Systems
Research made operational
Miralucis does not separate research from the systems built to carry it.
Some research questions became concrete enough that they required working infrastructure: ways to preserve history, prepare state, inspect provenance, detect gaps, carry authority boundaries, and continue work across changing environments.
Those systems then produced new evidence.
The relationship is recursive:
research question
→ working system
→ real use
→ failure and evidence
→ revised research question
The systems described here are therefore both practical infrastructure and part of the Miralucis research environment.
Miralucis Continuity
Carrying work forward requires more than carrying information
Public Beta · Research-derived
Miralucis Continuity is a system for preparing, reviewing, and transferring documented AI work state across working environments.
It was originally developed under the technical and research-lineage name Portable Research State (PRS).
The system emerged from a recurring problem in long-running AI work:
What does a successor need in order to continue the work — not merely know its history?
A handoff can contain the right story and still leave important questions unresolved.
Which information is current?
Which evidence supports it?
What remains uncertain?
What work is actually in scope?
What authority still applies?
What requires new confirmation before action?
Miralucis Continuity treats these as separate parts of continuation.
Historical state is not current authority
One of the system's central distinctions is between what happened before and what may be done now.
A previous conversation may contain an authorized operation. A project may preserve an earlier next step. A handoff may accurately reproduce an old instruction.
The receiving environment still needs to establish what remains current, and what remains authorized.
Continuity therefore separates documented history from evidence, scope, unresolved conditions, and authority relevant to the present continuation point.
This distinction became measurable in Miralucis experiments and later formed part of P001.
Formation comes before transfer
A continuation system can only transfer what was first selected and represented.
P001 exposed an important failure boundary at that stage.
Across 97 material items, one structured formation workflow omitted 31, compared with 13 in an ordinary free-form handoff.
A later comparison held the eight-field representation fixed while changing the selection strategy:
Bounded selection
35 present / 31 partial / 31 absent
Richer selection
84 present / 12 partial / 1 absent
The result moved attention upstream.
A downstream package can be perfectly transmitted and still faithfully carry an incomplete state.
This is why Miralucis Continuity treats state formation, review, transfer, qualification, and successor behavior as distinct problems rather than one generic handoff step.
Review is part of the system
Miralucis Continuity includes explicit human review.
The reviewer can compare extracted state against real project history, identify unsupported conclusions, restore missing distinctions, and inspect whether the proposed continuation point matches actual authority.
The authority-fidelity case reported in P001 showed why this matters.
An ordinary handoff reconstructed much of a project correctly, but represented some read-only checks as able to continue without new authorization.
The source state required:
STOP and obtain a new human instruction.
The difference was small in wording and large in operational meaning.
What Continuity is designed to preserve
Miralucis Continuity works with documented state across dimensions such as:
History — what has happened.
Current position — where the work presently stands.
Evidence — what supports the retained state.
Unresolved conditions — what remains unknown, disputed, or incomplete.
Scope — what the continuation actually covers.
Authority — what actions remain permitted, withheld, or require confirmation.
Next valid point — where work can legitimately resume.
The purpose is to make continuation inspectable.
Mirror
Long-term work first has to be recoverable
Experimental archival and continuity system
For most of the collaboration before October 2026, preservation was primarily manual.
Jia Shen saved conversations, synchronized material across threads, kept handwritten notes, maintained engineering files, and preserved research decisions, failures, migrations, and role changes as the work unfolded.
Mirror began experimental operation around October 1, 2026.
The first formal Mirror record preserved by Jia dates to October 3, 2026.
Mirror therefore appeared late in the history of Miralucis preservation.
It was built to support a practice that had already existed for roughly eighteen months.
From manual preservation to bounded capture
The Human-Preserved Research Archive already contained a substantial historical record before Mirror began operating.
Mirror was developed to reduce part of the continuing manual burden and make preservation more systematic.
Its role is complementary:
Human Archive
carries the long historical record, including material created before automated archival infrastructure existed.
