Separate liquidation messages, event rows, quantity and accounts

Define the observation unit before the statistic

Write down whether one observation means a received envelope, an item inside its data array, a unique provider event, an execution, an order or an account. These are not interchangeable. A feed example containing a data array establishes a possible batching structure; its exact counting semantics still need the channel definition. A local row identifier identifies your stored record and does not automatically identify a distinct exchange execution.

Choose a denominator matching the question. Average units per recorded event row is total compatible quantity divided by the number of rows retained under the declared replay policy. Average units per message uses a different denominator and changes when the provider batches differently. Neither becomes average loss per trader without an account-level loss dataset.

Keep the population and units fixed

For an event-rate comparison, freeze symbol coverage, observation window, time convention, batching rules and filtering. A jump in rows per minute after adding venues may reflect coverage rather than greater activity. A quote-value comparison additionally needs a consistent valuation basis. Do not compare ten contract units on one instrument with ten base coins on another as though equal quantities implied equal exposure.

Three messages contain six rows and 41 base units

Assume a fictional collector has verified that these six rows are distinct observations, each reporting an incremental executed quantity for the same linear instrument. They arrive in three messages over one fixed minute. Message A contains sizes 2, 3 and 5; message B contains 7; message C contains 11 and 13. There are three envelopes, six event rows and 41 base units. The units-per-row mean is 41 / 6 = 6.8333, while units per envelope is 41 / 3 = 13.6667.

If the provider instead batches exactly these same six observations into one message, row count and total size stay unchanged while envelope count falls to one. Units per envelope rises to 41 without any extra liquidation quantity. The public rows have no account identifiers. Assigning each row to a different account or assigning repeated partial reductions to the same account can fit the observed sizes. The data therefore do not identify a unique-account total, and still less a number of distinct people.

An absent account identifier cannot be reconstructed from size

Equal timestamps, equal sizes or repeated round quantities do not establish common account ownership. Different accounts can generate identical public fields; one account can generate different sizes. Do not infer identities or claim an account-level concentration measure from that resemblance. Even a provider's own account statistic would require its published counting method before comparison with public event rows.