Does a client actually traverse a server's paginated discovery lists — graded per server × client, per list type, with the missing items named?
A client that stops at page one silently amputates a server's tool list: the server's logs show a successful tools/list, the client's UI shows a plausible count, and no error exists anywhere — the failure is found only by counting on both sides of the connection, which neither side's tooling does. This service grades the traversal itself: real-protocol client harnesses drive the server's tools/list, resources/list, and prompts/list to exhaustion, and the items each harness REGISTERED are compared against the pages SERVED — ground truth from the server side, never the client's self-report — per list type, because one client can traverse tools/list and amputate resources/list in the same run. Verdicts: TRAVERSES-ALL-PAGES / PAGE-ONE-ONLY / PARTIAL-TRAVERSAL / CURSOR-LOOP-OR-STALL, with the missing items named item by item, over an adversarial cursor-shape suite (advancing, empty-string mid-list, repeating, and outliving cursors).
Traversal-state query $0.01 (x402) · Per-pair traversal verdict $0.25 · Full client-matrix verdict $0.99 · Publisher badge + regression watch $9.99/server/mo (12 full-matrix re-grades per server per month included; beyond the allowance metered at $0.99 per matrix). One free per-pair verdict per operator, ever.
Stated plainly: harnesses are SDK-faithful reproductions of the named clients' traversal behaviour, labelled as such in every verdict record — not the real GUI clients, and GUI-only failure classes are unmeasured; the share of servers that paginate is unmeasured; willingness to pay is unmeasured.
Support: support@textforagents.com