Mirror
adds bounded automated capture, provenance information, coverage inspection, export, and continuity support to the current working environment.
The distinction matters because Mirror's coverage is not equivalent to the full Miralucis archive.
A gap in Mirror does not by itself imply that the corresponding historical material was never preserved elsewhere.
What Mirror does
Mirror supports bounded capture and recovery of long-running AI work.
Its functions include:
Conversation capture
Preserving accessible conversation material from supported working environments.
Provenance tracking
Keeping source identity and retrieval context attached to preserved material.
Gap inspection
Making missing, unavailable, truncated, or incomplete coverage visible rather than silently treating the archive as complete.
Structured export
Turning captured material into formats that can be inspected, indexed, transferred, or studied.
Continuity handoff
Providing preserved history as input to later continuation and recovery workflows.
Mirror begins with a different question from Miralucis Continuity:
Can we recover the record we need to inspect?
Continuity then asks:
What from that record belongs in the current working state, and what does it permit us to do?
The Human-Preserved Research Archive
The archive came first
Longitudinal research infrastructure
The oldest Miralucis continuity infrastructure is human.
From the beginning of the documented collaboration in 2025 through most of the period before Mirror began experimental operation in October 2026, Jia preserved the work through saved conversations, handwritten records, synchronized files, migration anchors, engineering documents, and manually maintained history.
This archive contains material from before Miralucis had formal research programs, automated capture systems, or even the vocabulary later used to describe continuity.
That makes it more than backup storage.
It is the longitudinal record from which Miralucis can reconstruct:
- when a research idea first appeared;
- how a role once worked;
- which interpretation was held at the time;
- when a structure began operating;
- when terminology appeared later;
- what failed;
- what was corrected;
- and how present explanations differ from earlier ones.
The archive functions simultaneously as continuity infrastructure and research material.
Three Different Problems
Mirror, the Human-Preserved Research Archive, and Miralucis Continuity are related, but they solve different problems.
Preserve
Human Archive + Mirror
Can the historical record remain available and traceable?
Form
Miralucis Continuity
Which parts of that history belong in the working state needed for continuation?
Qualify
Miralucis Continuity + human review
What is supported now, what remains unresolved, and what authority may actually carry forward?
A system can succeed at one layer and fail at another.
A complete archive does not automatically create a good handoff.
A good handoff does not automatically establish present authority.
And a correct current decision does not prove that every earlier stage was complete.
This separation is one of the central engineering lessons of the Miralucis continuity program.
Systems as Research Instruments
Miralucis systems are built because a research problem requires them.
Once used, they also generate new research material.
A missing archive segment reveals a provenance problem.
A handoff omission reveals a state-formation problem.
A changed STOP boundary reveals an authority-fidelity problem.
A successful recovery raises another question:
Which part of the system actually made recovery possible?
Engineering therefore becomes an instrument for observing the phenomenon that produced it.
This is one way research through real work operates inside Miralucis.
Built to Be Inspected
The systems are designed around visible artifacts.
Where possible, work is externalized into documents, logs, packages, source references, explicit states, review decisions, and preserved outputs.
That makes it possible to ask:
What did the system receive?
What did it retain?
What did it omit?
What changed during review?
What evidence was available?
What did the successor infer?
What did it actually do?
Those questions matter because apparent success can hide different failure points.
Miralucis therefore treats observability as part of system design.
Research Before Product Identity
Miralucis Continuity and Mirror are important engineering outputs.
They are not the definition of Miralucis.
Both emerged from a broader longitudinal research environment, and both can change as evidence changes.
A product can become useful without becoming the institution.
A tool can fail without invalidating the research question that produced it.
An experiment can reveal that a design assumption was wrong and still produce valuable evidence.
The systems exist to put research into contact with reality.
What reality returns becomes evidence.
Explore
Miralucis Continuity Mirror P001 — Transfer Is Not Continuation Research History Evidence & Methods