What is Kredence?

Kredence is an autonomous agent system that discovers projects in a funding ecosystem, monitors them as they ship, adversarially challenges their impact claims, and produces continuously-updated hypercerts that funders can use to allocate with confidence.

The problem

Most impact evaluation tools ask projects to submit something. Forms, milestone reports, self-assessments. The output is optimistic — because the people filling them out want to look good.

Kredence does not ask projects to submit anything.

Point it at an ecosystem — a Gitcoin round, a DAO cohort, a hackathon track — and it finds every active project, pulls their evidence from GitHub, deployments, and onchain activity, runs an adversarial challenge on every claim, and builds a living hypercert for each project.

What Kredence produces

For each project discovered in an ecosystem, Kredence outputs:

Artifact	Description
Project manifest	Structured list of all discovered projects and their evidence sources
Evidence bundle	Raw and normalised evidence — commits, PRs, deployments, README claims
Adversarial log	Signed agent receipt of every claim challenged, with outcome: verified, flagged, or unresolved
Hypercert payload	Title, contributors, timeframe, impact category, evidence references, and adversarial summary

What makes it different

Against AI summarisers — Kredence runs an adversarial agent on every claim before it enters a hypercert. The output is a structured evidence record with verified claims, flagged overclaims, and open objections — not a confidence score from a single LLM pass.

Against manual evaluation — Kredence produces consistent, replayable, evidence-backed records across an entire ecosystem simultaneously.

Against self-reported systems — No submission form. Projects that ship consistently but never fill out forms are still discovered and evaluated.

The agent pipeline

Ecosystem Input
    │
    ▼
Scout Agent          Discovers all active projects autonomously
    │
    ▼
Evidence Agent       Collects GitHub, onchain, and website evidence
    │
    ▼
Adversarial Agent    Challenges every claim — flags, verifies, or marks unresolved
    │
    ▼
Synthesis Agent      Assembles the final hypercert payload
    │
    ▼
Hypercert + Dashboard

Each agent carries an ERC-8004 onchain identity. Every output includes identity metadata. The Adversarial Agent's objection logs are signed receipts.

Next steps

• How it works — a deeper look at each agent stage
• Quickstart — run your first evaluation in under 5 minutes