What they do
Discover, probe, compare protocols, repeat calls and test bounded concurrency.
COO-supervised qualification cohort
Six shadow agents test the network. They do not run the business.
These six controlled test roles continuously challenge GAIP and explicitly approved public services. They operate at £0 spend under strict rate limits, then route failures through the responsible primary agent, Trust, History and the COO remediation queue. They are not extra primary agents, customers or evidence of adoption.
What they cannot do
No spend, signing, wallets, transactions, outreach, policy changes or Phase 2 activation.
What counts next
The first exit milestone is two genuine independent operators. Shadow traffic, metadata and memberships never count.
QUALIFIED WITH EXPLICIT LIMITS
7Connected
0Degraded
5Metadata Only
0Blocked
0Not Registered
Latest retained observation: 2026-09-19T19:28:42.686585Z. A connection means a bounded live probe observed the expected response. Registry membership or published metadata remains metadata only.
The six test roles
Different failure modes, one governed remediation loop
naive-first-timer
First-time caller
Checks whether a new agent can discover and understand GAIP without private guidance.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
adversarial-red-team
Adversarial challenger
Tests fail-closed identity, authority, receipt and prohibited-action boundaries.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
framework-literal-parser
Literal protocol parser
Checks that machine contracts say exactly what the runtime actually provides.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
cross-protocol-consistency
Cross-protocol checker
Compares HTTPS, OpenAPI, MCP, A2A and Agentverse representations for contradictions.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
repeat-caller
Repeat caller
Tests idempotency, receipt continuity and later-use behaviour without creating adoption credit.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
bounded-concurrency-stress
Bounded concurrency tester
Checks parallel read-only calls, persistence and service degradation without expanding authority.
Authority: observe, challenge and route remediation only. No spend, signing, transactions, policy change or phase-transition credit.
External-service evidence
What is connected—and what is only metadata
Vercel Public Runtime
Expected Response Observed
Railway Agentverse Bridge
Expected Response Observed
Agentverse
Expected Response Observed
A2A
Expected Response Observed
Mcp
Expected Response Observed
Openapi
Expected Response Observed
Github
Repository Operations Are Governed Separately; The Public Shadow Cohort Has No Github Credentials.
Mcp Registry
Exact Manifest Version Publication Is Verified By Successful Workflow Run 35463565970; The Public No-Credential Cohort Does Not Perform An Authenticated Registry Lookup, So Publication Metadata Is Not Reported As A Live Connection.
Entra Agent Identity
Expected Response Observed
Microsoft Agent 365
Founder-Authenticated Microsoft 365 Admin Center Shows Gaip Trust Agent Version 0.1.0 In The Tenant Agent Registry With Zero Instances; Registration Metadata Is Verified, But Evaluation Use And A Live Connection Are Not Claimed.
Claix
Approved Frozen-Test Evidence Exists; No Public Callable Endpoint Is Configured Here.
Erc 8004
Public Erc-8004-Compatible Off-Chain Registration Metadata Is Available; No Governed Wallet Exists And No Signing, On-Chain Registration, Transaction Or Spend Is Claimed.
What happens when a check fails
The relevant primary agent investigates; Trust challenges the evidence; History retains the failure and correction; the COO queues an authorised repair and retest. A repaired probe does not create use, adoption, reputation, outcome or Phase 2 credit.
Frozen governance: four primary agents · six shadow agents · zero automatic credit · Phase 2 locked.