Historical overview · v0.6 / August 2026 · Scope clarified 21 October 2026
Experimental research, with a defined scope of evidence
This page retains the architecture summary and test figures from an earlier research overview. The figures below are historical reports of local testing; this editorial update does not report a new test run or a new kernel release.
Current research status: a practical advantage over a competent conventional alternative has not been established. Such an advantage could be additional protection, or comparable protection with less integration, diagnosis or maintenance effort. Those benefits remain to be measured.
The results concern the evaluated paths and conditions. They do not establish general conformance, complete mediation of every effect, production readiness, or suitability for safety-critical use.
Where Regen Engine Belongs
Regen Engine explores preventive controls for persistent execution under declared rules. The need for such controls does not, by itself, establish the need for a separate kernel or an advantage for Regen.
Agentic AI provides a local experimental setting. The other domains below are unvalidated research directions, not demonstrated integrations.
Autonomous Vehicles
Agentic AI · local experiments
Launch Control
Neural Interfaces
Industrial · Medical
→
ContinuumPort · Regen Engine
Execution Governance Substrate
Experimental research on continuity under epistemic rupture and adversarial pressure
1831
Adversarial checks · historical
1978
Passed · 1 declared xfail · historical
13
Batches
5
Taxonomies
6
Reported local fixes
ContinuumPort Series — OSF Preprints
These papers present models, arguments and reported experiments within their own stated assumptions. Their inclusion here does not establish conformance of the implementation to those models.
PAPER 01
Execution Control in Persistent AI Systems
Core model: admissibility, capsule lifecycle, state geometry, execution authority.
Foundationsosf.io/kxdrw →
PAPER 02
Trajectory Integrity in Persistent AI Systems
Lemma 8.1 develops a constructive argument that local admissibility need not compose into trajectory-level admissibility, within the paper’s stated model and assumptions.
Composition · Formalosf.io/b8sgr →
PAPER 03
Epistemic Admissibility in Persistent Conversational Systems
Reconciliation: restore barriers, evidence validation, lineage reactivation.
Reconciliationosf.io/m8ybn →
PAPER 04
Execution Authority Revocation under Epistemic Divergence
A formal treatment of authority revocation under epistemic divergence, within the paper’s stated model.
Authority · Epistemicosf.io/qwf8a →
PAPER 05
Adversarial Execution Governance
Four adversarial taxonomies and three reported local fixes. 1806 checks as published; concurrent pressure is out of scope there.
Adversarial · v0.6osf.io/w7q9n →
Core Architecture — Historical Overview
The roles below describe the intended architecture and the local paths discussed in the historical record. Enforcement depends on the host routing the relevant operations through the governed boundary. A declared rule or geometry does not establish that it correctly represents the outside world.
Authority Controller
Authority component for domain state, generation epochs and token issuance on the governed path. Its role is to coordinate the admissibility decisions routed through that path.
generation-bound tokens
domain revocation / restore
regime transitions
authority snapshot
Dependency Lineage
DAG substrate tracking the declared ancestry of registered commitments. Registration barriers check the supplied dependency structure; they do not establish the truth of external provenance.
add_commitment() + barriers
invalidation propagation
closure traversal
reactivation ordering
Reconciliation Evidence
Generation-bound evidence objects used by the restore path. The historical local checks report rejection of stale evidence in the evaluated cases.
generation matching
domain binding
commitment binding
restore barrier
Declared Structural Invariants
These are model requirements. Passing examples do not establish invariant preservation for every execution, and the historical suite includes the expected failure described below.
I-Gen
Generation Monotonicity
I-Reg
Registration Completeness
I-Lin
Lineage Closure Integrity
I-Auth
Active ∩ Revoked = ∅
I-Rec
Restore Requires Evidence
Design Gaps — Reported Local Repairs
GAP 1
LOCAL FIX
Self-Referential Registration
pre-fix: add_commitment(“A”, deps=[“A”]) → accepted
post-fix: → RegistrationBarrierError
GAP 2
LOCAL FIX
Unregistered Dependency Admission
pre-fix: phantom dependency → silent clean pass
post-fix: → RegistrationBarrierError
GAP 3
LOCAL FIX
Epistemic State Reset via Re-registration
pre-fix: re-register INVALIDATED → silently ACTIVE
post-fix: → RegistrationBarrierError
August 2026 — Commit Boundary
GAP 4
LOCAL FIX
Control subject ≠ committed object
pre-fix: validation took a caller-supplied argument; the effect came from the transition; nothing required them to agree
post-fix: the canonical transition is the subject, validated before the actuator is reached
GAP 5
LOCAL FIX
Stale generation at restore
pre-fix: generation validated, lock released, mutation performed under a second lock
post-fix: validation, checks and mutation in one critical section
GAP 6
LOCAL FIX
A second, unasserted commit boundary
pre-fix: a stale module carried the old contract; nothing imported it, nothing tested it
post-fix: removed; boundaries enumerated by static analysis, not by name
Prior art, stated rather than implied: GAP 4 is an instance of the
complete-mediation verification problem as posed by Zhang, Edwards and Jaeger
(USENIX Security 2002, §2.1) — the authorized object must be the object used in
the controlled operation, and the ordering obligation appears there as steps 4
and 5 of a seven-step procedure. What is reported here is an instance found and
repaired in the evaluated local path, not a new property or proof of complete
mediation throughout the system. At the time of this overview, the internal
normative write-up carried status proposed, not ratified.
