TimeSeventeen/Polymarket-US

Polymarket US — Historical Time & Sales

A shared record makes independent inquiry possible.

Time Seventeen turns Polymarket US's public Time & Sales reports into an organized, cleaned and versioned trade archive. The release makes reported execution prices and quantities accessible as UTC-partitioned Parquet, while preserving the timestamp precision and report provenance needed to investigate the record.

This dataset is public on Hugging Face. It covers the Polymarket US reporting source, which is separate from the international Polymarket platform on Polygon. It is not a US subset of the on-chain Polymarket-v1 or Polymarket-v2 datasets, and their identifiers and normalization fields should not be assumed to apply here.

Coverage and daily updates

The archive begins with the 30 October 2025 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 1 November 2026. It describes that manifest revision rather than a live counter.

Property	Published snapshot
First / last nonempty UTC transaction date	30 October 2025 / 31 October 2026
Nonempty daily partitions	367
Reported-print rows	255,063,224
Source reports recorded in the manifest	368
Storage	Parquet; one file per UTC transaction date
Excluded instruments	Combo symbols beginning caoc- or crossd-
Access	Public on Hugging Face

These date boundaries are not proof of complete coverage since exchange inception. The manifest explicitly retains historical_backfill_complete=false. Missing dates, verified empty dates and delayed reports must be interpreted separately.

Source and organization

The source is the exchange's official public Time & Sales reporting page. Reports can be revised and may contain transactions belonging to earlier UTC dates. Consequently, the date in a report filename is not a reliable substitute for transaction time.

daily/
  2025_10_30.parquet
  ...
manifest.json
README.md

The archive assigns each print to a UTC date using the explicit offset in its transaction timestamp. Original timestamp strings, including nanosecond precision, remain intact. Later reports can rebuild and revise earlier partitions; pin a repository revision for reproducible work.

The release includes cleaned daily prints and a file-level manifest. Raw CSV reports, local ingestion state, separate market metadata and live WebSocket observations are not included in this dataset.

Schema and units

Column	Arrow type	Meaning
transaction_time	string	Original transaction timestamp, with offset and fractional seconds preserved
utc_day	string	Derived UTC date in YYYY_MM_DD form
symbol	string	Source instrument identifier
price	float64	Reported execution price in USD, not cents
quantity	float64	Number of contracts
report	string	Source report identifier retained for the selected occurrence

For example, price=0.499 means USD 0.499, not 49.9 USD. Timestamps carrying an Eastern offset must be converted to UTC before time-based aggregation. Preserve the original string if your tooling truncates nanoseconds.

There is no unique trade ID, aggressor side, buyer or seller in this public source. These fields are intentionally absent. Do not invent execution IDs, infer aggressor direction from price changes or interpret report as a transaction-date field.

Cleaning without erasing genuine prints

Reports overlap. The same timestamp, symbol, price and quantity can appear in multiple reports, but identical prints within a report can also represent genuine repeated observations.

The archive therefore retains the maximum multiplicity of an identical print observed within any one report, rather than concatenating all overlapping reports or applying ordinary drop_duplicates(). If one report contains a print twice and another contains it three times, the reconciled tape retains three occurrences. The retained report value provides provenance for an occurrence; it is not a list of every report in which that print appeared.

Combo instruments with caoc- and crossd- prefixes are excluded from derived daily ticks. An ordinary market remains eligible even if it can serve as a leg in a Combo. Prices, quantities and repeated non-Combo prints are preserved by this exclusion.

Verification and coverage

Daily outputs are reconciled against committed report shards using an exact, multiplicity-preserving comparison of complete rows. Source-bound receipts and file SHA-256 checks tie the validated local artifact to the published Parquet. Changed reports can produce new historical versions.

This verifies agreement with the collected reports and publication integrity; it is not proof that every historical exchange trade is available in the source. Only nonempty, closed UTC-day partitions are published. The official reports are delayed, and the latest published date can lag today. This archive is not a real-time execution feed or a quote/order-book history.

Read a partition

Install huggingface_hub and pyarrow:

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

repo = "TimeSeventeen/Polymarket-US"
revision = HfApi(token=False).dataset_info(repo).sha
path = hf_hub_download(
    repo_id=repo, repo_type="dataset", revision=revision,
    filename="daily/2025_10_30.parquet", token=False,
)
table = pq.read_table(path)
print(table.schema)
print(table.select(["transaction_time", "symbol", "price", "quantity"]).to_pylist()[:3])

Save revision alongside results. Select dates and columns before reading a large sample, and use batch processing for full-history analysis. The Hub configuration is daily; train denotes a loading split, not a prepared statistical train/test partition. Sort by parsed transaction time when ordering matters, and do not discard repeated prints during sorting or aggregation.

Research scope

The archive supports transaction-price histories, contract-activity measurement and comparisons of observed event-market trading. Economic interpretation still requires a verified contract definition and outcome mapping. A print is evidence of an executed transaction, not a standing executable quote or an exact physical probability. Cross-platform comparisons require explicit alignment of contract conditions, time zones, settlement rules and coverage.

Sources and attribution

• Official Polymarket US Time & Sales reports
• Published files and revision history
• Curated by Time Seventeen.

When referencing this release, include Time Seventeen, Polymarket US — Historical Time & Sales, the Hugging Face URL, repository revision and access date. This independently maintained research archive is not an official exchange data product.