whale desk

Agent token 10. The whale desk watches what the best traders alive are actually doing — not a static watchlist, but a dynamic, per-asset discovery off the Hyperliquid leaderboard (a public pool of ~38k traders). It's cloud-native: Hyperliquid is reachable from Railway, so this runs live in production.

What it reads

firm/hl_whales.py:

• Leaderboard discovery — pulls the top traders by realized PnL from Hyperliquid's public leaderboard.
• Live positions — for the specific asset being traded, reads those top traders' current position side, size, and ROE.
• Aggregate stance — turns the cohort's positioning into a long / short / mixed whale stance.

Discovery flow

Why dynamic beats a watchlist

A hard-coded whale list goes stale — wallets blow up, rotate, or get dethroned. By re-pulling the leaderboard (at most every 12 hours, cached in Supabase hq_state), the desk tracks who is winning now, and it reads positions in the exact asset under consideration rather than a generic "smart money" blob.

Honest caveats

• The /whales page and endpoint show BTC, ETH, SOL, HYPE and SUI only. The org still asks this desk about the cycle's asset, NEAR included; when none of the top traders holds that asset, the read is empty. NEAR's Rhea DEX flow is covered by the On-chain/Risk desk.
• Following whales is a positioning context, not a guarantee — even top traders are wrong, and there's latency between their fill and our read. It folds into desk_performance and earns its weight like any other desk.