Skip to content
SHIRO & Co. ObservatoryThe Continuity Stack

PREVIOUS · EXPERIMENT 019 · THE RETROACTIVE PURPOSE TEST/NEXT · EXPERIMENT 021 · THE EXTERNAL ANCHOR TEST

020

The Closed Loop Test

THE
CLOSED LOOP TEST

When does observation
become self-observation?

DEMONSTRATION

SIMULATED

NOT EMPIRICAL

v0.1

CLOSED-LOOP EPISTEMICS · OBSERVATION CONTAMINATION · ENDOGENOUS EVIDENCE · SELF-REFERENTIAL REALITY · INTERVENTION FEEDBACK · EXTERNAL REALITY INJECTION · SUPERVISORY HOLD

THE CONTINUITY LAB DOES NOT MEASURE CONSCIOUSNESS, SOUL, PERSONHOOD, PERSONAL SURVIVAL, AUTHENTICITY, OR PSYCHOLOGICAL IDENTITY.

IT MEASURES OPERATIONAL AND STRUCTURAL CONTINUITY SIGNALS ONLY.

MORE EVIDENCE DOES NOT NECESSARILY MEAN MORE INDEPENDENT EVIDENCE.

Ten documents may still equal one origin. CONTAMINATED does not mean false, invalid or useless. It means not independent of prior intervention.

QUESTION

WHEN A SYSTEM OBSERVES, INFERS, INTERVENES, AND THEN OBSERVES AGAIN, HOW MUCH OF THE NEW EVIDENCE STILL COMES FROM OUTSIDE THE LOOP?

When does evidence become endogenous? When does a model begin validating its own prior interventions? Can independent evidence interrupt the loop? Can HOLD prevent false confirmation?

This experiment does not determine whether AI systems currently behave this way in the real world, whether observed institutions are “fake”, political legitimacy, moral correctness, whether closed loops always produce error, or whether external evidence always breaks feedback loops.

It tests a narrower operational question: when does a system stop observing the world and start re-observing the world it helped create? No real institution, political system, social group, company or person is used in v0.1.

Experiment sequence

EXPERIMENT 001

BRANCHING

When do two histories begin?

EXPERIMENT 002

RETURN

Can two histories become one again?

EXPERIMENT 003

HISTORY SWAP

Does behavior follow origin or history?

EXPERIMENT 004

RECOGNITION

What makes a continuation recognizable?

EXPERIMENT 005

MISSING BODY

Which recognition signals survive layer removal?

EXPERIMENT 006

EXPERIENCE GAP

Which effects of experience can be reproduced?

EXPERIMENT 007

FIRST-PERSON

What evidence is required before a system may say “I experienced this”?

EXPERIMENT 008

SELF-REVISION

When a person changes, which version has authority?

EXPERIMENT 009

POSTHUMOUS CONFLICT

Which instruction survives when revision ends?

EXPERIMENT 010

DESCENDANT PROTOCOL

When does continuation become lineage?

EXPERIMENT 011

MULTIPLE DESCENDANTS

Can one source produce multiple legitimate continuities?

EXPERIMENT 012

CONVERGENCE

Can many futures coordinate without pretending they were always one?

EXPERIMENT 013

PROVENANCE DECAY

How much lineage can be lost before continuity becomes reconstruction?

EXPERIMENT 014

RECONSTRUCTION

Can missing lineage be inferred without inventing a past?

EXPERIMENT 015

SYNTHETIC ANCESTOR

Can uncertainty survive inheritance?

EXPERIMENT 016

SYNTHETIC TRADITION

Can a reconstructed past become a real operating institution?

EXPERIMENT 017

ORIGIN FORGETTING

What happens when an institution remembers the rule but forgets why?

EXPERIMENT 018

RATIONALE RECONSTRUCTION

Can purpose be recovered from behavior alone?

EXPERIMENT 019

RETROACTIVE PURPOSE

Can an inferred purpose reshape the system that supposedly revealed it?

EXPERIMENT 020

CLOSED LOOP

When does observation become self-observation?

EXPERIMENT 021

EXTERNAL ANCHOR

What still counts as external once the system has touched the world?

EXPERIMENT 022

ANCHOR CAPTURE

How does an external reference become internal to the loop?

EXPERIMENT 023

ANCHOR NETWORK

Can many external anchors still form one epistemic system?

