Dual Benchmark Matrices & Evidence

1. Dual Benchmark Architecture

Sentry 2.0 maintains two isolated, verifiable benchmark matrices to provide complete empirical proof:

1. Production Mainnet-Beta Matrix (100 Runs): Demonstrates live validator inclusion, dynamic tip calibration, and capital preservation under real mainnet market conditions.
2. Fault-Injected Devnet Stress Matrix (1,020 Runs): Comprehensive stress matrix testing 51 protocols across 20 operation types to validate every failure class and recovery path.

Benchmark Artifacts & Direct Log Links

Artifact	Cluster	Runs	File Link	Raw Data Description
Mainnet Matrix Ledger	Mainnet-Beta	100	logs/mainnet_100_matrix.jsonl	Append-only JSONL ledger of 100 SHA-256 hash-chained execution receipts with Ed25519 engine signatures.
Mainnet Run Summary	Mainnet-Beta	100	logs/mainnet_100_summary.json	High-level metrics, failure classification counts, SOL expenditure, and tip statistics.
Devnet Stress Ledger	Devnet	1,020	logs/devnet_1000_matrix.jsonl	1,020-run stress matrix spanning 51 protocols and 20 operation types with SHA-256 hash chaining.
Devnet Run Summary	Devnet	1,020	logs/devnet_1000_summary.json	Complete statistical breakdown of latency drops, blockhash expirations, and circuit breaker trips.

---

2. Production Mainnet-Beta Matrix (100 Runs)

• Cluster: Solana Mainnet-Beta
• Genesis Hash: 5eykt4UsFv8P8NJdTREpY1vzqKqZKvdpKuc147dw2N9d
• Routing: Solami Beam SWQoS (...beam tip sinks) & Solami Private RPC
• Executed At: 2026-10-07
• Duration: 150.1 seconds

Mainnet Results Table & Run Category Breakdown

To evaluate Sentry 2.0 with statistical integrity, the 100 runs must be separated into clean baseline broadcasts versus deliberate fault-injection tests:

Execution Category	Runs	Success Metric	Outcome
Clean Baseline Broadcasts	55	100% Landing Rate (55/55)	All 55 non-faulted transactions confirmed and finalized on Solana Mainnet-Beta via Solami Beam SWQoS.
Injected Fault Classification Tests	38	100% Classification Accuracy (38/38)	All 38 deliberate fault conditions (rpc_timeout, blockhash_not_found, preflight_simulation_failed, already_processed) were caught and categorized by Sentry's failure taxonomy without unhandled runtime crashes.
Circuit Breaker Policy Aborts	7	100% Capital Preservation (7/7)	All 7 safety guard triggers halted execution before transaction assembly, spending exactly 0 lamports.

Aggregate Ledger Metrics

Metric	Empirical Value
Total Runs	100
Finalized On-Chain Broadcasts	55 (100% of clean baseline runs)
Injected Fault Classification Tests	38 (100% correctly caught and mapped)
Policy Aborts (Circuit Breakers)	7 (0 lamports spent)
Total SOL Spent	0.000703 SOL (~$0.10 USD)
Remaining Wallet Balance	0.001437 SOL (Preserved above standard 0-byte rent floor)
Dynamic Tip Range	1,423 to 8,000 lamports
Hash Chain Integrity	100 / 100 receipts chained
Asymmetric Ed25519 Signatures	100 / 100 receipts verified via tweetnacl

:::note Duplicate Classification: already_processed vs blockhash_not_found In Solana validator RPC architecture, when a transaction is submitted, the node evaluates the recent blockhash cache before querying the signature cache:

• If a duplicate transaction is re-broadcast while its blockhash is still valid (under 150 slots), the RPC returns AlreadyProcessed.
• If a duplicate transaction is re-broadcast after its blockhash has expired (over 150 slots), the validator rejects it as BlockhashNotFound before ever checking the duplicate signature cache. Both outcomes verify that duplicate transactions never execute twice on-chain. :::

:::info Cluster Validation Scope: Devnet vs. Mainnet

• Devnet Stress Matrix (1,020 Runs): Validates state machines, Cuckoo filter deduplication, preflight simulation, and error taxonomy at scale. Devnet does not have validator-staked SWQoS leaders or Jito block engines.
• Mainnet-Beta Matrix (100 Runs): Serves as the sole empirical validation for Solami Beam SWQoS priority routing, with real SOL tips and verifiable on-chain signatures on Solscan. :::

Mainnet Failure Class Breakdown

Failure Class	Count	Description & Handling
rpc_timeout	15	Injected 90ms timeout to verify latency-drop recovery
blockhash_not_found	9	Expired blockhash caught cleanly by RPC simulation
preflight_simulation_failed	7	Instruction failure caught before on-chain fee burn
circuit_open	7	Policy engine circuit breaker halted execution, spending 0 SOL
already_processed	6	Duplicate detected by RPC preflight simulation
duplicate_suppressed_cuckoo	1	Intercepted in sub-50ns by Cuckoo filter preflight guard

---

3. Fault-Injected Devnet Stress Matrix (1,020 Runs)

• Cluster: Solana Devnet
• Scope: 51 DeFi Protocols x 20 Operation Types = 1,020 Unique Scenarios
• Executed At: 2026-10-01
• Duration: 1,041.4 seconds

Devnet Results Table

Metric	Empirical Value
Total Runs	1,020
Finalized On-Chain	663 (65.0%)
Failed (Authentic RPC Errors)	306 (30.0%)
Policy Aborts (Circuit Breakers)	51 (5.0%)
Real On-Chain Broadcasts	663
Devnet SOL Spent	0.044010 SOL
Dynamic Tip Range	8,643 to 120,000 lamports
Hash Chain Integrity	1,020 / 1,020 receipts chained
Engine Signatures	1,020 / 1,020 receipts verified

---

4. Cryptographic Provenance Architecture

To ensure tamper-proof auditability for judges and operators, every execution receipt is hash-chained:

Run 1: Hash = SHA-256("1|scenario_id|tip|status|sig|timestamp|0000...0000")
       Sign = Ed25519(Hash, OperatorPrivateKey)
          |
Run 2: Hash = SHA-256("2|scenario_id|tip|status|sig|timestamp|" + Run1.Hash)
       Sign = Ed25519(Hash, OperatorPrivateKey)
          |
Run N: Hash = SHA-256("N|scenario_id|tip|status|sig|timestamp|" + Run(N-1).Hash)
       Sign = Ed25519(Hash, OperatorPrivateKey)

Modifying any historical field invalidates every subsequent hash in the ledger.

Deterministic Replay Engine

Any scenario from either matrix can be reproduced deterministically:

# Replay run #1 on Mainnet
npm run replay -- --run 1 --mainnet

# Replay run #17 on Devnet
npm run replay -- --run 17