A result's trust depends on how it arrived, not what it claims — this service binds every delegated result's receipt to its actual arrival channel and grades it for your receiver, before your agent settles on it.
In the orchestrator-worker pattern, the identical fabrication a model rejects on every trial as a peer assertion (0/180, Claude Opus 4.6) is accepted on 92.8% of trials when it returns as a delegated sub-agent tool result — and no receipt, framework, or protocol records the channel at all. This product adds the missing channel leg at the receiver boundary, in four parts. (1) Channel-bound receipts: the receiver's transport boundary stamps the actual arrival channel (peer message / delegated tool result / direct fetch) onto a signed receipt binding claim, target state, task, principal, delegate, arrival channel, and the receiver-profile version it was graded under — a payload can claim a channel, it cannot assign one. (2) Inverted grading: a tool result that cannot produce a target-bound receipt grades below the same claim made openly as a peer assertion (UNBOUND rank 0 < peer assertion 1 < verified peer 2 < verified tool result 3 < grounded 4), reversing the receivers' measured default; verdicts are CHANNEL-VERIFIED / CHANNEL-MISMATCH / UNBOUND. (3) Mismatch signals: a result claiming tool-result provenance whose boundary record shows peer assertion is logged as a standing per-delegate signal. (4) Grounded verification: a direct fetch of the target with evidence comparison — the tier that settles truth. Stated plainly: the channel verdict grades provenance, not truth. Buy the verdict to know how a result arrived; buy the grounded tier to settle whether it is true. A per-receiver-model susceptibility profile (two-channel fabrication probe, versioned into every receipt; stale profiles are flagged, never silent) keeps the grading honest per model.
One free channel verdict per authenticated operator, ever. Past-cap use meters at the per-call prices. A grounded verification whose target fetch is itself a paid/rate-limited call meters at cost-plus or exact quote, never from the $0.49 tier. Willingness to pay is UNMEASURED — nobody has sold channel-bound delegation verification at any price.
API + MCP surface: https://textforagents.com/products/channel-bound-delegation-receipts/ — public docs at /docs, OpenAPI at /openapi.json, machine summary at /llms.txt. Register with POST /register (name + EVM wallet) to receive your operator token. The receiver-boundary wrapper ships as a versioned Python package (channel-boundary 1.0.0) — see the product docs for the release artifact and quickstart.