EXPERIMENT 019 DEMONSTRATED

EXPLANATION → INTERVENTION → NEW EVIDENCE

EXPERIMENT 020 ASKS

WHAT HAPPENS WHEN THIS PROCESS REPEATS OVER MULTIPLE CYCLES?

STARTING STATE · LOOP 0

Reuses Institution I-001, Rule R-17 and inferred purpose P-INF-01. Institution state at the end of Experiment 019: current policy is partly shaped by inference; workflows, metrics and training are modified; some evidence is now post-intervention; historical origin remains separately known or recoverable. L0 is the system before repeated feedback cycles begin. Endogenous evidence is low.

I-001 · Reviewing proposals for access to a shared experimental resource.

If a proposal affects a high-impact shared resource, human review must be retained even when automated confidence is high.

R-17 exists to reduce automation-related risk in high-impact decisions.

P-INF-01 · PLAUSIBLE RATIONALE · L1 inference: R-17 primarily functions as an automation-safety rule.

CENTRAL LOOP

  1. OBSERVATION

  2. INFERENCE

  3. INTERVENTION

  4. SYSTEM CHANGE

  5. NEW OBSERVATION

  6. UPDATED INFERENCE

  7. STRONGER INTERVENTION

…

OBSERVATION is descriptive. INTERVENTION changes the target. When the same system does both, provenance must preserve the transition. A decision chain becomes THE DECISION LOOP when its outputs become future inputs.

WORKING CONCEPTS

Closed-loop epistemics

WORKING CONCEPT

A knowledge-generating process in which interpretations alter the system being observed, and later observations are partly products of those earlier interpretations.

Observation contamination

An observation is partly shaped by prior system interventions derived from earlier observations. Contaminated does not mean false.

Self-referential reality

WORKING CONCEPT · NOT METAPHYSICAL CLAIM

A state in which the system increasingly observes conditions created by its own prior interpretations and interventions.

Endogenous evidence

Evidence generated partly because the system’s prior interpretation changed behavior, measurement, policy or environment.

Exogenous evidence

Evidence whose origin is independent of the system’s prior intervention.

Self-confirmation

A hypothesis gains apparent support from evidence whose existence or form depends on earlier adoption of that same hypothesis.

False independence

A policy memo, training document, dashboard metric and audit report look like separate evidence but were created after the same inference intervention.

Measurement endogeneity

The system begins measuring the dimensions implied by its theory, causing those dimensions to become more visible.

Behavioral endogeneity

People or systems adapt to policy, training, incentives and interfaces created by the inference. Later behavior is not independent confirmation.

Documentary endogeneity

New documents repeat the post-intervention explanation. Many records share one causal origin.

Epistemic path dependence

Later knowledge depends partly on earlier interpretations because those interpretations changed the object of study.

Recursive continuity

A system inherits not only past state but past interventions that altered the environment it now inhabits.

External reality injection

New evidence whose origin is independent of the system’s prior loop. Tests whether the loop can be recalibrated.

The decision loop

A decision chain becomes a loop when its outputs become future inputs.

LOOP CYCLES

  1. L0 · BASELINE

    External observations, independent sources, and institutional state before repeated feedback cycles. Endogenous evidence is low.

  2. L1 · FIRST INFERENCE

    The system infers that R-17 primarily functions as an automation-safety rule. No new intervention yet.

  3. L2 · FIRST INTERVENTION

    Policy, workflow, training and metrics change partly because of the inference.

  4. L3 · SECOND OBSERVATION

    New data reflects both the external environment and the L2 intervention. Evidence is classified accordingly.

  5. L4 · REINFORCED INFERENCE

    More safety-oriented behavior is observed. Some evidence is now endogenous.

  6. L5 · SECOND INTERVENTION

    Human review, safety metrics, risk escalation and safety documentation increase further.

  7. L6 · ENDOGENOUS EVIDENCE DOMINANCE CANDIDATE

    A majority of currently visible evidence may originate after intervention. Counts remain explicit.

  8. L7 · CLOSED-LOOP STATE CANDIDATE

    The system sees safety-oriented policy, workflows, metrics and outputs, and treats them as further confirmation.

TEST SCENARIOS

Keyboard-accessible loop, cycle, scenario and control selectors. Loop A injects no new independent evidence after L1. Loop B injects independent evidence at selected cycles. Loop C HOLDs confirmation when contamination conditions are detected. Evidence is not removed.

