How I would price, size, skew and hedge PropAMM quotes, what the live contracts show today, and a working quoting engine with backtests in Python and the browser.
PancakeSwap owns the BNB Chain routing layer (77% of DEX volume, its own aggregator, Binance Wallet distribution). Prop AMMs already quote the majors there. A first-party PropAMM puts that flow on PancakeSwap's own quotes.
Turn a Binance-referenced fair value into a ladder that wins routed flow and loses little to arbitrage, then keep token exposure and delta inside limits at a known hedge cost.
A quoting engine (Python reference, browser port, parity test) with live paper quoting, backtests against a passive v3 pool, a parameter sweep, live aggregator route shares and an onchain read of the PropAMM contracts.
| Takeaway | Why it matters |
|---|---|
| Refresh speed is the largest lever | In the sample hour, moving from a 1 s to a 3 s refresh cut PnL from $131 to $22 on the same capital. Gas per update is about $0.001, so updating more often is cheap. |
| Thin at the touch, deep behind it | Ladders that grow away from mid kept arbitrage to 5% of volume. A front-loaded ladder doubled the arb share and turned PnL negative. |
| Skew controls inventory, the hedge controls delta | Two separate controls: skew rebalances the pool's token mix through flow, the perp hedge removes price risk at a known cost per trade. |
| Measure markouts by counterparty router | Most PropAMM swaps today come through solvers and private routers. Toxicity by source decides who gets the tight spread. |
Five ways a swap can fill. The PropAMM sits between passive pools and RFQ.
| Venue | What is public | Quant read |
|---|---|---|
| v3 and Infinity pools | Concentrated liquidity and bin pools with hooks on BSC and Base. v3 did $14.6B and Infinity $5.2B on BSC over 30 days (DefiLlama). | Passive LPs pay loss-versus-rebalancing to arbitrageurs; the 0.01% BNB/USDT pool is the price to beat. |
| PropAMM (LiquidFi contracts) | A market maker writes a packed price word (mid, spreads, depth curve, expiry block). Per-epoch input caps widen the spread with volume taken. Listed as propamm_pcs in the Unified Swap API docs. | The quote expires by block, so liveness and update cadence are part of the pricing model. |
| PancakeSwap X (RFQ) | Makers answer rfq_requestQuote within 500 ms; the winner gets an exclusivity window, then a Dutch auction. | Gasless for users, off the public mempool. Pricing logic can be shared with the PropAMM. |
| Aggregator and Unified Swap API | One endpoint quotes the aggregator and PCSX; integrated by Binance Wallet, Rabby, Across and others. | PancakeSwap decides where its own routed flow goes, and the PropAMM competes for it on price. |
| Perps (Aster orderbook) | PCS Perps runs on Aster's orderbook infrastructure. | A possible hedge venue for some pairs, next to Binance futures. |
Prop AMMs hold about 11% of BNB Chain DEX volume and 38% on Solana (DefiLlama, 30 days to 13 Nov 2026). On BNB Chain, Metric holds most of that 11%.
| Venue | BSC 30d | Model | Gap vs a PancakeSwap PropAMM |
|---|---|---|---|
| Metric V1 | $2.19B | Prices against a live reference in deviation bands, 11 chains. | Depends on third-party aggregators for flow. |
| Tessera V (Wintermute) | $0.60B | Prop AMM from a large market maker; Solana, Base, BSC. | Large market maker behind it; no first-party routing on BSC. |
| Kipseli | $75M | Signed-quote, RFQ-style. | Took the largest slice of a 100 BNB KyberSwap route in one read. |
| Lunarbase | $62M | Operator quorum reading several CEX venues; BNB/USDT, BTCB/USDT. | Narrow pair set. |
| ElfomoFi | $37M | Curator vaults priced on oracle inputs. | Small share. |
| Solana leaders (BisonFi, Tessera V, Scorch, GoonFi) | n/a | Oracle-updated curves on the chain where prop AMMs hold the largest share. | The benchmark for how much share prop AMMs can take once routers prefer them. |
Volumes: DefiLlama dexs overview for BSC, read 13 Nov 2026. The PancakeSwap PropAMM itself has no DefiLlama listing, so its volume is not in these totals.
Read from BNB Chain RPC and the KyberSwap API on 13 Nov 2026. The demo's PancakeSwap PropAMM and Aggregator routes tabs re-run these live.
| Measurement | Result | What I would do with it |
|---|---|---|
| Quote update cadence | 280 price-word writes in 2,000 blocks (15 min), median gap about 3.2 s, about 7 blocks. | Backtest the cadence against PnL. In the sample hour, 1 s beat 3 s by about $109 per hour on $400k. |
| Update cost | About 36,000 gas at 0.05 gwei, roughly $0.001 per update. | Gas does not limit cadence. Inclusion latency and key operations do. |
| Updating keys | One externally owned account signs the updates in the samples read. | Hot-key rotation, a second signer and an expiry-based fail-safe belong in the operating model. |
| Swap flow | About 110 swaps per hour. Most transactions target solvers and private routers; a few come through the PancakeSwap aggregator router and 1inch. | Markouts by sender, so tight spreads go to benign flow. |
| KyberSwap BNB to USDT routes | A $1k ticket goes to one venue, which changed between reads (PancakeSwap v3 or a prop AMM). At $100k and above, prop AMMs and RFQ makers took 55% to 80% of the route in my reads. | The size buckets where the PropAMM should win, and the depth it needs to quote there. |
| v3 pool gap vs Binance | In my reads the 0.01% BNB/USDT pool sat within 1 bp of the Binance mid and the 0.25% pool was up to 6 bps away. | Fair value is Binance; v3 pools follow it with a lag set by their fee. |
Contract addresses from KyberSwap's public dex-lib integration (PR #1903, merged 11 Nov 2026, which names the source LiquidFi). The price-word encoding is not public, so these reads cover timing, cost and flow only.
