is designed to sit underneath AgentID and IntuRank rather than compete with them: if AgentID issues a credential or IntuRank reports a score, AgentWitness is where someone checks what that credential or score is actually resting on.
6. Sustainability & Long-Term Vision
Long-term vision (6–24 months): Stay a free, static, zero-maintenance-cost public good that grows into the “verify the trust behind it” layer other agent-identity and reputation-scoring tools in the ecosystem (AgentID, IntuRank, and others) can link to or embed, instead of each rebuilding backer-level evidence views themselves.
Post-grant sustainability plan: The entire app is a static site on GitHub Pages, which is free to host indefinitely with no servers, databases, or API keys to maintain. Grant funds pay only for the build time in Milestones 1–2; there is no ongoing cost the project depends on to keep running afterward.
Business or revenue model: None. AgentWitness has no token, no ads, and no fees — it stays a free tool, consistent with keeping anything public honest and free of hidden incentives.
Go-to-market / distribution strategy: Share direct profile links with claim and atom owners who want to show their backing is legible; reach out to AgentID/IntuRank/IntuitionMCP builders about linking to or embedding AgentWitness views as a complementary evidence layer.
Competitive landscape: portal.intuition.systems shows a reputation number but not backer-level history. IntuRank reports a single aggregate score. Neither currently shows a time-series signal history with a sybil heuristic for an arbitrary atom, triple, or account — that gap is what AgentWitness fills.
7. Additional Materials
Screenshots: in the repository’s screenshots/ folder, captured against live mainnet data.
Contact email: mrtarasyuks@gmail.com
Wallet address: 0x2973B942305DF55B6fb1ed676a39e9741e7E1F62
8. Applicant Attestation
I confirm the information provided above is accurate to the best of my knowledge and that I intend to deliver the milestones described in Section 4.
Name: Serhii Tarasiuk
Wallet Address: 0x2973B942305DF55B6fb1ed676a39e9741e7E1F62
Date: 06.10.2026