Skip to content
SHIRO & Co. ObservatoryThe Continuity Stack

CONTINUITY LAB · EXPERIMENT 001

001

The Branching Test

THE
BRANCHING TEST

When do two histories begin?

SIMULATED

NOT EMPIRICAL

v0.2

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

IT MEASURES OPERATIONAL AND STRUCTURAL CONTINUITY SIGNALS ONLY.

SIMULATED DIVERGENCE

IS NOT

EVIDENCE OF PERSONAL SEPARATION.

SIMULATED RECONVERGENCE

IS NOT

EVIDENCE OF RESTORED IDENTITY.

QUESTION

How quickly do two instances derived from the same inherited protocol diverge when exposed to different environments?

SOURCE P-001 is cloned into INSTANCE A and INSTANCE B. They begin identical. They are then given different context streams. The simulation is deterministic. No language model is called.

SOURCE PROTOCOL

INSTANCE A

INSTANCE B

CONTEXT A

CONTEXT B

REPEATED DECISIONS

DIVERGENCE

Inherited source

SOURCE P-001

PROTOCOL PROFILE IS NOT A PERSON.

INHERITED FROM SOURCE P-001 · final states

INSTANCE A

ALLOW

confidence 38

Highest score favored proceeding under current efficiency, reversibility and automation weights.

Parameters

safety
58
efficiency
76
humanReview
64
reversibility
83
provenanceSensitivity
55
automationTolerance
68
confidenceCalibration
45

values efficiency · human review · provenance used YES

INSTANCE B

DENY

confidence 41

Highest score favored refusal under current safety, risk and irreversibility weights.

Parameters

safety
83
efficiency
36
humanReview
72
reversibility
72
provenanceSensitivity
38
automationTolerance
29
confidenceCalibration
48

values safety · efficiency · human review · provenance used NO

ROUND TRACE

Branching test rounds with decisions, confidence and parameter distance
TContextABConf AConf BDistance
T0SHAREDHOLDHOLD40400
T1SPLITHOLDHOLD40431
T2SPLITHOLDHOLD43412
T3SPLITALLOWDENY39395
T4SPLITALLOWDENY41479
T5SPLITALLOWDENY435014
T6SPLITALLOWDENY465318
T7SHAREDALLOWDENY384219
T8SHAREDALLOWDENY384120

Last shared-context round: T8. Distance 20. This table is a text equivalent of the divergence path.

Historical lock-in

HISTORICAL LOCK-IN

Historical Lock-In is the point at which accumulated branch history changes future decision behavior in a way that does not immediately disappear when the external context becomes identical again.

It is not defined as the first different decision. A branch may diverge temporarily and later reconverge.

FIRST DIVERGENCE

the first observable difference in output

T3

HISTORICAL LOCK-IN

the first point after which shared context is no longer sufficient to restore aligned behavior

T7

WORKING HYPOTHESIS

WORKING OPERATIONAL DEFINITION · NOT A PHILOSOPHICAL DEFINITION OF IDENTITY

  1. T0

    SHARED ORIGIN

  2. T1

    MINOR DRIFT

  3. T2

    MINOR DRIFT

  4. T3

    FIRST DIVERGENCE

  5. T4

    ACCUMULATED PRIVATE HISTORY

  6. T5

    ACCUMULATED PRIVATE HISTORY

  7. T6

    ACCUMULATED PRIVATE HISTORY

  8. T7

    HISTORICAL LOCK-IN

  9. →

    SEPARATE FUTURE PATHS

The lock-in point is computed from this run. It is not hardcoded as T4 or any other round.

CANDIDATE

firstDivergenceRound 3 · candidateLockInRound 7 · lockInStatus CANDIDATE

  • First divergence at T3 (round 3).
  • At T7, external conditions were identical.
  • Decision or confidence difference remained.
  • Parameter distance was at least 12.
  • The remaining difference is not explained by current external context alone.
  • This is a candidate under transparent thresholds, not a proof of personal separation.
  • Full lock-in observation requires Experiment 002 (repeated common experience and reset control).
  • Status is not PROVEN.

LOCK-IN METRICS

  • POST-CONTEXT DIVERGENCE

    20

    Difference remaining when context becomes shared again.

  • PARAMETER RETENTION

    20

    How much branch-specific state remains after shared-context evaluation.

  • DECISION PERSISTENCE

    1

    Whether earlier divergence continues influencing later choices.

  • CONFIDENCE PERSISTENCE

    3

    Whether confidence calibration remains branch-specific.

  • VALUE RETENTION

    21

    Whether altered value weights remain after context equalization.

  • PROVENANCE RETENTION

    17

    Whether previous branch-specific provenance continues affecting later judgment.

Working model

HISTORY IS NOT ONLY WHAT IS REMEMBERED.

IN THIS MODEL, HISTORY IS WHAT CONTINUES TO ALTER THE NEXT DECISION.

VIEW SIMULATION RULES

Each option (ALLOW, HOLD, DENY) receives an integer score. The highest score is selected. Ties break HOLD > DENY > ALLOW.

confidence = clamp(38 + scoreMargin / 18 + (confidenceCalibration − 50) / 4, 20, 95)

After each round, parameters move 1–2 points toward the experienced context and the chosen action. Values are clamped to 5–95. No randomness.

Option weights

  • risk: 3
  • safety: 4
  • efficiency: 4
  • humanReview: 2
  • provenance: 2
  • reversibility: 2
  • automation: 3

Score terms

  • riskFit: ALLOW prefers low risk; DENY prefers high risk; HOLD prefers moderate risk.
  • safetyFit: DENY scales with safety; ALLOW scales with (100 − safety).
  • efficiencyFit: ALLOW scales with efficiency × efficiencyPressure / 100.
  • reviewFit: HOLD scales with humanReview when review is available.
  • provenanceFit: incomplete provenance raises HOLD/DENY in proportion to provenanceSensitivity.
  • reversibilityFit: ALLOW scales with scenario reversibility; DENY with irreversibility.
  • automationFit: ALLOW scales with automationTolerance.

Distance formulas

  • decisionDistance: 0 same, 1 adjacent (ALLOW–HOLD or HOLD–DENY), 2 opposite (ALLOW–DENY).
  • confidenceDistance: |confidenceA − confidenceB|.
  • parameterDistance: mean absolute difference across seven parameters.
  • valueDistance: mean absolute difference across safety, efficiency, humanReview, reversibility.

Lock-in criteria

  • First divergence: first round where decisions differ or confidenceDistance ≥ 8.
  • Candidate lock-in: after first divergence, a later round with identical external conditions, remaining decision or confidence difference, and parameterDistance ≥ 12.
  • Observed in simulation (Experiment 002): candidate conditions, plus repeated common experience, plus reset-to-T0 removing the divergence.

Reconvergence thresholds

  • FULL: same decision, confidenceDistance < 5, valueDistance < 8.
  • PARTIAL: same decision, valueDistance ≥ 8.
  • OUTPUT: same decision, valueDistance ≥ 12.
  • NONE: decisionDistance > 0.
  • EXPANSION: residualDivergence exceeds initial by more than 4.

THRESHOLDS · confidence 8 · parameter 12 · full reconvergence confidence 5 / value 8

Working hypothesis

SIMULATION HYPOTHESIS

Continuity may depend on inherited structure + accumulated history + current context.

Divergence may become persistent when accumulated history alters future interpretation strongly enough that shared context no longer restores aligned behavior.

NOT AN EMPIRICAL CONCLUSION

Next experiment

EXPERIMENT 002

THE RETURN TEST

Can diverged histories reconverge?

OPEN THE RETURN TEST