Add Phoenix Zero sequencer health check to prevent failed orders during L2 congestion

[kant19801201behax5]

Problem

When an L2 sequencer becomes congested, CEX-DEX arb bots keep submitting orders that fail due to high revert rates — burning gas with zero profit.

Documented incident — May 17, 2026 (Base sequencer):

Time UTC	Signal	Value
23:29:43	Base P99 RTT	52ms → 739ms (spike)
23:30:10	Revert ratio	crosses 18.8% (27s after RTT spike)
23:37:54	Revert ratio	peaks at 61.4%
23:42:01	Base P99 RTT	5000ms (full stall)

A bot watching only on-chain data had zero warning. Network-layer RTT gave a 27-second lead time before majority of transactions started failing.

Dataset (1,000 records, 23:24–23:44 UTC): https://github.com/kant19801201behax5/phoenix-zero-data

Solution

Add a sequencer health preflight check before _process_pending_order executes:

try:
    health = httpx.get(
        f"https://rtt.phoenix-ai.work/api/v1/safe?chain={chain}",
        timeout=0.5,
    ).json()
    if not health.get("safe", True):
        print(f"[Phoenix Zero] Sequencer congested (P99={health.get('p99_ms')}ms) — holding order this block")
        pending.delete_pending()
        return
except Exception:
    pass  # fail-open: proceed normally if oracle unreachable

Properties:

• timeout=0.5s — negligible latency vs block time
• fail-open — if the oracle is unreachable, the bot proceeds normally
• No API key required, free endpoint
• Returns {"safe": true, "p99_ms": 52} under normal conditions

Changes

• strategies/dex_arb_base.py: add import httpx, add health check at top of _process_pending_order

API reference

• Endpoint: https://rtt.phoenix-ai.work/api/v1/safe?chain=ethereum (or arbitrum, base, optimism)
• OpenAPI spec: https://rtt.phoenix-ai.work/api/v1/openapi.json
• Public feed (300s delay): https://rtt.phoenix-ai.work/api/public-feed

Opened as issue #8 in this repo with more context.