En
← All articles

Databento’s funding puts reproducible market data in focus

Market-data quality depends on time, instrument identity and usage rights, not the size of a funding round.

ColumnTechnologyAuthorOpen Market NotesTypeArticle
A paper tape printed with market charts passes through a clock; a red stamp marks the tape on the right.

Databento announced a $100 million Series C on November 12, 2026. Company announcement Capital raised does not establish whether a dataset suits a particular investigation. For market-data infrastructure, the more useful test is whether another researcher, with the same permissions and conditions, can explain and reproduce the result.

Open Market Notes’ analysis is that a data service creates value not merely by accelerating downloads, but by reducing hidden differences in interpretation. Prices, instrument identity, time semantics and usage rights jointly define the research input. A successful API response establishes that a request was processed, not that all those conditions were satisfied.

A price table needs two notions of time

Nasdaq’s TotalView-ITCH 5.0 specification expresses timestamps as nanoseconds since midnight. Exchange specification That is a format definition, not a guarantee of any vendor’s end-to-end speed. Fine granularity alone cannot remove ambiguity about time zones, trading-day boundaries or record ordering.

Researchers need to distinguish when an event occurred from when their system could obtain the information. Suppose a market event reaches a strategy program after it happens. Scheduling decisions using the earlier time rather than the receipt time could make a simulation look more favorable than the actual operating conditions. This is a hypothetical research problem, not an allegation about Databento’s quality.

Procurement should therefore establish who generates each time field, where it is recorded in the processing chain and how late messages are handled. Combining sources also requires an ordering rule. Two equally timestamped records cannot necessarily be interchanged. Preserving original markers and documented conversions can be more useful than turning everything into one tidy column.

Versions matter too. After a historical correction, a fresh download can differ from the first research input. Reproducibility does not require a ban on corrections; it requires explaining what changed. Teams should retain query scope, acquisition date, file checksums and transformation-code versions. Compare data revisions separately from strategy changes or risk assigning an outcome to the wrong cause.

Today’s ticker is not a complete historical identity

Databento opened its Reference API update preview on November 10, with effectiveness planned for December 10, including point-in-time historical listing status. Update notice Preview availability and production effectiveness remain distinct stages at publication.

The analytical issue is sample construction, not field count. Selecting historical securities from those still present today may omit instruments that have left the market. Even perfectly accurate individual prices cannot repair that selection: the research then examines the past of today’s survivors rather than the market available at the time.

Symbols can also change with corporate events, listing venues or share classes. A team needs explicit identity-mapping rules that distinguish an issuer, a security and a particular listing relationship. Different instruments bearing a similar company name should not simply be concatenated into one price history.

There is a further distinction between when something became effective and when a researcher could know it. Historical status helps answer what was true at a given time. Testing a decision that could actually have been made then also needs a boundary around available information. A point-in-time field is not proof that every form of look-ahead bias has disappeared.

For an interface migration, compare old and new outputs in parallel on a fixed sample. Explain changes in inclusion, exclusion and mapping, and retain a difference report. Establish whether the research definition changed before deciding whether its results improved. That is a stronger acceptance test than checking that requests no longer return errors.

Archive the join rules too

Joining prices to a security master requires more than matching a code: effective dates and listing relationships also constrain the match. Suppose a name change causes one price history to match two identity records. The merged row count grows, but the research may not throw an error. This is a hypothetical engineering example, not a discovered product defect. Teams should inspect row counts before and after joins, unmatched entries and multiple matches rather than relying on a smooth final chart.

Permission is a separate boundary. CME’s API guidance distinguishes display, non-display and distribution uses. Market operator guidance Customers should check applicable agreements against their research and archiving workflows. Read access does not establish unrestricted transfer or indefinite retention. This is not a legal conclusion about any particular Databento contract.

Accept an explainable result, not just a larger archive

Centralized purchasing and a unified interface can reduce duplicate connection work. They can also make several teams dependent on the same cleaning and mapping assumptions. The first saves engineering effort; the second can propagate one misunderstanding. These effects are not contradictory. Shared infrastructure needs traceable transformations and a clear correction process alongside convenient access.

A limited pilot can select one defined market, a bounded period and a research question agreed in advance. Beyond coverage and cost, test how omissions become visible, how revisions are communicated, how identity changes can be rerun, and whether licensing permits the final use. These are proposed evaluation criteria, not performance tests conducted by this publication.

Financing supplies resources to the provider, but does not perform the customer’s validation. A useful delivery lets researchers explain what they saw, when they could have seen it and what they may do with it, then locate the reason when results change. Valuation is neither a data-quality measure nor a signal to buy any security.

Information only. No investment, legal, tax, or financial advice.