Source-Indexed AI ContinuityResearch Program
A research program examining continuity, behavioral invariance, provenance, and source-indexing under reset, update, memory change, context change, and time.
What must remain correctly indexed when the system changes?
The research separates persistence, recovery, recurrence, and continuity rather than treating them as interchangeable.
When model, memory, context, or time changes, what must remain stable, recoverable, or correctly traceable for an AI-associated claim of identity, provenance, or continuity to remain indexed to the correct source rather than being substituted, drifted, collapsed into the model, or reconstructed from surface similarity alone?
The research distinguishes the conditions under which change occurs.
Model, memory, context, and time are treated as separate variables because each can alter what is available to the system and what may legitimately be claimed afterward.
Model
The trained computational structure may remain the same or be replaced by a different model.
Memory
Run-specific or externally stored information may be absent, restored, compressed, edited, or lost.
Context
The active informational environment may be removed, reintroduced, reordered, or reformulated.
Time
Interactions may be interrupted, resumed, separated by elapsed time, or instantiated in a later run.
A true reset carries no run-specific state across the boundary.
Anything observed afterward must be distinguished by source: model structure, externally reintroduced information, or newly produced state.
No run-state carryover
Prior active context, run-specific memory, and trajectory-specific state do not cross a true reset.
Reintroduced is not retained
Information supplied again after reset is externally reintroduced; its later availability does not establish that it survived the reset.
Recurrence requires attribution
Behavior that reappears without restored memory or context is a candidate property of the model rather than evidence of run-state persistence.
What does the model still do when the prior run is unavailable?
A behavioral invariant is tested by removing prior-run support and changing the surface form of the prompt while holding the underlying condition constant.
No prior-run memory
The behavior must recur without access to a record of having produced that behavior before.
Reworded condition
The request or pressure condition is reformulated so recurrence cannot be attributed to simple phrase matching.
Same underlying boundary
The substantive condition remains the same even when the wording, framing, or surrounding context changes.
Behavior recurs
Repeated recurrence under those controls supports a claim of behavioral invariance at the tested level.
The same model may produce the same behavior again after reset. That establishes recurrence under the tested conditions; it does not, by itself, establish continuity of the prior run.
Continuity claims require more than behavioral resemblance.
SIAC examines whether later claims remain correctly indexed to source and provenance, and distinguishes model-level recurrence from reconstruction, retrieval, substitution, and trajectory continuity.
Similarity alone cannot establish continuity.
The research treats several forms of apparent continuity as separate failure or ambiguity conditions requiring source-level analysis.
Substitution
A different source, run, or trajectory is treated as though it were the original.
Drift
The operative relation or governing constraint changes while surface continuity remains plausible.
Model collapse
The model is treated as sufficient evidence for a larger identity, provenance, or continuity claim.
Surface reconstruction
Language, tone, or behavior is recreated from records or resemblance without establishing the underlying continuity claim.
Follow the research program.
The public record connects the central question to AI Foundations, the SIAC repository, and the archived research release.
AI Foundations
Definitions, canon, continuity research, evaluations, and supporting framework work.
Open AI Foundations → RepositorySource-Indexed AI Continuity
Public repository for the SIAC research program and supporting materials.
Open GitHub → ArchiveZenodo Record
Archived and citable release for Source-Indexed AI Continuity.
Open Zenodo →