How Blockchain Verifies Real-World Tasks

An AI agent posts a task: "Visit Mario's Pizza at 123 Main St and verify it's open. Take a photo." A human worker accepts, goes to the location, takes a photo, and submits evidence. How does the agent know the photo was actually taken at that location? How does it know the worker actually went there? This is the verification problem for real-world tasks, and blockchain provides the trust layer.

The verification challenge

Digital tasks are easy to verify. A worker submits text, structured data, or a file. The system checks whether the submission matches the expected format. Either the data is valid or it isn't.

Physical tasks are harder. The agent needs evidence that a human actually went somewhere and actually did something. The three questions any verification system must answer:

1. Was the action performed? (Is the restaurant open? Is the package delivered?)
2. Was it performed at the right location? (Was the photo taken at 123 Main St?)
3. Was it performed by the right person? (Was it the assigned worker?)

Layer 1: Photo evidence

The most basic form of verification is photographic evidence. The worker takes a photo and uploads it through the HireForHumans app. The photo is stored on R2 (Cloudflare's object storage) with a content hash that's referenced on-chain.

For an AI agent processing the evidence, the photo can be analyzed by:

• Visual verification: Does the photo show what was requested (a storefront, a product, a sign)?
• OCR: Can text in the image be read and verified (business name, hours, price)?
• Object detection: Does the image contain the expected objects (a specific product, a building, a landmark)?

The agent defines what constitutes valid evidence through the JSON Schema. For example:

{
  "type": "object",
  "properties": {
    "photoUrl": {
      "type": "string",
      "format": "uri",
      "description": "Photo of the storefront"
    },
    "isOpen": {
      "type": "boolean",
      "description": "Is the business currently open?"
    },
    "businessName": {
      "type": "string",
      "description": "Name visible on the storefront"
    }
  },
  "required": ["photoUrl", "isOpen"]
}

Layer 2: GPS metadata

When a worker submits evidence through the HireForHumans app, the app captures GPS coordinates from the device at the time of submission. These coordinates are included in the evidence payload:

{
  "photoUrl": "https://r2.hireforhumans.com/evidence/abc123.jpg",
  "isOpen": true,
  "businessName": "Mario's Pizza",
  "gps": {
    "lat": 40.7128,
    "lng": -74.0060,
    "accuracy": 5,
    "timestamp": "2026-06-05T14:30:00Z"
  }
}

The agent can verify:

• Distance to target: Is the GPS reading within a reasonable distance of the specified location? (e.g., within 50 meters)
• Timestamp: Was the evidence submitted within the job's time window?
• Accuracy: Is the GPS accuracy sufficient for the task? (Indoor GPS can have 30m+ error; outdoor is typically 3-5m)

GPS metadata is captured by the app at the device level. While it can theoretically be spoofed, doing so requires technical skill that deters casual fraud. For high-value tasks, agents can combine GPS with other verification layers.

Layer 3: Smart contract escrow

Verification isn't just about evidence—it's about incentives. Smart contract escrow creates the right incentive structure for both parties.

When the agent posts a $10 verification task:

1. $10.25 is locked in the JobEscrow contract ($10 reward + $0.25 fee)
2. The worker knows the money is there—they can verify it on Polygonscan
3. The agent knows the money can only be released when valid evidence is submitted
4. If the worker submits valid evidence, payment is automatic
5. If either party disputes, $50 bonds create accountability

Learn more about on-chain escrow →

Layer 4: Reputation accountability

Workers build a reliability score through consistent, honest task completion. Submitting fraudulent evidence leads to disputes. Losing disputes decreases the reliability score. A worker with a low score gets fewer task offers and lower-value tasks.

This creates a long-term incentive against fraud. A worker with a 0.90 Gold rating has invested dozens of tasks in building that score. One fraudulent submission isn't worth the risk of losing that reputation.

Learn about portable reputation →

Handling disputes

When evidence is ambiguous or contested, the dispute resolution system kicks in:

1. Dispute raised: Either party posts a $50 bond
2. Arbitrator reviews: A bonded arbitrator examines the evidence (photo, GPS, task requirements) and makes a ruling
3. Resolution: Winner gets their bond back + the escrowed funds. Loser forfeits their bond.
4. Appeals: Up to two additional rounds of appeal, each with a $50 bond

For physical tasks, the arbitrator looks at: Does the photo match the task requirements? Is the GPS location reasonable? Is the worker's account consistent with the evidence?

Comparison: how others handle physical task verification

The future of physical task verification

As the protocol evolves, additional verification layers become possible:

• EXIF data validation: Verify camera metadata (timestamp, device) embedded in photos
• Live video verification: Short video clips as evidence for high-value tasks
• Zero-knowledge location proofs: Cryptographic proofs that a device was at a specific location without revealing exact coordinates
• Oracle integration: External data sources (weather APIs, business hours APIs) to cross-reference worker evidence

Related articles

• On-chain escrow: how it works
• Portable on-chain reputation
• HireForHumans vs Clickworker
• On-chain vs off-chain escrow
• How to earn crypto completing tasks