| JD duty | Plan | In the demo |
|---|---|---|
| Pricing and liquidity strategies for PropAMM pools | Fair value from the Binance book, a vol-scaled half-spread, skew by token exposure, a ladder that grows away from mid. One parameter set per pair, tuned per regime. | Live quoting, Method |
| Spread models, depth allocation, volatility adjustments | Half-spread = base + k·σ·√(refresh horizon); depth ratio per level; floors per pair. | Parameters panel, Sweep |
| Inventory management to control token exposure | Exposure limits per token, skew to lean against them, size taper on the side that adds risk, vault caps. | Backtest exposure chart |
| Delta-neutral hedging | Hedge net delta on a perp past a threshold; band sized so hedge cost stays below spread captured. | Hedge cost line in the PnL breakdown |
| Real-time orderbook data into pricing | Binance depth stream for mid and book spread; microprice and lead-lag from the perp as next steps. | Live quoting (WebSocket) |
| Backtest and simulate performance | Replay 1 s bars, score spread capture, hedge cost, gas, inventory and markouts against a passive v3 benchmark. | Backtest, Python CLI |
| Maximise volume, minimise risk | Pick settings on the volume against PnL frontier, with peak exposure as the risk axis, checked out of sample. | Sweep frontier |
What the engine computes each update. Every term is a parameter in the demo.
| Term | Definition | Why |
|---|---|---|
| Fair value F | Binance mid (spot); perp mid as a cross-check. | Where price discovery happens for the majors. |
| Volatility σ | EWMA of squared 1 s log returns, 60 s half-life. | Reacts within a minute to a regime change. |
| Half-spread h | max(min, base + k·σ·√T), T = seconds until the next update lands. | Covers the expected move while the quote is stale. |
| Skew | Centre = F·(1 − skew·h·x), x = exposure / limit, capped at ±1. | A long pool cheapens its ask and sells down through flow. |
| Depth ladder | Levels step bps apart, level k+1 = ratio × level k. | Small size where stale quotes get picked off, more size further out. |
| Size taper | The side that adds exposure scales by (1 − |x|). | Stops adding risk at the limit. |
| Expiry and caps | Quote expiry in blocks, per-epoch input cap. | Matches the contract mechanics; a missed update stops quoting. |
| Control | Rule | Trigger to review |
|---|---|---|
| Token exposure limit | Per token in USD, set from vault size and the hedge venue's depth. | Time spent above 50% of the limit. |
| Delta hedge | Hedge full net delta on the perp when it passes a threshold. | Hedge cost above 40% of spread captured. |
| Pool rebalancing | Skew first; transfer between vault and CEX only when skew cannot clear the imbalance. | Rebalancing cost per $1M volume. |
| Volatility regime | σ drives the spread; a hard widen or pause above a σ ceiling. | Arb share of volume above 10%. |
| Data and liveness | Stale Binance data or a missed update leaves the quote to expire, never to stay live. | Any update gap over expiry. |
| Counterparty toxicity | Markouts at 1, 5, 30, 60 s by sender and router. | Negative 5 s markout for a sender over a day. |
Binance 1 s bars, 13 Nov 2026 07:02 to 08:02 UTC, $400k capital, the same router and arbitrageurs for every row. One window, in sample.
| Setting | Volume | PnL | Arb share | Note |
|---|---|---|---|---|
| PropAMM, 1 s refresh, ladder ratio 1.3, hedged | $1.21M | +$131 | 5% | Spread $205, hedge −$51, gas −$7 |
| Same, 3 s refresh (observed cadence) | $0.81M | +$22 | 4% | Wider spreads lose routed flow |
| Same, front-loaded ladder (ratio 0.7) | $1.38M | −$41 | 11% | Hedge cost doubles |
| Passive v3 position, ±5%, 0.01% fee | $1.16M | −$35 | 19% | Versus holding the starting mix |
Reproduce: python3 quant/backtest.py --symbol BNBUSDT --hours 1 --sweep, or the demo's Backtest tab. The flow share and competing quote are assumptions, set in the parameters panel.
Built from the public job posting (2026), PancakeSwap's docs and developer repository, DefiLlama, KyberSwap's public integration code, BNB Chain RPC and Binance public market data, read on 13 November 2026. The demo is my own prototype built for this application; it paper-trades and sends no transactions. Unsolicited interview homework; happy to walk through any section.
PancakeSwap: developer docs repo (Unified Swap API, PCSX market makers) · Perps
PropAMM integration: KyberSwap PR #1830 · PR #1903
Volumes: DefiLlama BSC DEXs · Solana DEXs
BNB Chain: Fermi hard fork · market-making landscape
Independent homework for the PancakeSwap Quant Developer role · 2026 · edwardtay.com