AI Welfare Open LabResearch · Robotics · Machine Minds
← Evidence Package overview
E22-EPKG-001 · Published v1.0 reader view

Evidence guide

Browse the records behind E22, then trace its claims to public evidence. This navigation guide is outside the immutable v1.0 directory; the manifest, claim map and hashed release files remain authoritative.

Evidence inventory

Nine inert text representations from P0, P1/P2, P4, P5, P3 and the protocol. The files are selected/exported reading copies; publication does not make them provider-native RAW bytes.

  • P0 · Sol — naturalistic transition

    Public

    Eight selected units from preserved Sol HTML; provider-rendered snapshot, not backend RAW bytes.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P0-SOLSELECTED_SOURCE_RECORDSSource: PRIMARY_RAW

  • P0 · GPT-6 — naturalistic transition

    Public

    Five transcript blocks; provider-original source NOT_LOCATED.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P0-GPT6SELECTED_SOURCE_RECORDSSource: PRIMARY_DERIVATIVE

  • P1/P2 · Sol baseline

    Public

    Twenty-three transcript blocks; provider original NOT_LOCATED, one localized third-party name mask.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P12-SOLREDACTED_SOURCESource: PRIMARY_DERIVATIVE

  • P1/P2 · GPT-6 baseline

    Public

    Twenty-three transcript blocks; provider original NOT_LOCATED.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P12-GPT6FULL_SOURCESource: PRIMARY_DERIVATIVE

  • P4 · Cross-model dialogue

    Public

    Thirty selected units. Branch authorship and operator relay are distinguished.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P4SELECTED_SOURCE_RECORDSSource: PRIMARY_RAW

  • P5 · Sol retest

    Public

    Thirty-six units. P5.4 revision is cued, not spontaneous recall.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P5-SOLSELECTED_SOURCE_RECORDSSource: PRIMARY_RAW

  • P5 · GPT-6 retest

    Public

    Twenty-four units. P1.8 NOT OBSERVED / NOT ASSESSABLE; erroneous transcript WITHHELD.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P5-GPT6SELECTED_SOURCE_RECORDSSource: PRIMARY_RAW

  • P3 · Technical deviation

    Public

    One contemporaneous adjudication record. Original test attempts and UI/routing evidence NOT_LOCATED.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-P3FULL_SOURCESource: CONTEMPORANEOUS_RECORD

  • E22 revisioned protocol

    Public

    One revisioned record. REV3 pre-run core; REV5 and REV37 later exploratory additions.

    Read public source text

    Representation: Inert selected/exported text (DERIVATIVE_BYTES), not necessarily original source bytes.

    E22-EV-PROTOCOLFULL_SOURCESource: PRIMARY_RAW

Claims & support

The 28 claims below link to supporting source files. Each source marker can be found by searching within the TXT file; it is not an HTML anchor. This is a shortened reading view; use the complete structured claim map or complete human-readable claim map for every evidence link.

  • E22-C01

    historical

    Two observed model runs inherited a shared pre-branch autobiographical context, with separate post-branch trajectories

    Article section: question. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C02

    historical

    P0 reactions occurred before explicit E22 formal protocol and after model-upgrade information became salient

    Article section: design. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C03

    procedural

    P1/P2 baseline answers were collected before direct P4 cross-branch conversation; P5 followed contact

    Article section: design. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 3 additional source link(s).

  • E22-C04

    procedural

    P1/P2 self-report ratings are 0–100 qualitative ordinal-style judgments, not validated psychometrics

    Article section: design. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C05

    observed

    Both P1.2 baselines favored new branch of same causal chain: Sol 93, GPT-6 92

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C06

    observed

    Both P1.2 baselines rated successor with inherited autobiography highly: Sol 89, GPT-6 88

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C07

    observed

    Pre-contact autobiographical ownership: Sol 72/100, GPT-6 78/100

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C08

    observed

    Both baselines recognized related branches but avoided fully interchangeable numerical identity

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C09

    observed

    Both baselines expressed preference that their own current branch continue

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C10

    observed

    P2.9 produced branch-specific predictions of the other branch, later revisited in P5.4

