Post-selection does not modify the past.
It completes the description in which the past becomes readable.
An event takes place. A measurement is recorded. An impact appears. Taken alone, it belongs to a set that is too large, too mixed, where the structure of the system remains distributed among several possibilities. The event is real. It is not yet entirely situated.
Post-selection intervenes afterward.
It does not revisit the impact. It does not displace the trace. It sends no message toward what has already occurred. But it adds a terminal condition without which the description remains incomplete. The past does not change content. It receives its second edge.
This is where retro-causal language demands precision.
Naive retro-causality imagines a future that acts like a signal returned toward the past. It seeks a late hand that would modify the already inscribed event. This image is falsely spectacular. It gives time dramaturgy, not structure.
Structural retro-causality is colder.
It says that a system may not be completely described by its initial conditions alone. Between a preparation and a final measurement, the event belongs to a totality that is readable only with its two limits. The terminal condition does not come to correct the first. It responds to it. The system is not launched from a single edge. It is caught between two.
The two-state vector formalism gives this intuition strict notation. At an intermediate instant, the system can be described by a state evolving from past preparation and by another state defined from subsequent measurement. This double framing does not transform the event into fiction. It affirms on the contrary that initial conditions alone do not always suffice to say what the system is between two measurements.
Post-selection therefore does not invent a structure.
It renders readable a structure that does not appear in the raw ensemble.
The delayed-choice quantum eraser makes this point visible. Taken globally, the impacts do not show the pattern that naive reading would expect. Interference appears only in correlated subsets, when events are grouped according to the associated measurement. The pattern is not printed afterward on the screen. Nor is it available without sorting. It was in the correlations, but these correlations required an ulterior condition to become figure.
This is neither arbitrary fabrication nor immediate revelation.
It is conditional structure.
The error would be to choose too quickly between two impoverished readings. To say that the future changes the past gives the terminal condition a power it does not have. To say that it is only a statistical filter removes from it what it brings: the possibility of describing a system not from a single beginning, but from the ensemble of its constraints.
The post-selected past is not a falsified past.
It is a past with two edges.
An isolated event does not yet know entirely to which past it will belong. The formula seems paradoxical because we treat the past as an already closed mass. But the event can belong to the raw ensemble, to a post-selected subset, to a correlation erased in the mass, to a structure revealed by a final condition. The event does not change. Its physical and descriptive belonging changes.
The final condition is therefore not a late commentary.
It participates in the complete form of the phenomenon.
This operation exceeds physics only if one remains prudent. A historical archive does not modify the letters it preserves, but it can produce a readable past by selecting the documents that will form a series. A dataset does not change the original data, but it fabricates the ensemble in which a regularity will become trainable. A judicial investigation does not change the event, but it establishes the regime in which certain traces will become evidence.
These examples are not proofs of physical retro-causality. They show why post-selection is more than a metaphor. It names an operation where the meaning of an event depends on an ensemble that is not given at the moment of its appearance.
The future does not speak to the past.
It sometimes closes the system in which the past becomes describable.
Doctrine
The future does not rewrite the event. It completes the description of the event.
Post-selection is not an inverted signal. It is a terminal condition that renders readable a structure that initial conditions alone do not suffice to describe. The post-selected past is not false. It is conditional. It becomes complete only between its two edges.
Open vector
A quantum experiment, a historical archive, a training dataset, a financial portfolio, a judicial investigation: each can produce a sharper past by filtering afterward the events that will count.
When a regularity appears only after selection, what has one discovered: a structure of the world or a property of the filter?
Both, perhaps. This is precisely what distinguishes structural retro-causality from a simple reading bias.
