Limit Order Book

Returns reconstructed limit order book snapshots for Kalshi and Polymarket prediction markets. The API replays raw order book deltas from R2 parquet files and captures book state at configurable intervals.

Request

GET /api/v1/lob

Parameters

Name	Type	Required	Description
market	string	Yes*	Market ticker (e.g., KXHIGHNY-26MAR04-B51)
series	string	Yes*	Series ticker — returns all markets in series
source	string	No	Data source: kalshi (default) or polymarket
date	string	Yes**	Single date (YYYY-MM-DD)
from	string	Yes**	Range start (YYYY-MM-DD)
to	string	Yes**	Range end (YYYY-MM-DD, max 31 days)
interval	integer	No	Snapshot frequency in seconds (1–3600). Default: 60
levels	integer	No	Book depth per side (1–50). Default: 10
hour	integer	No	Filter to single hour (0–23)
format	string	No	Output format: json (default), csv, parquet, ndjson
columns	string	No	Comma-separated column names for csv/parquet

* One of market or series required. ** One of date or from/to required.

Snapshot interval

The interval parameter controls how frequently the order book is sampled (in seconds):

Interval	Snapshots/day	Use case
1	~86,400	HFT analysis, microstructure research
60	~1,440	Standard analysis, spread tracking (default)
300	~288	Lightweight overview
3600	~24	Daily book shape summary

Book depth

The levels parameter controls how many price levels are returned per side (bids and asks). Default is 10. Max is 50.

Output formats

Format	Content-Type	Best for
json	application/json	Small queries, single-date
ndjson	application/x-ndjson	Streaming, multi-date, large ranges
csv	text/csv	Spreadsheet export
parquet	application/vnd.apache.parquet	Analytics, DataFrame workflows

Result limits

Format	Max rows
JSON	50,000
NDJSON, CSV, Parquet	500,000

At the default interval=60, a single date produces ~1,440 rows per market — well within limits.

Concurrency

LOB reconstruction is CPU-intensive. A maximum of 4 concurrent LOB requests are processed. Additional requests receive 429 Too Many Requests.

Example

# Single market, single date
curl "https://api.mostlyright.xyz/api/v1/lob?market=KXHIGHNY-26MAR04-B51&date=2026-03-01"

# All markets in a series, hourly snapshots
curl "https://api.mostlyright.xyz/api/v1/lob?series=KXHIGHNY-26MAR04&date=2026-03-01&interval=3600"

# Multi-date range as NDJSON (streaming)
curl "https://api.mostlyright.xyz/api/v1/lob?market=KXHIGHNY-26MAR04-B51&from=2026-03-01&to=2026-03-07&format=ndjson"

# Polymarket source
curl "https://api.mostlyright.xyz/api/v1/lob?market=0xabc123:0xdef456&source=polymarket&date=2026-03-01"

from therminal.markets import MarketsClient

markets = MarketsClient()

# Single market LOB
snapshots = markets.lob(market="KXHIGHNY-26MAR04-B51", date="2026-03-01")

# As DataFrame
df = markets.lob(
    market="KXHIGHNY-26MAR04-B51",
    date="2026-03-01",
    interval=300,
    levels=5,
    as_dataframe=True,
)

# ML features
from therminal.ml import LOBFeatures
from sklearn.pipeline import make_pipeline
from sklearn.ensemble import GradientBoostingClassifier

pipe = make_pipeline(
    LOBFeatures(market="KXHIGHNY-26MAR04-B51", interval=60, levels=10),
    GradientBoostingClassifier(),
)
pipe.fit(dates_train, y_train)

Response

[
  {
    "ts": 1709258460000000000,
    "market_id": "KXHIGHNY-26MAR04-B51",
    "best_bid": 45,
    "best_ask": 46,
    "spread": 1,
    "mid": 45.5,
    "bids": [
      {"price": 45, "qty": 1000},
      {"price": 44, "qty": 500}
    ],
    "asks": [
      {"price": 46, "qty": 800},
      {"price": 47, "qty": 300}
    ]
  }
]

Field	Type	Description
ts	integer	Nanosecond epoch timestamp
market_id	string	Market identifier
best_bid	integer	Highest bid price (0–99 cents)
best_ask	integer	Lowest ask price (0–99 cents)
spread	integer	Ask − bid (or −1 if no book)
mid	float	(bid + ask) / 2
bids	array	Bid levels sorted high-to-low
asks	array	Ask levels sorted low-to-high

Each level has price (integer, cents) and qty (integer, contracts).

Try it

GET /api/v1/lob Send →

market

date

source

interval

levels

Click "Send →" to try it live

Data source

LOB data is ingested separately from a dedicated collector and stored as consolidated parquet files in Cloudflare R2. Each file contains order book deltas (price changes) for a single series on a single date.

The API reconstructs book state by replaying these deltas and snapshotting at the requested interval. This means the data is reconstructed on-the-fly — not pre-computed snapshots.