Kalshi — Historical Trade Tape

A market is a sequence of decisions. This dataset preserves their traded expression.

Time Seventeen organizes Kalshi's public execution records into a research-ready, UTC-partitioned Parquet archive. The release provides transaction prices, contract quantities, market identifiers and execution timestamps for studying how event markets absorb information over time.

This is a public, non-combination trade dataset. It is an independently maintained research release, rather than an official Kalshi data product. The underlying source is Kalshi's public market-trades API; the published schema is the archive's schema, not a promise that today's API returns exactly the same fields.

Coverage and daily updates

The archive begins with the 30 June 2021 UTC transaction-date partition and is updated daily, publishing completed-day data as it becomes available from the source and passes validation. The start date describes this archive's first published partition, not an independently established exchange launch date.

Update cadence and published coverage are separate: the dated manifest snapshot below reports the files actually available at that revision. It must not be read as a live “updated through today” counter. Source delays, ingestion gaps and subsequent corrections can affect publication; use the repository files and manifest to establish the latest available date. The current UTC day is not presented as complete.

At a glance

The following snapshot is calculated from the published manifest.json, last updated 26 October 2026. Counts describe that manifest revision, not a continuously refreshed claim.

Property	Published snapshot
First / last nonempty UTC trade date	30 June 2021 / 25 October 2026
Nonempty daily partitions	1,943
Execution rows	1,759,100,164
Storage	Parquet; one file per UTC trade date
Excluded instruments	Combination / multivariate tickers beginning KXMV
Access	Public on Hugging Face

The first and last dates are archive boundaries, not proof of uninterrupted or exchange-complete coverage. Consult the manifest for file-level row counts, checksums, verified empty dates and source progress.

What the release contains

daily/
  2021_06_30.parquet
  ...
manifest.json
README.md

Each row represents one observed execution. Records from open and closed markets are assigned to the UTC date of created_time, independently of the contract's closing or settlement date. Deduplication uses trade_id within each UTC date. Row order is not guaranteed; sort explicitly for time-series work.

Only nonempty, closed UTC-day partitions are published. A missing file is not automatically a zero-trade day: verified empty days are recorded separately from unavailable coverage in the manifest.

Schema and units

Column	Arrow type	Meaning
trade_id	string	Source execution identifier
ticker	string	Market identifier
created_time	timestamp[us, UTC]	Execution time; determines partition date
count_fp	string	Decimal contract quantity preserved as text
count	float64	Convenience representation of contract quantity
yes_price, no_price	float64	YES and NO prices in cents
yes_price_dollars, no_price_dollars	string	Decimal USD price representations
taker_side	string	Source-reported taker side, where available

Keep prices and quantities in one consistent unit system. A YES price of 76 cents corresponds to 0.76 USD. Prefer decimal dollar strings for precision-sensitive calculations; use Decimal, rather than binary floats, when exact decimal arithmetic matters. Do not round fractional quantities to integers. Check null availability before interpreting a side or price field.

The two complementary price columns describe the same execution, not two independent trades. Summing YES and NO notionals would change the economic measure. State whether “volume” means contract quantity, execution count or a particular dollar-notional convention.

Construction and verification

The pipeline collects public execution records, partitions them by UTC execution time, deduplicates by execution ID and excludes KXMV combinations. A completed daily partition is bound to all 24 local source-hour receipts, scanned for schema, date bounds, valid IDs, quantities and prices, and checked for combination exclusion. After publication, remote file size and SHA-256 are checked against the uploaded artifact.

These checks establish local compaction consistency and transfer integrity. They do not independently prove that the upstream historical API supplied every execution. The snapshot's historical_backfill_complete flag describes pipeline progress, not a stronger source-completeness guarantee. Revised data should be studied at a pinned repository revision.

Read a partition

Install huggingface_hub and pyarrow, then download only the dates needed:

from decimal import Decimal
from huggingface_hub import HfApi, hf_hub_download
import pyarrow.parquet as pq

repo = "TimeSeventeen/Kalshi"
revision = HfApi(token=False).dataset_info(repo).sha
path = hf_hub_download(
    repo_id=repo, repo_type="dataset", revision=revision,
    filename="daily/2021_06_30.parquet", token=False,
)
table = pq.read_table(path)
print(table.schema)
print(table.select(["ticker", "created_time", "count_fp"]).to_pylist()[:3])
# For a non-null dollar-price string: Decimal(value)

Record revision with research outputs. For large samples, use partition selection, column projection and batch reads instead of loading the entire archive into memory. The Hub dataset configuration is daily; train is a loading split, not a curated machine-learning training/test division.

Research scope

The tape supports transaction-price histories, activity profiles and event-market microstructure research. It does not contain historical order-book depth, quote changes, cancellations, participant identities or a joined market-rule/settlement metadata table. A traded price is an observed transaction price, not a currently executable bid or ask. Calibration and event-study work require an independently verified mapping to contract rules and outcomes, with point-in-time joins to avoid look-ahead bias.

Sources and attribution

• Kalshi public market-trades API documentation
• Published files and revision history
• Curated by Time Seventeen.

When referencing this release, include Time Seventeen, Kalshi — Historical Trade Tape, the Hugging Face URL, the repository revision and your access date.