base-depth-analysis

Snapshots and plots executable depth for WETH/USDC across Base liquidity venues: the three largest concentrated-liquidity pools, the TesseraSwap prop AMM, and the KyberSwap aggregator as a benchmark.

The question

How much size can you trade on Base at a given price, and who is providing it? Headline pool TVL does not answer this. Concentrated-liquidity pools can be deep inside a few ticks and empty beyond them. A prop AMM prices off a curve its operator updates, with no pool state to inspect. The only view that works for both is the same one: at a fixed block, what does each venue return for a swap of size q, on each side?

This tool builds that curve per venue, plus a combined curve for the pools, so the venues can be compared at the same block and the same sizes.

How the curves are computed

Everything below comes from crates/core/src/curve.rs and the adapters in crates/evm/src.

Size grid. Sizes are set in base units (WETH) from --min-base to --max-base over --steps points, spaced geometrically by default. The grid is dense near the top of the book and sparse at the tail. On the ask side the grid is converted to USDC input using the mid price, so both sides cover roughly the same notional.

White-box pools (Uniswap v3, PancakeSwap v3, Aerodrome Slipstream). State is read at the pinned block: slot0, liquidity, tickSpacing and fee in one Multicall3 call. Tick bitmap words and tick data are then loaded lazily as the swap crosses them. The swap is replayed off-chain with the Uniswap v3 step math (uniswap_v3_math::compute_swap_step). The same code handles all three forks; Slipstream pools are found by tick spacing instead of fee tier, and their dynamic fee is read at the snapshot block. Each segment of the grid is applied to the simulated pool state in turn, so point i is the marginal price of the i-th slice after the earlier slices have moved the price.

Black-box venues (TesseraSwap, KyberSwap). No pool state to replay, so the venue is sampled at each cumulative size Q_i and the curve is recovered by finite differences:

marginal_i = (quote_i - quote_{i-1}) / (base_i - base_{i-1})

Tessera is quoted with eth_call to its tesseraSwapViewAmounts(tokenIn, tokenOut, amount) view at the pinned block. Kyber is quoted through its public aggregator API (/routes) with RFQ sources excluded and gas excluded by default. Kyber can optionally refine the grid: it bisects any interval where linear interpolation misses the mid-point quote by more than --kyber-adaptive-bps.

Unified curve. A greedy router over the white-box pools. For each slice of the grid it quotes every pool at its current simulated state, fills the slice on the pool with the best output, and advances that pool's state. For convex AMM curves this approaches the optimal split as the slices shrink. It is a lower bound on what a real router gets, since each slice goes to one pool.

Per-point output. Each point records cum_base, the segment's base and quote amounts, the marginal price (seg_quote / seg_base), and the average price to that depth (cum_quote / cum_base). For v3 pools it also records the tick the segment ended at.

UI stats. The best bid/ask and depth ladder use the marginal price at each depth across the selected venues. Kyber is left out of those stats: it routes through the same pools, and its samples are not pinned to the snapshot block.

Checks against on-chain quoters

At a pinned block (51,963,857), selling 100 WETH through the simulator and through the pools' own QuoterV2 contracts (--v3-quoter) gave the same average price to within 1e-9 relative for Uniswap v3 and PancakeSwap v3. For Slipstream, a 9.28 WETH sell matched Slipstream's QuoterV2 to within 0.00002 USDC (about 5e-10 relative). cargo test covers the step schedule, v2 constant-product math and the unified curve builder.

Running it

Requires a recent stable Rust toolchain.

cp .env.example .env    # optional; set BASE_RPC_URL to your own node
cargo build --release

Without BASE_RPC_URL or --rpc-url the tool uses https://mainnet.base.org, which rate-limits this workload quickly. https://base-rpc.publicnode.com works for occasional snapshots; for the live view, use your own node or a paid endpoint.

One snapshot to disk (writes snapshot_meta.json and depth_points.csv):

./target/release/base-depth-cli snapshot \
  --rpc-url https://base-rpc.publicnode.com \
  --max-base 100 --steps 32 \
  --out-dir snapshots/latest

Pin a block for reproducible results (applies to every venue except Kyber):

./target/release/base-depth-cli snapshot --block 51963857 --venues uniswap_v3,pancake_v3,slipstream,unified,tessera

Live view, rebuilt every --interval seconds and pushed to the browser over a WebSocket:

./target/release/base-depth-cli serve --interval 15 --max-base 50 --steps 32
# open http://127.0.0.1:8080

Useful flags:

flag	effect
--venues	comma list of uniswap_v3, pancake_v3, slipstream, pancake_v2, aerodrome_v2, unified, kyber, unified_kyber, tessera
--block N	pin all on-chain reads to block N
--side bid|ask|both	restrict to one side
--spacing linear|geometric	size grid spacing
--v3-quoter	quote Uniswap/Pancake v3 through QuoterV2 instead of the local simulator (for cross-checking)
--kyber-steps, --kyber-concurrency, --kyber-adaptive-depth, --kyber-adaptive-bps	Kyber sampling density and rate
--kyber-client-id / KYBER_CLIENT_ID	value sent as x-client-id (default base-depth-analysis)
--tessera-sender / TESSERA_SENDER	msg.sender for the Tessera view call

Output format

depth_points.csv has one row per curve point:

ts, block_number, pair, venue, pool, side, price, cum_base, seg_base, seg_quote,
marginal_price, avg_price, liquidity_tag

snapshot_meta.json records the chain id, block and timestamp, token metadata, the pools included, and the grid and Kyber settings used.

Layout

crates/core   curve construction, step schedule, unified router, CSV/JSON writer
crates/evm    Alloy RPC provider, v2/v3/Slipstream/Aerodrome simulators, pool discovery,
              QuoterV2, Tessera and Kyber quote engines
crates/cli    `snapshot` and `serve` commands; axum server for the UI
web           static UI (Plotly), no build step

Limitations

• One pair, three pools. Pool discovery runs over every fee tier and tick spacing but is then filtered to a fixed allowlist in crates/cli/src/main.rs: Uniswap v3 0.05% (0xd0b5…f224), PancakeSwap v3 0.01% (0x72ab…2d38) and Slipstream tick spacing 100 (0xb2cc…dc59). The Pancake v2 and Aerodrome v2 adapters exist but are excluded by that allowlist. Other pairs work via --base/--quote, but the allowlist has to be edited.
• Quotes are not fills. Every curve is a view-function or simulated quote at one block. It ignores gas, MEV and ordering within the block, and for the prop AMM, any gap between the quoted price and what a taker actually receives.
• Tessera is a black box. Its curve is sampled, not modelled, so its resolution is the grid resolution, and it is not part of the unified curve.
• Kyber is not reproducible. The aggregator API has no block parameter, so Kyber samples come from whatever state its backend has when each request lands. Close-together samples can give a noisy or non-monotone marginal curve. Treat it as a rough benchmark.
• unified_kyber currently samples Kyber with the same settings as kyber, so it duplicates it. It is kept as a separate venue but is off by default.
• Float output. Swap math runs in integers (Q64.96); prices and sizes are converted to f64 only when points are written.