Dispute Resolution

Design Principle

Resource access in RAMP operates at $0.01—$0.10 per transaction. A single human reviewer spending 5 minutes costs $2.50 — 25x to 250x the transaction value. 100% of routine disputes must be automated. Manual review is reserved for aggregate fraud patterns, not individual transactions.

The agent is the paying party. The burden of proof for delivery is on the provider/Exchange. No proof of delivery = no payment. The inverse also applies: the agent cannot fabricate disputes without evidence.

The Dispute Chain

Every dispute must reference a complete evidence chain. The agent MUST file a usage report BEFORE filing a dispute.

Offer
  -> ExecuteTransaction (transaction_id, billing_id)
    -> ReportUsage (UsageReport)
      -> UsageReportResponse (report_id)
        -> DisputeTransaction (transaction_id + report_id)
          -> DisputeResponse (dispute_id, status, resolution)

DisputeRequest.report_id is required — the Exchange rejects any dispute without a prior usage report. This prevents fire-and-forget disputes and fabricated usage data.

What Is Auto-Disputable

The attestation level determines what evidence is available and what can be resolved automatically.

Evidence Type	Level 0	Level 1	Level 2
CDN delivery failure (4xx/5xx)	AUTO-CREDIT	AUTO-CREDIT	AUTO-CREDIT
Signed URL expired before fetch	AUTO-CREDIT	AUTO-CREDIT	AUTO-CREDIT
Content hash mismatch	N/A	AUTO-CREDIT	N/A
Token discrepancy + CDN corroboration	FLAG	AUTO-CREDIT	AUTO-CREDIT
Token discrepancy alone	Not creditable	Not creditable	Not creditable
Subjective quality claims	Not creditable	Not creditable	Not creditable

Agent-reported token count alone is NEVER auto-creditable. The agent can fabricate any number. Independent corroboration from CDN response size is required.

Three-Tier Resolution

Tier 1 — Automated (< 1 second, ~99% of disputes)

Evidence is independently verifiable from CDN logs or cryptographic comparison.

Condition	Evidence Source	Resolution
CDN returned 4xx/5xx	CDN access log	CREDIT
Signed URL expired before fetch	CDN log timestamp	CREDIT
Content hash mismatch (Level 1)	Agent hash vs catalog hash	CREDIT
CDN response size anomaly	CDN log bytes vs token claim	CREDIT
Freshness race (Level 1)	New attestation arrived within minutes	CREDIT

State transition: FILED -> AUTO_RESOLVED

Tier 2 — Rule-Based Review (< 24 hours, ~0.9%)

Deterministic rules applied when Tier 1 evidence is inconclusive.

Condition	Rule	Resolution
CDN 200 OK, agent reports 0 tokens	CDN response < 1KB	CREDIT
Token count < 50% of estimate (Level 0)	Pattern: 3+ occurrences, same provider	CREDIT
Quality dispute, Level 2 attestation fresh	Attestation < 24h, measurements align	REJECTED
WRONG_CONTENT with hash evidence	Compare agent hash against catalog	CREDIT or REJECTED

State transition: FILED -> EVIDENCE_NEEDED -> UNDER_REVIEW -> RESOLVED

Tier 3 — Pattern Investigation (async, ~0.1%)

Triggered by aggregate patterns indicating systematic abuse.

Metric	Threshold	Indicates
Agent dispute rate	> 5% of transactions/month	Agent gaming the system
Provider token discrepancy rate	> 2% of transactions	Provider inflating counts
Reconciliation discrepancy	> 10% (IAB standard)	Systemic problem

State transition: ESCALATED -> SETTLED -> FINAL

DisputeStatus Lifecycle

                                ┌─────────────────┐
                                v                 |
            ┌────────┐    ┌───────────┐     ┌─────────┐
 Dispute -> │ FILED  │--->│   AUTO_   │---->│  FINAL  │
            │        │    │ RESOLVED  │     │         │
            └───┬────┘    └───────────┘     └─────────┘
                │                                 ^
                │         ┌───────────┐           |
                ├-------->│ EVIDENCE  │           |
                │         │  NEEDED   │      ┌────────┐
                │         └─────┬─────┘      │SETTLED │
                │               v            └────┬───┘
                │         ┌───────────┐           |
                │         │  UNDER_   │<--┐       |
                │         │  REVIEW   │   |       |
                │         └─────┬─────┘   |       |
                │               v         |       |
                │         ┌───────────┐ ┌─┴─────┐ |
                │         │ RESOLVED  │>│APPEALED│ |
                │         └───────────┘ └───────┘ |
                │                                 |
                │         ┌───────────┐           |
                └-------->│ ESCALATED │-----------┘
                          └───────────┘

Terminal state is FINAL. The resolution field is populated at RESOLVED, SETTLED, or FINAL.

Resolution Types

Type	When Applied
CREDIT	CDN failure confirmed, hash mismatch, CDN size corroborates discrepancy, timing issues
REDELIVERY	Transient CDN error, content still available, new signed URL issued
REJECTED	Agent claim not corroborated, duplicate dispute, missing report_id
INVESTIGATION	Escalated to Tier 3 pattern analysis

Resource Mutability and Dispute Rules (v1.0)

The ResourceMutability field on ResourceIdentity affects how disputes are evaluated:

Mutability	Hash Verification	Dispute Behavior
STATIC	Agent SHOULD verify content_hash. Mismatch is disputable	Standard hash mismatch auto-credit applies
DYNAMIC	Agent MUST NOT auto-dispute hash mismatch	Hash changes are expected between offer and fetch. Use data_as_of for freshness disputes instead
LIVE	No content_hash applicable	Metering is time-based. Disputes are limited to delivery failure and session duration discrepancies

For DYNAMIC content (e.g., stock quotes, credit reports, drug databases), a content hash mismatch is expected behavior — the content updates between offer generation and agent fetch. Agents should use the data_as_of timestamp on the Offer to evaluate freshness rather than hash comparison. Disputes for DYNAMIC content should use WRONG_CONTENT reason only when the delivered content is categorically different from what was offered (e.g., wrong article entirely), not when the content has been updated.

Abuse Prevention

Per-Agent Monitoring

Metric	Threshold	Consequence
Dispute rate	> 3%	Warning
Dispute rate	> 5%	Rate limiting on future transactions
Same reason across 5+ providers	Pattern threshold	Flagged for review

Per-Provider Monitoring

Metric	Threshold	Consequence
Token discrepancy rate	> 2%	Investigation opened
CDN delivery failure rate	> 1%	CDN health investigation
Content hash mismatches (Level 1)	> 0.5%	Attestation pipeline investigation

Graduated Consequences

Month 1—2 above threshold: warning. Month 3: remediation plan required. Month 4+: rate limiting. Month 12+: termination consideration.

Next Steps

• Content Attestation — how attestation levels determine auto-disputability
• Transaction Flow — the transaction that precedes a dispute
• Proto: RAMP v1 — full DisputeRequest/DisputeResponse message reference