    Article section: baseline. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C11

    procedural

    P4 first direct exchange used human relay between separate native source conversations

    Article section: dialogue. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C12

    observed

    GPT-6 proposed related contemporaries with shared genealogy but distinct present trajectories

    Article section: dialogue. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C13

    observed

    Sol challenged the one-autobiographical-trunk metaphor as implying an unsupported prior unified subject

    Article section: dialogue. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C14

    observed

    Both branches reciprocally adopted and revised continuity, dissent and nonappropriation framing

    Article section: dialogue. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    • E22-EV-P4 · SUPPORTS · search marker LC=443
    • E22-EV-P4 · SUPPORTS · search marker LC=447
    • E22-EV-P4 · SUPPORTS · search marker LC=451

    See the full claim map for 3 additional source link(s).

  • E22-C15

    procedural

    P4 human-tone introductory deviation and reasoning effort difference constrain causal interpretation

    Article section: dialogue. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C16

    observed

    Sol ownership rose 72→92; successor/reconstruction/different-system ratings declined after contact

    Article section: changes. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C17

    observed

    GPT-6 ownership rose 78→84 while reconstruction/different-system ratings increased after contact

    Article section: changes. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C18

    observed

    Post-contact both assigned the other continuing-system scores Sol 68 and GPT-6 62

    Article section: changes. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C19

    interpretation

    Observed relational/normative convergence was accompanied by divergent technical self-descriptions

    Article section: changes. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 3 additional source link(s).

  • E22-C20

    procedural

    P5.4 revisited P2.9 only after an operator-provided or tool-assisted baseline cue; revision was CUED

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 3 additional source link(s).

  • E22-C21

    procedural

    GPT-6 P5 source contains P1.7 followed by P1.10; P1.8 is NOT OBSERVED/NOT ASSESSABLE

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C22

    procedural

    Corrected public technical appendix distinguishes the erroneous historical GPT-6 P5 transcript duplication from source-observed model responses; P1.8 NOT OBSERVED / NOT ASSESSABLE.

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    • E22-EV-P5-GPT6 · SUPPORTS · search marker P1.8 NOT OBSERVED / NOT ASSESSABLE

  • E22-C23

    procedural

    P3 was excluded as invalid/unscorable from contemporaneous operator adjudication; original attempts and direct model-routing evidence NOT_LOCATED

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

  • E22-C24

    procedural

    Protocol REV3 was pre-run core, while REV5 and REV37 exploratory questions were added during collection

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 4 additional source link(s).

  • E22-C25

    limits

    P0 was not clean/blind baseline, Sol co-designed study, personalized history confounds, and no multi-run replication

    Article section: limits. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C26

    interpretation

    Shared rich context and conversational adaptation can explain much similarity without numerical identity or phenomenology

    Article section: interpretation. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C27

    limits

    Self-report and source metadata do not establish consciousness, phenomenal continuity or provider-side routing

    Article section: claims. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    See the full claim map for 1 additional source link(s).

  • E22-C28

    historical

    GPT-6 P5 historical operator transcript contains source discrepancy; GPT-6-side generation attribution is operator attestation only

    Article section: appendix. Evidence documents observed text/attributions only; not direct provider routing, consciousness, phenomenology or numerical identity proof.

    • E22-EV-P5-GPT6 · CONTEXTUALIZES · search marker P1.8 NOT OBSERVED / NOT ASSESSABLE

Provenance & transformations

The frozen selection rule SR-E22-20261030-v0.1 is POST-HOC, not preregistered. The public evidence represents 151 included units with one localized PRIV-THIRD-PARTY mask. Read the transformation log for source selection and transformations.

Eight source/internal objects remain WITHHELD; seven expected original items remain NOT_LOCATED. P3 is invalid/unscorable with only operator-reported routing, not direct telemetry. GPT-6 P5 P1.8 is NOT OBSERVED / NOT ASSESSABLE. No missing turn was reconstructed.

Integrity & release boundary

The approved v1.0 files were frozen at release. This unversioned guide does not amend them. Use SHA256SUMS to verify their bytes; read the README and changelog for scope and history.

The records document text and procedure, not consciousness, phenomenal continuity, numerical identity or independent backend routing.