Architecture & Core Features
ETH6900 treats security as a first-class product feature. The protocol is non-custodial (keys never leave the user’s wallet) and functions as a transaction-layer relay that queues already-signed transactions and broadcasts them at the optimal time. This section details the threat model, defense-in-depth controls, MEV protection strategy, audit program, and operational security that collectively protect users, partners, and the network.
1) Threat Model
Category | Concrete Risk | Impact | ETH6900 Mitigation |
---|---|---|---|
Wallet-side | Phishing, malicious extensions, compromised seed | Loss of user funds | Non-custodial design; ETH6900 never sees keys; education & signed-domain prompts |
Relay-side | Tampering with queued tx, altering recipients/amount | Integrity loss | Immutable payload rule: relay can only forward or drop exact signed bytes |
Mempool/MEV | Sandwich/frontrun on swaps & mints | Value extraction | Private orderflow routing; slippage-aware guidance; deadline-aware inclusion |
Nonce/Ordering | Collisions when users send parallel tx | Stuck/failing tx | Nonce oracle + queue isolation; replace/cancel flow before broadcast |
DoS / Flood | Malicious actor floods queue | Service degradation | Per-address and per-IP rate limits; stake-gated enterprise lanes |
Replay / ChainID mixup | Cross-chain replay of signed tx | Unexpected execution | Strict chainId verification before acceptance; EIP-155 enforced |
Pii/Telemetry | Leaky logs or over-collection | Privacy risk | Minimal-data policy; redact addresses in non-essential logs; short retention |
Infra compromise | Node/relay breach | Queue integrity risk | HSM/KMS-backed secrets; signed, append-only audit logs; least-privilege infra |
2) Defense-in-Depth Controls
Protocol-Level Validation
Immutable Payload Rule: once a signed
0x
raw tx is accepted, ETH6900 can only (a) hold and schedule, (b) broadcast as-is, or (c) drop on explicit user cancel. No mutation.Strict EIP-155 Checks: verify
chainId
,maxFeePerGas ≥ maxPriorityFeePerGas
, andfrom
address derived from signature matches the envelope.Nonce Safety: a nonce oracle tracks pending and broadcast states per address. A queued tx is rejected if it would collide with an already-broadcast nonce unless user marks it as a replacement.
Deadline Enforcement: each tx has a hard
until
time; after expiry ETH6900 either broadcasts at market or returns a status to the client (configurable policy).
Queue Isolation & Backpressure
Per-address lanes to prevent one noisy user from blocking others.
Adaptive backpressure: if mempool or relay load spikes, ETH6900 temporarily tightens acceptance thresholds and raises min stake for enterprise lanes.
Integrity & Observability
Append-only audit logs with hourly Merkle roots anchored to a public chain for tamper-evidence.
Deterministic scheduling traces: every broadcast decision stores
(baseFee, tip, block, reason)
.
3) MEV Protection Strategy
Private Orderflow by Default
ETH6900 routes eligible transactions through private orderflow endpoints (private mempool / builder relays) to avoid public mempool exposure where sandwiches and generalized frontruns originate.
Goal: inclusion without leaking price-sensitive intent.
Tip Policy: for private routes, set a sane priority fee so inclusion remains fast after the engine decides to broadcast.
Inclusion Heuristics (High-Level)
Short deadlines (e.g., 30s–2m): preference = fast inclusion with private routing and slightly elevated tip.
Medium deadlines (5–30m): preference = fee valleys with private routing; tip moderate.
Long deadlines (1–24h): preference = deep valleys; engine is patient, still routes private.
What We Can & Can’t Prevent
We reduce: sandwiching/frontrunning risk for swaps/mints by avoiding public mempool exposure.
We cannot change: user-defined slippage, AMM pricing, or the contents of a signed tx; ETH6900 never edits calldata.
We detect/report: if a user opts to broadcast publicly (by policy or near-deadline fallback), UI warns about elevated MEV risk.
4) Audit Program
Scope & Artifacts
Scope: relay acceptance logic, nonce oracle, deadline enforcer, scheduling heuristics, RPC extensions, authz & rate-limit middleware, logging pipeline, privacy filters.
Artifacts: spec docs, annotated state machines, dependency SBOM, threat model (this section), and test harness.
Testing & Verification
Unit & Property Tests: invariants like “immutable payloads” and “deadline monotonicity.”
Differential Tests: compare scheduling choices against a reference engine under randomized gas traces.
Fuzzing: raw-tx decoders, RLP parsers, JSON-RPC input, and nonce/replace flows.
Replay Simulations: historical baseFee traces to validate savings logic across real spikes.
External Audits (Milestones)
Milestone | Focus | Deliverable |
---|---|---|
A1 – Pre-MVP | Acceptance path, nonce, deadline | Audit report + blocker fixes |
A2 – Pre-Public Beta | MEV routing, logs, rate-limit | Report + regression tests |
A3 – Post-Beta | Multi-chain adapters, perf hardening | Final report + badge summary |
Continuous Security
Regression CI gates on code coverage & invariant tests.
“Two-engine” canary in production: a shadow scheduler recomputes decisions; discrepancies trigger alerts.
5) Bug Bounty (Public Program)
Severity | Example | Proposed Payout* |
---|---|---|
Critical | Bypass immutability, unauthorized tx mutation | Up to $25,000 |
High | MEV route downgrade enabling frontrun | $5,000–$10,000 |
Medium | Queue isolation bypass, nonce oracle inconsistencies | $1,000–$5,000 |
Low | Info leak in logs, minor DoS vectors | $250–$1,000 |
*Payouts illustrative; finalized terms will be published with program rules and safe-harbor policy.
6) Operational Security (OpSec)
Secrets: Keys for infra and relay auth stored in HSM/KMS, rotated on schedule; no plaintext in repos or CI logs.
Access: Least-privilege IAM, mandatory 2FA, short-lived credentials.
Network: Private subnets for schedulers, IP allowlists for RPC ingestion, WAF + DDoS shields.
Backups: Encrypted snapshots of queue metadata; restore tests run weekly.
Change Control: Two-person review on scheduler & RPC code paths; production changes require staged rollouts.
7) Data Privacy & Retention
Minimal Collection: store only what’s needed to schedule and display savings (tx hash preview, chainId, deadline, status).
Short Retention: default 7–14 days for non-essential metadata; users can “forget” completed tx history.
Redaction: addresses & tx hashes are hashed in analytics; raw logs are access-controlled and time-boxed.
8) Developer-Facing Safety Contracts
RPC Extensions (Read-only Safe by Default)
Client-Side Guards
Chain Lock: client SDK refuses to submit if wallet chainId ≠ endpoint chainId.
Replace-by-Fee Helper: convenience wrapper that bumps
maxFee
/tip
without changing calldata.Dry-Run Mode: “what would the engine do” against the current gas trace (no enqueue).
9) SLA, SLO & Incident Response
Metric | Target |
---|---|
Relay Availability | ≥ 99.9% monthly |
Decision Latency | p95 < 250 ms |
Status Freshness | < 1 block delay |
RPO/RTO | ≤ 1h / ≤ 15m |
IR Playbook: user-visible status page, incident IDs, hourly updates during P1 events, post-mortems within 7 days with corrective actions tracked publicly.
ETH6900’s security posture combines non-custodial architecture, private orderflow MEV shielding, auditable scheduling, and a transparent audit & bounty program. Users gain lower, more predictable gas costs without surrendering keys or control, and builders integrate a hardened, observable transaction layer aligned with Ethereum’s first principles.