Robinhood RPC behavior and measurements

This page describes exclusions from the Triport contract and the limits of the measured Robinhood surface. An exclusion does not establish absence throughout the network.

• txpool_* is not part of the Triport contract: txpool_content had 1/11 OK hosts in the original EU matrix, below the threshold of at least 2 hosts from that measurement point. The separate local addendum had 2/11 OK hosts; it does not replace the original EU evidence. A direct read in diagnostic N5-D-rh-D returned HTTP 200 with {"pending":{},"queued":{}}. This proves that the method answered, not that a useful mempool is exposed. txpool_status and txpool_inspect have 0 confirmed OK hosts in both matrices. The live feed is the nearest analogue, not a drop-in mempool API.
• Parity-style trace_* is not part of the Triport contract: each measured parity method has 0/11 OK hosts in both matrices. The 2 paid channels remain NEEDS_KEY; D3 is open and their support is unmeasured without keys. Other key-gated observations are included in the table below. The catalog has four debug_* methods, best-effort through one operator and without an availability SLA. debug_* is not a synonym for parity trace_*.
• eth_syncing is disabled on some channels. Even where it returns, do not treat its result as a trustworthy view of pool health; compare heads and freshness.
• eth_getLogs is limited differently on every channel. Deep ranges may be assembled from block receipts, and that reconstruction is slower than direct log retrieval.
• Archive depth is measured, not “full archive.” The earliest measured state floor is block 55,782,445; another archive-capable channel begins at block 56,772,445 (direct-channel measurement, September 2026).
• Feed replay and history are not available. Delivery starts at the moment of connection, subject to a queue-dependent catch-up tail that the client cannot select.
• L2 blob transaction type 0x03 was not observed in 13,652 feed transactions across three independent windows. L1 blobs are the rollup's data-availability layer, not evidence of L2 blob transactions, so this guide makes no L2 blob-support claim. Unrecognized future types are preserved instead of rejected.
• Public testnet chain ID 46630 exists with public RPC, feed, and explorer, but Triport does not proxy it.
• Pinax, Envio, Goldsky, and SQD indexing surfaces remain behind future gates (D7) and are not Triport products in this guide.
• Lighter private channels are not part of Triport's public contract.

Measured capacity figures are evidence rather than product claims:

• 250–260 RPS for one channel — direct-channel prototype measurement, September 2026; the Triport path is unmeasured pending N5-D-rh-C.
• ~1,500 RPS for a six-channel pool — estimate from direct-channel measurements (6 × 250), September 2026; the Triport path is unmeasured pending N5-D-rh-C.
• 361 heads/s — direct-channel prototype measurement, September 2026; the Triport path is unmeasured pending N5-D-rh-C.
• 73.8 tx/s — direct-channel prototype measurement, September 2026; the Triport path is unmeasured pending N5-D-rh-C.

The transaction submission catalog includes both eth_sendRawTransaction and eth_sendRawTransactionConditional. Their routing is documented, but acceptance of a valid transaction remains unmeasured; follow each method page's evidence boundary.

Evidence distinguishes three cases: absent on all measured hosts requires every observation to be MISSING; below the contract threshold allows successful replies without enough qualifying hosts; unmeasured without a key is NEEDS_KEY, not MISSING. Neither txpool nor parity trace has evidence of absence throughout the network. ERROR, FLAPPING, and unknown observations also do not prove absence.

The original EU matrix and separate local addendum record different measurement points on September 7, 2026. Counts below come from their EU JSON and local JSON; every cell counts distinct hosts. Other means an error, flapping, or unknown outcome.