There are two halves to a market data product. One is a live computation over a book that anybody with the same inputs can reproduce. The other is a record of what that computation said at every moment since you started running it.
Only the second one is scarce.
Why it cannot be recreated
Positioning is computed from an order book and an options chain as they stood at an instant. Those states are not archived by the venues in a form that reconstructs the derived view, and the derived view depends on choices — which books count, how open interest is filtered, where the boundaries fall — that would need reproducing exactly. In practice, a reading not stored at the time is gone permanently.
A competitor entering the market tomorrow with ten times the funding starts their history tomorrow. That is the whole moat, and it is the only one in data that money genuinely cannot shorten.
What the history is for
Two things, mainly. Context, so that a reading can be expressed as a percentile against its own past rather than as a bare number nobody can scale. And base rates — what has actually tended to follow the configuration currently on screen, measured rather than asserted.
Both of those require a long, unbroken, honestly recorded series. A gap in the record is not just missing data; it biases every statistic computed over it, usually in the direction of calm, because gaps tend to happen during the exact conditions that break collectors.
Append-only, on purpose
A record you can retroactively edit is a record nobody should trust, including you. Ours is append-only: a reading that turns out to have been wrong is annotated, never overwritten, because the alternative is an archive that always agrees with whatever we believe today.