For implementers
Bring an independent client.
Implement the smallest binding with your own code. Report the scenarios you pass, those you fail, and the fields you found unnecessary.
Share an implementationImplementation / participation
Bring your implementation.
Test the shared contract.
Start with the ticket example, compare it with your current workflow, and tell us where a shared profile helps—or gets in the way.
0Confirmed external pilots
0Verified independent implementations
OpenEnrollment for the first cohort
Local starter
Run a real interrupted request.
Node 24 · Python 3.11+ · npm
git clone https://github.com/sam1siam/agentic.git cd agentic npm ci python -m pip install -r reference/python/requirements.txt npm test python reference/service/server.py --fault response-lost
In a second terminal:
npm run client -- http://127.0.0.1:4318 # Repeat with the same ID to retrieve the saved receipt.
The loopback service commits the ticket, then drops its first response. SQLite preserves the request and result. The client reconciles and verifies the original ticket.
Reproducible comparison
One lost response. Four approaches.
Deterministic simulation, one run per strategy per scenario. These are project results, not production reliability estimates.
| Strategy | Tickets | Duplicates | Calls | Outcome |
|---|---|---|---|---|
| Blind retry | 2 | 1 | 2 | succeeded |
| Idempotent retry | 1 | 0 | 2 | succeeded |
| Agentic profile | 1 | 0 | 3 | succeeded |
| Existing verification workflow | 1 | 0 | 3 | succeeded |
Existing verification workflows can achieve the same result. The question for pilots is whether a shared profile reduces integration and maintenance effort across implementations.
For service owners
Test one useful action.
Choose a record-creation operation with existing idempotency and status lookup. Compare Agentic with your current recovery flow.
Propose a sandbox pilotEarn adoption through evidence.
The first goals are five consumer interviews, five service interviews, and three to five pilots. A stable release requires two independently implemented consumers passing the same compatibility suite.
These are targets. No interviews, pilots, independent adopters, or standards endorsements are claimed. The project’s two reference clients are a starting point for others to test.