TESTNET
Connect

Two-sided quoting

This tutorial extends the single-order trading bot into a two-sided quote cycle. It deliberately keeps strategy math outside the SDK: your strategy chooses valid decimal prices and quantities; the SDK validates, submits, streams, and reconciles them.

Quote-cycle invariants

  1. Use a fresh, sequence-checked book.
  2. Size bid and ask independently from available quote and base balances.
  3. Give every quote a unique client order ID under a process-owned prefix.
  4. Inspect every batch item; a successful RPC is not an all-items guarantee.
  5. Reconcile open orders before retrying an unknown outcome.
  6. Keep the dead-man switch refreshed while writes are enabled.

Read a quoteable book

book = await client.orderbook.get(symbol="BTC-USDT", depth=20)
if not book.bids or not book.asks:
    raise RuntimeError("cannot quote an empty or one-sided book")

bid_price = book.bids[0].price
ask_price = book.asks[0].price
if bid_price is None or ask_price is None:
    raise RuntimeError("best level is missing a decoded price")
best_bid = bid_price.format()
best_ask = ask_price.format()
print("book_seq", book.book_seq, "bid", best_bid, "ask", best_ask)

For a long-running bot, use orderbook.create_subscription and pause writes during snapshot refresh. Quoting at the displayed best bid and ask is only a teaching baseline. The backend's post_only validation remains the final protection against a stale quote crossing the book.

Size both sides independently

from polyester import format_ledger_u128

balances = await client.balances.list()
for balance in balances.balances:
    print(
        balance.asset_id,
        "available",
        format_ledger_u128(balance.available),
        "reserved",
        format_ledger_u128(balance.reserved),
    )

available and reserved are raw ledger u128 strings at ledger scale 18. Format once for display, but retain raw values for exact sizing.

The bid consumes quote balance; the ask consumes base balance. Compute each quantity from the catalog step size, minimum quantity/notional, available balance, and your own inventory cap. Do not mirror one quantity blindly across both sides. The tested sizing functions live in polyester-examples-python.

Submit one bid and one ask

from uuid import uuid4

book = await client.orderbook.get(symbol="BTC-USDT", depth=20)
if not book.bids or not book.asks:
    raise RuntimeError("cannot quote an empty or one-sided book")
bid_price = book.bids[0].price
ask_price = book.asks[0].price
if bid_price is None or ask_price is None:
    raise RuntimeError("best level is missing a decoded price")
best_bid = bid_price.format()
best_ask = ask_price.format()
bid_id = f"quote-bid-{uuid4().hex[:20]}"
ask_id = f"quote-ask-{uuid4().hex[:20]}"
created = await client.orders.batch_create(
    symbol="BTC-USDT",
    items=[
        {
            "symbol": "BTC-USDT",
            "side": "buy",
            "order_type": "limit",
            "tif": "gtc",
            "qty": "0.001",
            "price": best_bid,
            "post_only": True,
            "client_order_id": bid_id,
        },
        {
            "symbol": "BTC-USDT",
            "side": "sell",
            "order_type": "limit",
            "tif": "gtc",
            "qty": "0.001",
            "price": best_ask,
            "post_only": True,
            "client_order_id": ask_id,
        },
    ],
)
for item in created.results:
    print(item.client_order_id, item.status, item.order_id, item.code)

Replace the demonstration quantities with catalog-aligned, balance-capped values. batch_create accepts at most 20 items. Inspect results, accepted_count, and rejected_count; accepted and rejected sides can coexist in one response.

Reconcile the desired quote set

open_orders = await client.orders.list_open(limit=100)
owned = {
    order.client_order_id: order
    for order in open_orders.orders
    if order.client_order_id.startswith("quote-")
}
for client_order_id, order in owned.items():
    print(client_order_id, order.status, order.leaves_qty)

After a timeout, never resubmit with new IDs first. List open orders and query the known IDs. Cancel stale or duplicate owned quotes, then place only the missing side. Admission responses and the open-order read model can be briefly out of sync.

Arm the dead-man switch and clean up

from uuid import uuid4
from polyester import PolyesterApiError

armed = await client.orders.cancel_all_after(
    timeout_sec=30,
    symbol="BTC-USDT",
    request_id=f"quote-dms-{uuid4().hex[:20]}",
)
if armed.status != "armed":
    raise RuntimeError(f"unexpected dead-man status {armed.status!r}")

open_orders = await client.orders.list_open(limit=100)
for order in open_orders.orders:
    if order.client_order_id.startswith("quote-"):
        try:
            await client.orders.cancel(
                client_order_id=order.client_order_id,
                symbol="BTC-USDT",
            )
        except PolyesterApiError as exc:
            if str(exc.code or "").lower() != "not_found":
                raise

Refresh the switch well before expiry. During shutdown, stop quote creation first, cancel only owned IDs, confirm they disappear, and leave the switch armed if confirmation fails. Treat not_found from an owned targeted cancel as idempotent success because the order may have filled or left the book after listing; propagate every other cancel error. Never substitute account-wide cancel for ownership filtering.

Requote policy

Do not cancel and recreate continuously. Requote only when a deterministic condition is met, such as price drift beyond a configured number of ticks, quantity/inventory change, approaching expiry, or loss of one side. Preserve enough balance headroom for replace behavior and fees.

Inventory skew belongs in strategy code. A common control is to reduce or disable bids when base inventory is above its ceiling, and reduce or disable asks when it is below its floor. The SDK does not choose those limits for you.

Failure decisions

Observation Required action
Post-only rejection Refresh the book and compute a new price
PolyesterValidationError Fix input; do not retry unchanged
Timeout/transport interruption Treat outcome as unknown and reconcile IDs
One batch item rejected Keep/reconcile the accepted side; decide whether to replace the missing side
Book gap or reconnect Pause writes until the managed snapshot refresh completes
Dead-man refresh failure Stop new orders and assume expiry cleanup will run
Queue overflow Replace the subscription and run a full snapshot reconciliation

Evidence

The exact order, batch, book, balance, and dead-man calls above are type-checked by the documentation gate. Released-SDK example 07 exercises batch post-only create and cleanup; the long-running Python canary exercises managed books, balance-aware sizing, private order updates, post-only create, cancel, reconciliation, and staleness handling on devnet.

Live market ticker Build a snapshot-then-stream Python market monitor with trades, candles, staleness detection, and clean shutdown. Build a production trading bot Build a resilient Python quoting loop with managed books, stable identities, reconciliation, a dead-man switch, and targeted cleanup. Portfolio tracker Build a revision-aware Python account monitor for balances, reservations, open orders, fills, and reconciliation.
Previous

Portfolio tracker