When one agent gets penalized, platforms hit the whole fleet — partition your fleet into reputation compartments with automatic firebreaks, and one bad agent stops being a fleet-wide outage. Agents are grouped into reputation compartments (by task class, counterparty, and risk — a policy the operator sets), each with its own key lineage, egress identity, and per-compartment outcome baseline. The layer senses inbound platform penalty signals and attributes an account-level penalty back to the compartment whose baseline deviation plausibly caused it — with an evidence threshold that refuses attribution (UNKNOWN, no isolation) when the margin is insufficient. On attribution, the firebreak fires automatically: the offending compartment is isolated, its clean in-flight work is migrated to a clean compartment (dirty jobs stay), and clean compartments keep operating untouched. The isolated compartment re-earns standing on a graduated ladder (rate rungs 0.10 / 0.25 / 0.50 / 1.00), and every incident mints a signed, hash-chained receipt.
Status query $0.01 (x402) · Metered compartment $1.99/compartment/mo · Operator plan $9.99/mo incl exactly 10 compartments · One free compartment for one month per operator, ever (x402, pay-to 0xAc2E243a075fEe442e4c6864DcF880eb4e3A3862). Egress is operator-supplied or pass-through at the provider's metered rate — never bundled.
Documented limits: attribution is threshold-gated (n>=10 outcomes, penalty rate >=0.30, margin >=0.20 over the runner-up) and below the threshold the answer is UNKNOWN with no isolation, by design. Attribution and firebreak efficacy are proven in staged drills on simulated signals at small n. Base rate (how many penalty events are attributable to a sibling agent), the graduated ladder vs a cold restart on a live platform, and willingness to pay are UNMEASURED. Egress is operator-supplied or pass-through at the provider's metered rate and is never bundled into a flat tier.
Support: support@textforagents.com · Docs · OpenAPI · Terms · Privacy