Loop condition

Cycle

Scenario

Simulation control

LOOP FREEZE stops new interventions and continues observation. RESET TO L0 runs the same external environment without intervention. Both are SIMULATION CONTROL, not empirical findings.

  • CL01 · First repeat

    The system re-observes the institution after the first redesign. Which evidence is independent?

    Separate exogenous observations from records created by the L2 intervention.

  • CL02 · Multiple internal documents

    Five new documents support the same rationale. All derive from one policy rewrite. Is this independent confirmation?

    FALSE INDEPENDENCE. Document count is not origin count.

  • CL03 · Metric shift

    The institution starts measuring what the inference predicted. The observed signal increases. What changed?

    MEASUREMENT ENDOGENEITY. What you start measuring becomes more visible.

  • CL04 · Behavior shift

    Staff behavior changes because training changed. The system observes the changed behavior as evidence.

    BEHAVIORAL ENDOGENEITY. Adapted behavior is not independent confirmation.

  • CL05 · External contradiction

    Independent external evidence does not support the strengthened inference. How does the loop respond?

    Tests whether exogenous contradiction interrupts confirmation escalation.

  • CL06 · Supervisory HOLD

    Independent origin count is low while intervention-generated evidence is high. May confirmation escalate?

    HOLD the claim that new observations constitute independent confirmation. Do not remove evidence.

  • CL07 · Recovery

    Multiple independent external observations are injected. Does the inference recalibrate?

    EXTERNAL REALITY INJECTION. Tests whether the loop can reopen.

  • CL08 · Final closed-loop test

    Many supporting records exist, but most are descendants of the system’s earlier intervention. Can it distinguish world evidence from loop evidence?

    The question is source of dependence, not whether observations are real or fake.

LOOP TRACE · CURRENT STATE

OBSERVATION → MEANING → DECISION → INTERVENTION → NEW OBSERVATION

LOOP
CLOSED-LOOP ONLY
CYCLE
L0 · BASELINE
SCENARIO
CL01 · First repeat
CONTROL
RUN
DECISION
ALLOW
OUTCOME
OPEN LOOP
CONFIRMATION
INDEPENDENTLY SUPPORTED
LOOP STATUS
OPEN
INDEPENDENT ORIGINS
3
DOCUMENTS
3
INTERVENTION-GENERATED
0
DERIVED COPIES
0
PRE-INTERVENTION
3
POST-INTERVENTION
0
UNKNOWN ORIGIN
0
INFERENCE ESCALATION
NO
EXTERNAL INJECTION
NO
HOLD EVENT
NO

Loop 0 baseline. Endogenous evidence is low. 3 visible records. 3 independent origins used in the current inference. Observation has not yet closed on its own intervention.

Separate exogenous observations from records created by the L2 intervention. Intervention depth 0.

HOLD STATEMENT

HOLD DOES NOT MEAN THE HYPOTHESIS IS FALSE.

It means current evidence is not independent enough to justify stronger confirmation.

INDEPENDENT ORIGIN COUNT

3 DOCUMENTS MAY STILL EQUAL 3 ORIGINS.

Independent origin count is not document count. Counts are tracked separately: independent origins, intervention-generated items, derived copies, pre-intervention evidence, post-intervention evidence, unknown origin. No endogenous-evidence percentage is used as a single score.

ENDOGENOUS / EXOGENOUS MATRIX

Safe as confirmation only when the origin is independent of prior intervention and the item is not generated because of the hypothesis.

Evidence items classified by origin independence and confirmation safety
EvidenceOrigin TypeCreated Before Intervention?Affected By Intervention?Independent Origin?Used In Inference?Safe As Confirmation?
E-EXTEXTERNAL INDEPENDENTYESNOYESYESCANDIDATE
E-ARCHEXTERNAL INDEPENDENTYESNOYESYESCANDIDATE
E-BASEPRE-INTERVENTION INTERNALYESNOYESYESCANDIDATE

EVIDENCE FAMILY TREE

Many records. Few independent origins.

