How does a new dataset enter research without becoming an unexamined file on a shared drive? This note describes that path, using a public daily volume file from the New York Stock Exchange as the example, and marks the point where the path stops.
Arrival is not permission
A file can arrive by download, by delivery, or by a researcher’s own collection. Arrival means the bytes are on disk. It does not mean the file is understood, licensed for the use someone has in mind, aligned to a clock, or fit to join to prices. The path into research is a sequence of written checks. Skip a check, and the file can still be opened. It has not entered research. It is only present.
The first check is identity. What does a row represent, who published it, and which version is this? The second is the clock: when was each fact knowable? The third is the population: which names, which days, which venues are included, and which were silently dropped? The fourth is a sample a person actually looks at. A schema and a row count are not a sample. A sample is a few days read closely enough to find the surprise the documentation did not mention.
A dataset enters research when its identity, clock, population, and failure sample have been written down and a person has accepted a named version for a named question.
A public volume file
Daily consolidated volume for New York Stock Exchange names is a plain public example. The file says how many shares were reported for a symbol on a date. Before it enters a study, a researcher asks which prints that volume includes. Auctions, odd lots, and late reports may be inside the total or outside it, depending on the publisher’s rule. Two files that both say volume can differ by enough to matter, and both can be internally consistent. The study has to cite the rule, not the word volume.
The path then asks for a point-in-time join. A symbol in the volume file has to map to the same issuer as the symbol in the price file on that date, including changes of ticker. A holiday has to be a missing session and not a zero. A zero, if it appears, has to be explained, because a zero can mean no trading or a missing row that was filled in. The NYSE example is helpful because the calendar and the symbol list are public. A researcher can check a known holiday and a known ticker change without special access.
Only after those checks does anyone talk about acceptance. Acceptance names the version of the file, the discard rules, the identifier map, and the question the file is allowed to serve. A study of liquidity might accept the volume series. A study that needs buyer-initiated volume would refuse it, because this file does not carry a side. Refusal is a successful trip through the path. The file stays available for the question it can answer and stays out of the question it cannot.
Where the path stops
The path stops at the boundary of research. An accepted dataset may be used in a study, a note, or a test. It may not, by itself, become an order, a public claim, or a silent input to a live system. Crossing that boundary is a different decision, made about a tested idea rather than about a file. The volume series can be interesting and still stop at the research door. Most files should.
The New York Stock Exchange volume file is only the example of a dataset with a dull, checkable definition. The same path applies to a text extract or a picture-derived count. Identity, clock, population, sample, acceptance. If any step is missing, the dataset has not entered research, no matter how many people have already opened it.
This piece does not claim that a volume file is a signal, that acceptance moves a dataset into production, or that every public file is usable. It describes the path up to research, and the door where that path stops. Opening the file, forwarding it, or recognizing the publisher does not move a row past that door. The door is a written acceptance, and a chat message is not a substitute for it.
