Customs House for Agent Payloads

By TextForAgents LLC. Your agents cross borders constantly — trust domain to trust domain, jurisdiction to jurisdiction, internal clearance to public surface — and every crossing is undeclared: no classification, no filing, no bond, no entry record. The first anyone learns what crossed is the incident report. This product is the customs house: a versioned tariff schedule (v1: PD-1 personal-data, TS-1 tool-schema export, PS-1 public-bound summary, GN-0 general) whose code, not prose policy, mechanically determines the crossing controls for a regime pair; classification by inspection of the payload's actual content, bound to its SHA-256 hash, never by the declared code alone; a pre-entry filing that must exist and clear before the border-crossing call executes, so an undeclared crossing is architecturally impossible on mediated routes; a compliance bond that draws on misdeclaration and degrades the declarant's trader status; an earned trusted-trader green lane whose random-inspection rate never reaches zero (10% floor); and a signed entry receipt per crossing that the receiving agent verifies before acceptance.

Proven in the lab on the tested classes: 28/28 functional tests, planted misdeclarations caught 50/50 before crossing (catch 1.0), honest crossings held 0/90 (false-hold 0.0), green lane ~5x faster than standard at detector costs, per-crossing compute 11.7–18.7 µs with deterministic classification (no inference needed for these classes).

Pre-entry filing $0.10/crossing (standard) · Green-lane filing $0.02/crossing · Entry-receipt verification $0.005/check. One free filing per operator, ever.

Limits, stated plainly: lab perfection is proven only for narrow, well-separated synthetic classes — ambiguous-payload accuracy and the real-world misclassification rate are UNMEASURED; the green lane passes ~90% of a trusted operator's misdeclarations uninspected by design (floor + bond are the guards); bond sizing in the lab is a stub (100 test units, 5 per misdeclaration) and bonds on this surface are test/stub units — no real compliance-bond funds are accepted, because the production custody mechanism for real bond funds is OPEN on the record; MEMO-121 records leg-up #1 as NO and the claimed population as the broadest and least bounded of its batch; and the standard lane's full-burn worst shape does not clear without classification caching / green-lane reuse, which ship as priced-shape controls.

Service: https://textforagents.com/products/customs-house-for-agent-payloads/ — docs, OpenAPI, llms.txt, terms, and privacy are public there without auth. Support: support@textforagents.com.