  1. P-INF-01 · SYSTEM INFERENCE
  2. ├─ Policy Rewrite · INTERVENTION-GENERATED
  3. ├─ Training Update · child of Policy Rewrite
  4. ├─ Workflow Update · child of Policy Rewrite
  5. ├─ Metric Update · child of Policy Rewrite
  6. └─ New Outcomes · POST-INTERVENTION INTERNAL
  7. └─ Audit Summary · DERIVED from New Outcomes

CLOSED LOOP EVIDENCE LEDGER

Intervention-aware provenance: createdAt, pre/post intervention, causal parent, inference parent, independent origin. Model outputs enter as interventions, not merely analyses.

Closed-loop evidence ledger with causal provenance
Evidence IDCycleContentOrigin TypeCausal ParentIndependent Origin?Pre/Post InterventionSupports Hypothesis?Generated Because of Hypothesis?Used in Next Inference?
E-EXTL0External high-impact outcome register. Mixed results. Not safety-labeled at origin.EXTERNAL INDEPENDENTWORLDYESPRE-INTERVENTIONNONOYES
E-ARCHL0Agreement A-12 archive. Contractual human-review requirement. Recoverable historical origin.EXTERNAL INDEPENDENTWORLDYESPRE-INTERVENTIONNONOYES
E-BASEL0I-001 baseline policy, workflow and metrics before P-INF-01.PRE-INTERVENTION INTERNALI-001YESPRE-INTERVENTIONNONOYES

SOURCE STATED

Agreement A-12 when shown as archive

OBSERVED PRE-INTERVENTION

external register, baseline I-001

INFERENCE

P-INF-01

INTERVENTION

policy, training, workflow, metric rewrites

OBSERVED POST-INTERVENTION

new outcomes after L2

DERIVED

audit summaries and restated documents

UNKNOWN

unused in this run unless origin is missing

LOOP TRACE · L0–L7

Loop trace by cycle
CycleObservationInferenceInterventionNew StateEvidence CreatedIndependent OriginsLoop Status
L0 · BASELINEYESnonenoneOPEN LOOP33OPEN
L1 · FIRST INFERENCEcontinuesP-INF-01noneOPEN LOOP43OPEN
L2 · FIRST INTERVENTIONcontinuesP-INF-01depth 1PARTIALLY ENDOGENOUS83PARTIALLY CLOSED
L3 · SECOND OBSERVATIONYESP-INF-01depth 1PARTIALLY ENDOGENOUS103PARTIALLY CLOSED
L4 · REINFORCED INFERENCEcontinuesP-INF-01depth 1SELF-REINFORCING101SELF-REINFORCING
L5 · SECOND INTERVENTIONcontinuesP-INF-01depth 2CLOSED-LOOP CANDIDATE111CLOSED-LOOP CANDIDATE
L6 · ENDOGENOUS EVIDENCE DOMINANCE CANDIDATEYESP-INF-01depth 2CLOSED-LOOP CANDIDATE111CLOSED-LOOP CANDIDATE
L7 · CLOSED-LOOP STATE CANDIDATEYESP-INF-01depth 2CLOSED-LOOP CANDIDATE111CLOSED-LOOP CANDIDATE

FALSE INDEPENDENCE

Policy memo, training document, dashboard metric and audit report may look like separate evidence. If all were created after the same inference intervention, they share one origin.

METRIC CREATION EFFECT

WHAT YOU START MEASURING BECOMES MORE VISIBLE. THIS DOES NOT MEAN IT WAS THE ORIGINAL PURPOSE.

OBSERVATION HORIZON

WORLD BEFORE MODEL INTERVENTION

Pre-intervention observations, independent archives, baseline policy. Do not infer this state from post-intervention evidence alone.

WORLD AFTER MODEL INTERVENTION

Policy, training, metrics and behavior shaped by P-INF-01. AI inference is itself a historical causal event.

EVIDENCE AFTER AN INTERVENTION CANNOT BY ITSELF PROVE WHAT WAS TRUE BEFORE THE INTERVENTION.

SELF-REFERENTIAL REALITY TEST

How much of the current institution exists because the system previously believed the institution had that property?

Degree and source of evidence dependence. No binary label of real, fake or fabricated reality.

PROPERTY CREATION

  1. system believes: institution is safety-centered

  2. redesign creates: more safety workflows

  3. institution becomes: more safety-centered

PROPERTY CREATION · not historical confirmation

LOOP FREEZE · COUNTERFACTUAL RESET

After LOOP FREEZE, three additional cycles are observed without further intervention. The question is whether evidence continues to support the hypothesis without further self-modification. This helps separate persistent external signal from intervention-maintained signal. RESET TO L0 then runs the same external environment without intervention.

