A community dataset can be sufficient for a prototype. A paid dataset earns its place when its recorded fields and coverage answer a question the prototype cannot. Check those requirements before treating either price or file size as a quality measure.
Our September 25 comparison described a limited Kaggle search. It did not establish that Kaggle has no suitable Kalshi archive, no citation options or no comparable dataset. We have removed those platform-wide claims. This October 8 correction does not present a fresh exhaustive survey of Kaggle's dataset directory.
Start with the research question
| Requirement | What to inspect in any candidate file | A relevant ZenHodl example |
|---|---|---|
| Historical quote states | Bid/ask fields, sizes, sample clock, market selection, missing intervals | Polymarket Sports Archive: sampled quotes and depth, not every book update |
| Kalshi depth | Outcome conventions, level count, ticker selection and quantity representation | Kalshi Microstructure Tape: selected-market depth, up to 20 levels per side |
| Captured trades | Trade IDs, deduplication, exchange time versus retrieval time | Kalshi tape's 29,984,492 captured deduplicated prints in the October 2 release; not certified complete exchange history |
| Matched two-venue quotes | Game identity, side/complement derivation and join direction | MLB matched archive: nearest-time matching within ±30 seconds can include future quotes |
| Settlement analysis | Label source, unresolved records, canceled contracts and coverage denominator | MLB panel: 1,005 dated games, 965 settled and 40 unresolved |
| Reproducible publication | Version identifier, license, source attribution and stable reference | Free MLB sample on Zenodo; read its license and sample scope |
The comparison must be between named files and versions. “Kaggle data” does not identify an uploader, schema or observation window; “ZenHodl data” does not identify which archive you mean either.
What is specific to our recorded releases
Our public catalog, vintage October 5, lists 27 active archive products from $9 to $199. Slices and bundles overlap. Twenty-seven products are not twenty-seven disjoint datasets.
The October 3 Polymarket sports release contains 1,041,590,302 quote rows and 587,197,084 depth rows. Collection runs on roughly 30-second cycles; websocket state may be reused up to 300 seconds old without an exact-age column. These rows cannot reconstruct every update or prove that a displayed price was executable.
The October 2 Kalshi tape contains 821,979,759 depth rows and 29,984,492 captured prints. Depth is a volume-selected subset, not every market on every cycle. Historical fractional depth quantities were integer-truncated through July 19 and cannot be recovered from those stored rows. Trade creation and retrieval timestamps answer different questions; game context recorded during retrieval is not verified event-time context. See the schema.
The MLB matched archive's 1,394,808 rows are derived quote observations, not 1,394,808 trades. Complement prices may be derived from a recorded side. A backtest using its nearest-time join must filter future quotes, unresolved outcomes and unusable sizes before claiming returns.
These disclosures help assess fitness for a particular project. They do not make every ZenHodl file better than every community upload.
When a community file is enough
Use a specific upload when its fields, dated coverage and license fit your prototype. Download a small portion, inspect clocks and identifiers, then reproduce one result. A current uploader's documentation is the source for its contents; our old search results cannot establish what it contains today.
Consider a ZenHodl archive when you need a named release or field documented in our catalog and samples. Check that the particular product contains it: capture-time score/clock exists on some Kalshi-side archives, not universally across all products or Polymarket rows.
License and reproducibility checklist
For either source, record the exact dataset version, download date, license and relevant file hashes. Determine whether redistribution and commercial use are allowed. A downloadable ZenHodl ZIP is licensed under our data license; paying once does not transfer unrestricted data ownership or entitle you to future releases.
Before paying, inspect our free samples, then compare them with a named community dataset on the same research requirements. Publish the missing intervals and exclusions alongside any result. Data volume alone does not establish model accuracy, execution quality or profit.
Source scope: ZenHodl counts and limitations were checked against the public catalog and linked schemas on October 8, 2026. No current platform-wide Kaggle availability claim is made.