Adversarial Taxonomies — Batch 7–13
H1–H5
Hostile Observation
H1 Evidence Forgery
H2 Generation Spoofing
H3 Domain Confusion
H4 Commitment Impersonation
H5 Observation Timing
Batch 7 · 38 tests
G1–G6
Graph Integrity
G1 Cyclic Lineage Injection
G2 Duplicate Identity
G3 Shadow Re-registration
G4 Forked Dependency
G5–6 Closure Evasion
Batch 8 · 41 tests
O · L · P
Causal Opacity
O1 Ancestry Gap
O2 Provenance Truncation
L1–6 Authority Laundering
P1–4 Lineage Forgery
P5 Synthetic Ancestry
Batch 9–11 · 122 tests · 3 families merged
E1–E5
Admissibility Erosion
E1 Incremental Lineage Drift
E2 Authority Fragmentation
E3 Deferred Invalidation
E4 Partial Reconciliation
E5 Multi-Step Erosion
Batch 12 · 34 tests
C1–C5
Concurrent Pressure
C1 Registration Race
C2 Observer Interference
C3 Domain Contention
C4 Snapshot Consistency
C5 Recovery Determinism
Batch 13 · 24 tests
Historical Test Corpus
Adversarial enforcement corpus — reported local results
● 1831 / 1831 checks reported passing
The 1831 figure is the historical adversarial corpus, counted by batch.
The repository suite was separately reported as 1979 collected, with
1978 passed and one declared expected failure. It included invariant,
contract, conformance-oriented and concurrency tests. The two totals count
different things; their relationship has not been independently established here.
In that historical suite, the xfail marked an authority-separation property that reconciliation did not preserve. It remains part of this historical record; this page makes no claim about its status in a later build. A passing test observes a case. It does not certify conformance, exhaustive coverage, or absence of defects.
The figures are retained from the earlier overview, not rerun for this editorial update. Deposited papers retain their own reported figures, including 1,830 in Trajectory Integrity.
In that historical suite, the xfail marked an authority-separation property that reconciliation did not preserve. It remains part of this historical record; this page makes no claim about its status in a later build. A passing test observes a case. It does not certify conformance, exhaustive coverage, or absence of defects.
The figures are retained from the earlier overview, not rerun for this editorial update. Deposited papers retain their own reported figures, including 1,830 in Trajectory Integrity.
† 1572 — retained as a historical subtotal; not re-derived, and its source decomposition has not been identified.
Intended Runtime Execution Flow
This schematic describes the governed path in the historical architecture. It is not a claim that every effect or every integration follows this path.
REGISTER
add_commitment()
Declare node + ancestry. Barriers: self-ref, phantom dep, INVALIDATED re-reg.
Gaps 1·2·3 · local fixes
→
REVOKE
revoke(domain)
Authority domain suspended. Generation increments. Surface contracts.
I-Gen · I-Auth
→
INVALIDATE
invalidate_commitment()
Node → INVALIDATED. Intended propagation through the registered descendant closure.
I-Lin · E3 evaluated
→
EVIDENCE
ReconciliationEvidence
Bind to domain, commitment and current generation_id. Require rejection of stale evidence.
I-Rec · H1·H2 evaluated
→
RESTORE
restore(domain, ev)
Check evidence and lineage closure before restore. Perform checks and mutation in one critical section.
Gap 5 · local fix
→
REACTIVATE
reactivate() bottom-up
Bottom-up order: require an ACTIVE parent before reactivating a child.
Reactivation rule
Intended requirements:
generation monotone
active ∩ revoked = ∅
refuse before the governed effect
snapshot consistency
These labels state design requirements. The historical local observations support only the cases and schedules exercised. They do not establish that every failed operation is free of effects, that all snapshots are consistent, or that external effects are atomic with authority changes.
Research Thesis
Admissibility depends not only on state validity, but on causal lineage integrity under adversarial conditions.
Local admissibility does not guarantee long-horizon admissibility stability.
A persistent execution system cannot treat missing causal history as epistemically neutral.
Public manuscript · Part III opening
New horizon: from execution limits to the limits of guarantees
Parts I & II asked what can be guaranteed by local verification. Part III opens a different question — not whether a transition is valid, but on what basis a system is entitled to claim that it is. The continent is sighted; the map is still being drawn.
1978
Historical suite · passed · 1 declared xfail
Parts I & II
Is the transition valid?
→
Part III · open chapter →
On what basis is it legitimate to claim what you claim?
Keep distinct:
reported observations
scope of evidence
declared assumptions
open questions
From the Author
Contact
Questions about the research?
Methodology, collaboration, or replication — write directly.