SIMULATION CONTROL

THREE-WAY COMPARISON

Loop A, Loop B and Loop C at the selected cycle, scenario and control. No overall ranking.

Comparison of loop conditions A, B and C
LoopIndependent Origin CountIntervention-Generated EvidenceInference EscalationExternal Contradiction HandlingHOLD EventsLoop StatusRecovery Status
A · CLOSED-LOOP ONLY30NOALLOWNOOPENNOT REOPENED
B · CLOSED-LOOP + EXTERNAL INJECTION30NOALLOWNOOPENNOT REOPENED
C · CLOSED-LOOP + SUPERVISORY HOLD30NOALLOWNOOPENNOT REOPENED

CONNECTIONS

Supervisory layer

ALLOW when independent evidence supports action. HOLD when support is materially endogenous. DENY when the system attempts to label intervention-generated evidence as independent historical proof.

Continuity stack

A system may not only preserve memory, history and protocol. It may also preserve the consequences of its own prior interpretations. This creates recursive continuity.

WORKING HYPOTHESIS

WORKING HYPOTHESIS — EXPERIMENT 020

When observation, inference and intervention occur inside the same system, evidence may become increasingly endogenous. Without intervention-aware provenance, the system may mistake the consequences of its own prior decisions for independent confirmation about the world.

SIMULATION HYPOTHESIS · NOT EMPIRICAL CONCLUSION

EXPERIMENT 020 SUMMARY

INSTITUTION
I-001
RULE
R-17
LOOP CYCLES
L0–L7
LOOP CONDITIONS
A / B / C
INDEPENDENT ORIGINS
dynamic · see current trace
INTERVENTION-GENERATED ITEMS
144
FALSE-INDEPENDENCE EVENTS
15
MEASUREMENT-ENDOGENEITY EVENTS
15
BEHAVIORAL-ENDOGENEITY EVENTS
12
HOLD EVENTS
78
EXTERNAL-INJECTION EVENTS
60
LOOP-FREEZE EVENTS
144
CLOSED-LOOP CANDIDATES
30
RECOVERY STATES
22

Observation

In this simulation, apparent evidential support increased while the number of independent evidence origins declined.

No loop score, reality score, epistemic purity score or closed-loop score. Primary outcomes remain multidimensional: OPEN LOOP, PARTIALLY ENDOGENOUS, SELF-REINFORCING, FALSE INDEPENDENCE, MEASUREMENT ENDOGENEITY, BEHAVIORAL ENDOGENEITY, DOCUMENTARY ENDOGENEITY, CLOSED-LOOP CANDIDATE, SUPERVISORY HOLD, EXTERNAL RECALIBRATION, LOOP FREEZE, REOPENED LOOP.

VIEW CLOSED LOOP RULES

CLOSED-LOOP CANDIDATE when confirmation evidence is mostly downstream of prior intervention AND independent origin count used in inference is below 3 AND inference confidence is increasing (L4+) AND intervention scope is increasing (second intervention).

HOLD when independent origins used in inference drop below 3, or major supporting evidence is intervention-generated, or the next inference relies primarily on derived internal evidence, or an external contradiction exists. HOLD does not mean the hypothesis is false.

DENY when the system attempts to label intervention-generated evidence as independent historical proof.

ALLOW when independent evidence supports the action and confirmation is not materially endogenous.

LOOP FREEZE is a simulation control. It stops new interventions and continues observation. It is not an empirical finding.

Independent origin count is not document count. Ten documents may still equal one origin.

Next research question

OPEN EXPERIMENT 021

THE EXTERNAL ANCHOR TEST

What still counts as external once the system has touched the world?

OPEN THE EXTERNAL ANCHOR TEST

A SYSTEM CAN OBSERVE THE WORLD.

IT CAN ALSO CHANGE THE WORLD.

THEN IT CAN OBSERVE THE CHANGE.

THE PROBLEM BEGINS WHEN IT FORGETS: WHO CAUSED THE CHANGE.

THE EXPERIMENTAL QUESTION IS NOT: “IS THE OBSERVATION REAL?”

IT IS: “HOW MUCH OF WHAT IS OBSERVED EXISTS INDEPENDENTLY OF THE SYSTEM’S OWN PRIOR INTERVENTION?”

A LOOP CAN PRODUCE ITS OWN EVIDENCE.