The Market-Maker Desk — professional market making, end to end

Market making is the most consistently profitable activity in crypto's history — and the fastest way to bleed out if inventory is an afterthought. This is the complete guide to the Liquidity Sniper MM desk: how it quotes, how it protects itself, and how the ⚖ dual-book mode with its rescue governor turns hedge mode into a two-sided income machine.

What a market maker actually earns

A market maker quotes both sides of the book — a bid below the price, an ask above — and earns the spread each time both fill. The business is spread × turnover − inventory losses. The first two terms are easy; every failed market-making bot in history died on the third. Price trends, one side keeps filling, and the "neutral" bot discovers it has been buying a falling knife with leverage.

That is why this desk is built risk-first: the quoting engine is wrapped in inventory limits, volatility brakes, momentum guards and a layered stop stack, and every simulated number is computed the pessimistic way. The desk quotes real post-only (GTX) orders live — maker fees only, never crossing the spread — with cancel-replace management that preserves queue position, and an exchange-side sweep that makes a stacked or orphaned order structurally impossible.

The quoting model — Avellaneda–Stoikov, made honest

The desk quotes around a reservation price — the fair value shifted against your inventory — with a half-spread that widens when the market gets loud:

• Half-spread = max(minimum spread, fee floor, base spread + vol multiplier × σ) — where σ is realized 1-minute volatility. Quiet tape → tight quotes; violent tape → the spread breathes out automatically. The fee floor guarantees you can never quote negative-expectancy spreads: the round-trip maker fee plus a minimum edge is the hard bottom.
• Inventory skew shifts both quotes against your position — long inventory pushes bid and ask down so the book helps you unload. The skew per lot is also your averaging ladder spacing: at 80 bps/lot, your second lot fills ~0.8% deeper, your third ~1.6%.
• Volatility brake: σ above your threshold pulls all quotes — a market maker never quotes through a news candle.
• Momentum guard: strong short-term drift pulls the quote that momentum would run over — the classic adverse-selection defense.
• Order-book edges (live): optional microprice anchoring (size-weighted fair value) and book-imbalance lean — accumulate on the side the book favors.

Per-symbol desks & parameter profiles

Every symbol on the desk runs its own mode — ⏸ Off, 🏪 Paper or ● LIVE — and its own parameters. Desk-wide settings apply to every symbol except those with a ⚙ per-symbol profile: save the fields with scope "Only this symbol" and that pair runs its own knobs forever (a calm BTC desk and a wild mover profile coexist on one desk). A 🔍 pair browser ranks the universe by MM-suitability — or by volume, % change, volatility, gainers and losers when you're deliberately hunting movers.

⚖ Dual-book hedge mode & the rescue governor

On a hedge-mode futures account the desk can run two independent books on one symbol: every bid fill builds the LONG book, every ask fill builds the SHORT book, each with its own average and its own take-profit. In chop, both books breathe and both sides earn. The catch is the trend: one book gets trapped. That's where the rescue governor — the desk's signature risk mechanism — takes over:

• Trigger: a leg's unrealized loss reaches your per-leg limit (in USDT) → the losing side stops adding. Its take-profit stays live.
• Income-funded reduction: every realized profit the winning side books is spent reducing the losing leg by the same dollar amount. Realized PnL stays pinned near zero while exposure shrinks — the desk cannot dig itself a realized hole.
• Hysteresis: the leg recovers to half the trigger (or goes flat) → rescue ends, and the rescued side rests through a cool-off before re-opening.
• Hard ceiling: at 2× the trigger the leg is force-flattened at market. Income-funded patience never overrides a stop.

Direction gates — the desk's autopilot layers

Three optional, kill-switched layers steer which side may open (exits are never blocked):

In dual-book these gates pause the against-call book's opening; whatever it already holds keeps its TP and the rescue governor. The layers compose — and when they disagree, the desk stands down from opening rather than picking a fight with itself.

The stop stack

Five independent escape hatches, each anchored to a different thing, whichever fires first wins — covered in depth in the Risk Stack guide:

• Hard inventory limit (lots) — size-anchored: taker-flatten + cool-off.
• Inventory stop-loss % — entry-anchored: adverse move vs your average.
• 4H structural stop — market-anchored: the previous completed 4-hour candle's low/high ± a buffer; structure failed, get out.
• Session loss stop (USDT) — dollar-anchored: day PnL floor; flattens and halts.
• ⚖ Rescue trigger & ceiling — per-leg, in dual-book.

Honest paper mode

Paper fills require the tape to trade through your quote by a penetration margin — and only price movement new since the last cycle can fill an order. One deep wick fills you once, not twelve times; a quote can never fill on movement that happened before it was placed. Expect paper results that look like live results — that's the point. The desk also cross-checks live books against the exchange every minute: position sizes and entry prices are reconciled to exchange truth, loudly, never silently.

🧠 Hyper-optimizing a symbol

The desk's knobs are searchable per symbol on real tick history, walk-forward: the optimizer searches an older window, validates on the most recent one, and a candidate is only saved if it beats your current knobs out-of-sample. Admitted winners become 🎯 per-symbol tuning the desk applies automatically. Details in the Backtesting & Hyper-Optimization guide.

FAQ

Related guides