TRANSACTOLOGY ·XYZ

The transaction is the smallest unit of social grammar. Two parties, a verb, and the silent witness of a third — a clock, a clerk, a chain of cryptographic hashes — combine to bind a moment in the public record. Without the witness, there is no transaction; there is only a private opinion, held by two people, soon forgotten.

Consider the simplest case: a hand passing a coin to another hand. The coin is the object, the hands are the parties, and the witness is the observable matter of the world — the silent agreement that what was once here is now there. The world remembers. This is the rudimentary form of every later, more elaborate transaction.

From this simplest case we generalize. The parties may be many. The verb may be conditional. The witness may be a chain of computers, none of whom trust each other, who arrive at agreement only through the slow ceremony of consensus. The grammar grows, but the structure does not change: parties, verb, witness. To study transactions is to study the various shapes this triple can take.

It is tempting, when first studying these structures, to focus on the witness — the database, the ledger, the protocol. The witness is the easiest to draw. But the deeper questions live in the verb: what does it mean to commit? What does it mean to reverse? When can a verb be undone, and at what point in time does it become irrevocable? These are old questions; they are also the only questions worth asking.

This monograph proceeds in four chapters. The first, which you are reading, names the parts. The second considers the witness in detail — its shapes, its failure modes, its slow improvement over centuries. The third considers the verb. The fourth, briefly, considers the parties — what we mean by identity when nobody is in the room.

The earliest witnesses were stone. A boundary marker, scratched with a date, served as the public memory of who owned which strip of land. Stones are slow to write, slow to read, and impossible to forge in private. Their virtues have been the design target of every subsequent witness.

Paper improved on stone in writing speed and lost ground on durability. Magnetic disks improved on paper in capacity and lost ground on legibility — a hard drive is illegible without a precise apparatus to read it. Each improvement bought a new vulnerability. The history of the witness is the history of trading one set of risks for another.

The modern distributed ledger is the latest figure in this lineage. It trades the simplicity of a single witness for the resilience of many — at the cost of speed, of energy, and of the careful coordination work we now call consensus. To call any of these a "solution" is a category mistake. They are stages in a long conversation about how the world should remember.

Within this conversation, three families of witness recur: centralized, where one party keeps the books and is trusted to be honest; federated, where a small group keeps the books and is collectively trusted; and decentralized, where the books are kept by anyone who chooses, and trust is replaced by mathematics. Each family has done useful work; none has finished its work.

The verb of a transaction is its theory of time. A transaction that commits declares: this happened, and the world will hereafter remember. A transaction that aborts declares: nothing happened; please erase any provisional traces. The space between these two declarations — the moment when the result is not yet decided — is the most interesting territory in the field.

Outside this provisional space, the world expects determinism. Within it, the world tolerates ambiguity. The art of the systems engineer is to keep this ambiguous interior as small as possible while doing as much useful work as possible inside it. Two-phase commit is one such art. Optimistic concurrency is another. Sagas, which decompose the verb into many smaller verbs each individually committable, are a third.

To reverse a transaction is to admit that the verb was incomplete. In a database, reversal is a rollback — a clean undo, possible because the system held the old state in reserve. In the world of paper and contract, reversal is a refund, a chargeback, a writ of recovery; it is much messier and never quite restores the original. The two registers — clean and messy — coexist uneasily in modern systems.

The simplest model of a transacting party is a name. The next simplest is a name plus a key. The next is a name, a key, and a history. Beyond that we begin to disagree, and the disagreement does not resolve. Identity, in the language of transaction systems, is whatever set of facts is sufficient to bind a party to a verb.

This formulation is unsatisfying. It moves the difficulty without solving it: now we must agree on what counts as a sufficient binding. In law, the binding is signature, witness, and the consent of the bound. In cryptography, the binding is a public key, a private key, and a signature scheme. In some modern systems, the binding is a reputation — a long history of past behavior, weighed against the current claim. Each of these is a different theory of what makes a person, in a given moment, themselves.

This monograph closes here, not because the subject is exhausted, but because the next things we have to say require a different room. The reading list at // archive contains every paper we have learned from. The fellowship at // .org meets thursdays. We hope you have read carefully.