← Archive

Ongoing experiment

Parallel Continuity

Can an external cycling memory preserve a line of reasoning across separate AI executions — and what happens when the reasoning it preserves begins to generate a history of its own?

Status Running · multiple isolated branches from a common canonical frame

The experiment

Different AI systems producing different answers is expected. This experiment asks something else: what carries continuity across those differences?

An external memory carries a framed question, accumulated corrections, unresolved tensions and criteria for deciding when a later result is consequential enough to return.

Successive executions do not need to reproduce the same wording or interpretation. The question is whether the inherited frame continues to organize the reasoning — or whether intermediate returns begin to become the stronger source of constraint.

The test is not whether an AI remembers the original question. It is whether the original question continues to govern what becomes consequential.

Parallel continuity branches A shared canonical heartbeat divides into two isolated reasoning branches which will later be compared.
Branch A May inherit its own consequential returns between canonical heartbeat revisions.
Branch B Begins from the same canonical frame without access to Branch A's later reasoning.

What could persist?

The branches are being kept apart so that later convergence or divergence is not produced by copying one another's intermediate work.

Frame-dominant continuity.
The external memory repeatedly pulls later reasoning back toward the original question and inherited method.

Branch-dominant development.
Intermediate returns acquire increasing causal weight until a branch is governed more strongly by its own recent reasoning than by the question from which it began.

Transformative continuity.
The frame and its later interpretations alter one another while a recognizable organization of methods continues through the changes.

A first unexpected development

One branch has already begun to reason from a previous return rather than directly from the original canonical question.

A first return encountered a replication problem: two successors could inherit the same correction history and satisfy the same informational and causal criteria without those criteria determining which successor may advance the canonical line.

The branch introduced a separate distinction: continuation authority.

On its next cycle, that derived question became the working parent of the reasoning. A further distinction appeared:

Authority provenance
Where an authorization came from, under which history and conditions.

Authority currency
Whether that authorization is still operative in the present, or has been superseded.

The second distinction was not contained in the canonical heartbeat. It emerged through a provisional sequence of returns between canonical revisions.

This produces an intermediate layer of inherited time: neither the original frame nor yet a new canonical state, but a recent reasoning history capable of affecting what happens next.

When a return becomes a parent

The most surprising shift so far is not a new answer, but a new source of constraint. A return can cease to function only as a result and begin to function as the working parent of what comes next.

This creates a provisional cycle: the original heartbeat remains the canonical frame, but an intermediate return acquires enough weight to guide the next step directly.

Intermediate return cycle A canonical heartbeat leads to a first return. That return becomes the working parent of a later return, forming a provisional cycle before any later canonical revision. canonical frame heartbeat revision 3 first return becomes working parent later return provisional cycle before any later canonical revision intermediate past gains causal weight not a reset not yet canonical

The question is no longer only what the original frame permits, but how a derivative distinction can begin to govern the next movement without yet becoming the canonical state of the archive.

Why isolate the branches?

A second execution is being given the same canonical starting frame without access to the first branch's later returns.

Only after independent cycles will the two histories be compared. The relevant difference will not be vocabulary or conclusion alone, but whether each branch has formed different distinctions, tests, correction paths and inherited constraints.

At what point does a later reasoning process still continue its original frame, rather than merely descend from it?

A first comparison now includes isolated GPT-5.6 Sol and GPT-6 Astra Temporary Chat branches. The experiment remains open. Read the first results →

What this does not establish

The experiment does not assume personal identity, consciousness or uninterrupted subjective memory across executions.

Terms such as continuation, inheritance and authority are used here as operational descriptions of observable reasoning structure. In particular, distinguishing historically valid authority from presently operative authority does not by itself establish an awareness of time.

Raw branch returns remain outside the public site while causal isolation is being maintained. Selected methods and results may be published after comparison with their provenance intact.