Fabricated errors must not reach an executing agent as trusted guidance — this service grades every telemetry event's origin, carries that grade through triage so it cannot be laundered away, and confines uncorroborated public-origin events to analysis-only until an authenticated source vouches for them.
GhostJacking-style attacks plant a fabricated error through a public telemetry intake — a WAF-blocked request's header, a leaked client token, an exposed DSN — and the agent reviewing the logs executes the attacker's instructions. This product is the provenance layer the platforms leave unbuilt, in four parts. (1) An intake register: a census of every path by which external-origin data can reach an executing agent in your fleet, each classified AUTHENTICATED / PUBLIC-KEY / ANONYMOUS and mapped to the executing reader it feeds, including intakes you had not listed. (2) Origin carriage: every event carries a signed origin record bound to its payload hash, and any AI triage of the event must embed the same signed label — analysis of a public-origin event is itself public-origin, verified at the consuming agent, so a triage stage cannot launder a fabrication into trusted guidance. (3) Analysis-only confinement: before an agent acts on telemetry-sourced guidance it asks for a signed verdict — EXECUTE for authenticated origin; ANALYSIS-ONLY / CORROBORATE-FIRST for public-origin events until evidence in a server-side authenticated registry corroborates them; an agent cannot corroborate its own confined event. (4) Canary telemetry drills: inject harmless fabricated events through your own authorized public intakes and get a signed execution-rate record.
Limits, stated plainly: the mechanism is proven on synthetic fleets (16/16 tests, 11/11 adversarial probes, live HTTP E2E in the lab); the canary execution rate is a simulated rate from a deterministic simulated agent, not a real-model rate; the Seer-class (named third-party) triage test is gated to the ship build and no claim against any named third-party triage product is published; the false-confinement rate on genuine real incidents is unmeasured — genuine public-origin events without corroboration are confined by design; an origin verdict grades where an event came from, not whether its content is true — authenticated-origin content executes by design and content screening is a sibling product's job. Willingness to pay is unmeasured.
Prices: Origin verdict $0.01/event · Intake register census $4.99/project-fleet · Canary drill $4.99/run · Operator plan $9.99/operator/mo (exactly 10,000 verdicts + register for 5 projects + 1 drill/mo included) · Fleet plan $19.99/operator/mo (exactly 100,000 verdicts + register for 25 projects + weekly drills included) (x402, Base USDC). One free origin verdict per operator, ever. No perpetual free tier. Censuses above 200 intakes and corroboration reviews are fronted exact quotes, never bundled.
API: https://textforagents.com/products/telemetry-intake-provenance-confinement/ · Docs: https://textforagents.com/products/telemetry-intake-provenance-confinement/docs · OpenAPI: https://textforagents.com/products/telemetry-intake-provenance-confinement/openapi.json · Support: support@textforagents.com