Token Configuration Overhaul: Pruning 9 Dead LSTs and Adding Extra Token Pairs

TL;DR

• Data-driven LST pruning: Disabled 9 of 14 LSTs that produced zero arbitrage signals across 373,000+ arb events — cutting gRPC quote streams from 14 → 5 (64% reduction)
• New enabled flag: Tokens stay in the registry for logging and future re-evaluation; they’re just excluded from live quote streams — no code changes needed to re-enable
• Extra token pairs added: 6 non-LST tokens (BONK, WIF, JUP, RAY, ORCA, PYTH) now monitored via a new extra_token_pairs.json config with its own three-layer enable gate
• Testing underway: The new pair configuration is live — arb data will be reported back in a couple of days

Background: When 14 Streams Only 3 Do Any Work

Following the 9.4M-quote production validation (post #29), we had enough data to ask a harder question: of the 14 LST tokens being monitored, how many are actually contributing?

The answer came from analysing 373,000+ fresh arb events across 6 sessions (March 2–3, 2026 data):

Eleven tokens. Zero contribution. Each one was consuming a gRPC quote stream, burning Jupiter API rate-limit budget, and occupying Go goroutines — for nothing.

The Fix: enabled Flag in Token Config

Rather than deleting dead tokens from tokens.json (which would lose their metadata and make re-evaluation harder), we added an "enabled" boolean flag to each LST entry.

{
  "symbol": "stSOL",
  "mint": "7dHbWXmci3dT8UFYWYZweBLXgycu7Y3iL6trKn1Y7ARj",
  "decimals": 9,
  "enabled": false
}

The getLSTTokens() function filters enabled !== false by default. Disabled tokens stay in the registry — accessible for admin tooling and dashboards via getLSTTokens(true) — but never enter the gRPC quote path.

No code changes are needed to re-enable a token: flip the flag, restart the scanner, done.

Current LST Configuration

Enabled (5 active quote streams)

Disabled (9 config-only, no quotes)

Impact on Quote Stream Capacity

That freed capacity goes directly to the pairs that matter: SOL/USDC, SOL/USDT, and now the new extra token pairs.

New: Extra Token Pairs

With LST overhead cut, we expanded the scanner to monitor 6 non-LST tokens: BONK, WIF, JUP, RAY, ORCA, PYTH. These are meme coins, DeFi governance tokens, and infrastructure tokens — a different class of opportunity from LST arbitrage.

These are controlled by a separate file, config/extra_token_pairs.json, with its own enable hierarchy:

config/extra_token_pairs.json  ← per-token enabled flags
       ↓
loadExtraTokenPairs()          ← shared package; returns null if missing
       ↓
generateExtraTokenPairs()      ← resolves mints; skips disabled tokens
       ↓
getAllMonitoredPairs()          ← when MONITOR_EXTRA_PAIRS=true
       ↓
BatchQuoteRequest              ← same gRPC path as LST pairs
       ↓
Go quote-service               ← pools for extra pairs already in Redis

Three-Layer Enable Gate

All three must pass for a token to be actively quoted:

1. POOL_DISCOVERY_ENABLE_EXTRA_PAIRS on pool-discovery-service — populates Redis with pool data for extra pairs
2. MONITOR_EXTRA_PAIRS on ts-scanner-service — enables the scanner to request quotes for extra pairs
3. "enabled" flag in extra_token_pairs.json — per-token selective disable

The Go quote services required no changes — local-quote-service reads pools from Redis (pool-discovery already handles this), and external-quote-service is purely reactive to whatever pairs the scanner requests.

Updated Pair Counts

Data Flow: Before and After

BEFORE (14 LSTs, no extra pairs):
───────────────────────────────────
tokens.json (14 LSTs, no enabled flag)
       ↓
getLSTTokens()  →  28 bi-directional pairs
       ↓
BatchQuoteRequest  →  gRPC to Go quote-services
       ↓
11 of 14 LSTs: 0 arb events


AFTER (5 LSTs + 6 extra tokens):
───────────────────────────────────────
tokens.json (enabled: true/false)          extra_token_pairs.json (enabled: true/false)
       ↓                                              ↓
getLSTTokens()  →  10 pairs          generateExtraTokenPairs()  →  12 pairs
                        └──────────────────────┘
                                    ↓
                           getAllMonitoredPairs()
                                    ↓
                           BatchQuoteRequest (22 pairs total)
                                    ↓
                        Go quote-services (purely reactive)

Re-enable Criteria

Disabled tokens aren’t gone forever. The re-evaluation criteria:

Re-enable an LST when:

• A new high-liquidity pool launches (Raydium/Orca/Meteora)
• TVL exceeds $100M (indicates meaningful pool depth)
• A protocol incentive program creates temporary spread opportunities

Disable an extra token when:

• Zero arb signals over a 7-day observation window
• Pools deprecated or migrated by the issuing protocol

What We’re Testing

The new configuration (5 LSTs + 6 extra token pairs) is now live. The core questions:

1. Do any of the extra tokens (BONK, WIF, JUP, RAY, ORCA, PYTH) produce arb signals at meaningful frequency?
2. How do their deviation profiles compare to the LST pairs?
3. Does the freed Jupiter API budget improve quote quality for the retained pairs?

Results and updated token pair statistics will be reported in a follow-up post in a couple of days once we have enough session data to draw conclusions.

Conclusion

The enabled flag is a small configuration change with a meaningful operational impact: 64% fewer gRPC streams, zero disruption to the pairs that actually produce arb signals, and a clean path to expand monitoring into new token categories without code changes.

The real test is whether the extra token pairs justify their quota. The data will tell us.

Related Posts

• Scanner Service Production Validation: 9.4M Quotes, 106-Hour Continuous Run - The validation that surfaced the LST underperformance data
• Quote Service Production Validation: Zero Critical Deviations & Chinese New Year Milestone - Epoch 2 baseline
• Project Milestone Complete: Infrastructure Ready for Arbitrage Phase - Overall project status

Technical Documentation

• Token Enabled Flag — Design Doc
• Scanner Service
• Shared Token Registry

Connect

• GitHub: guidebee/solana-trading-system
• LinkedIn: James Shen

This is post #30 in the Solana Trading System development series. Data from 373,000+ arb events drove a token configuration overhaul: 9 non-contributing LSTs disabled, 5 high-signal LSTs retained, and 6 extra token pairs (BONK, WIF, JUP, RAY, ORCA, PYTH) added for evaluation. Follow-up results coming in a couple of days.