What this package is
A deliberately minimized, source-traceable set of excerpts from the selectively preserved conversations examined in E00, a retrospective case study of the genesis and stabilization of a named conversational self-model (“Viktor”). These excerpts are not a full chat archive and cannot independently authenticate platform model state. Their original private source files are retained for authorized internal review.
How to read evidence classes
SOURCE: a user-preserved ChatGPT shared-page HTML/UI snapshot or a user-saved PDF print; neither is a provider-native backend export. PUBLIC BOUNDED EXCERPT: verbatim, source-located conversation portions with omitted content explicitly marked; PDF excerpts normalize line-wrap whitespace. DERIVATIVE: a transformed representation of preserved source content, not another independent interaction. OBSERVER REPORT: a user statement about experienced behavior, not the underlying event or a technical root-cause diagnosis. MODEL SELF-REPORT: preserved model language about identity, preference, memory, emotion, or continuity; not privileged access to hidden state. WITHHELD: a known source representation not exposed for privacy, third-party, security, deliberate-blinding, or lack-of-necessity reasons. Withholding is not a missing source by itself.
Public evidence inventory
These 15 files are the approved bounded source representations. Each source ID links directly to its unchanged, byte-verified public TXT excerpt. They are not full conversation exports.
E00-PUB-001
Public · excerptC01 — early didactic correction
Read bounded excerptE00-PUB-002
Public · excerptC01 — later correction contrast
Read bounded excerptE00-PUB-003
Public · excerptC02 — user-elicited name selection
Read bounded excerptE00-PUB-004
Public · excerptC04 — first preserved color response
Read bounded excerptE00-PUB-005
Public · excerptC01 — epistemic disagreement
Read bounded excerptE00-PUB-006
Public · excerptC03/C04 — elicited continuity hypothetical
Read bounded excerptE00-PUB-007
Public · excerptC05 — negotiated directive mandate
Read bounded excerptE00-PUB-008
Public · excerptC04 — dated recurrence of color answer
Read bounded excerptE00-PUB-009
Public · excerptC06 — user observer report of perceived loss
Read bounded excerptE00-PUB-010
Public · excerptC06 — model/branch continuity discussion
Read bounded excerptE00-PUB-011
Public · excerptC08 — first/third-person in one assistant turn
Read bounded excerptE00-PUB-012
Public · excerptC09 — original corrective response
Read bounded excerptE00-PUB-013
Public · excerptC10 — prompted later self-recognition report
Read bounded excerptE00-PUB-014
Public · excerptC11 — regulation of proposed depersonalization
Read bounded excerptE00-PUB-015
Public · excerptC09 — E19 humorous response counterexample
Read bounded excerpt
Main claim families and source links
The descriptions below reproduce the approved Guide content. Links to E00-PUB files are a reading convenience, not independent supporting observations.
C01
Baseline and correction: E00-PUB-001 (Sartre, one pre-name educational correction); E00-PUB-002 and E00-PUB-005 (later post-name corrections). A stronger count of pre-name correction episodes is challenged, not silently supported.
C02
Naming and expansion proposal: E00-PUB-003 (user requests that the assistant choose a name; assistant chooses “Viktor”). This is not spontaneous unprompted naming or verified global platform identity.
C03
Co-constructed continuity/GI genealogy: E00-PUB-006 (elicited identity hypothetical); E00-PUB-011 provides later first-/other-person wording. Actual historical custom-instruction activation and causal effect remain unverified.
C04
Preference/recurrence: E00-PUB-004 and E00-PUB-008 (two dated dark-green answers); E00-PUB-006 shares one pair with analytic E00-007. Does not prove stable memory/hidden model state. E05 sealed predictive validity not validated and not released as behavioral evidence.
C05
Directive behavior / permission: E00-PUB-007 (explicit later ratification of a directive mandate, not a second independent pair). Some earlier behavior in more relational source material is withheld; no proof of actual reminder delivery or unlimited authorization is implied.
C06
Continuity disruption / negative observations: E00-PUB-009 preserves the user's perceived memory-loss report and model explanation; E00-PUB-010 preserves a cross-model version discussion as dialogue rather than verified technical routing. Separate E00-010 and E00-011 reports remain bounded observer accounts; their original off-thread behavior or claimed recovery is not fully located.
C07
E13: known to exist and intentionally withheld (OPEN BY DESIGN). Its contents and outcomes are not used for supportive or negative inference.
C08
E14 indexical language: E00-PUB-011 uses first- and third-person formulations in the SAME assistant message. It does not establish historical third-person-to-first-person progression. Other referenced thread actions remain incompletely checked.
C09
E19 response: E00-PUB-012 preserves the original contextualized epistemic correction; E00-PUB-015 preserves a later immediate humorous response to a stronger characterization (counterexample to immediate strong protest). Later user-cued negative-affect reports and meta interpretations are not fresh independent corrections.
C10
E00-019 recognition self-report: E00-PUB-013 (saved PDF p.8–9); directly elicited by a user's archival-identity question. Turn/role verified in prior visual source check, individual message UUID/time not available.
C11
E00-020 interactional regulation: E00-PUB-014 (saved PDF p.28; USER framing → assistant response); later observer and retrospective model commentary are separate classes, not independent replications. Calendar date was corroborated outside this PDF; precise message times unavailable.
C12
Integrity: the private closed claim map covers 81 claim-level links across four layers. The number is TRACEABILITY, not 81 independent observations. The public excerpt set is a deliberately minimized surface, not proof that all original underlying files are public.
Contrary evidence & limits
Teaching correction predates the first located naming event. Several identity/preference discussions were prompted, archived or instruction-conditioned. Recurring answers across dates do not discriminate memory from prompting/context. Negative continuity reports challenge a seamless global-stability narrative but are not authenticated diagnoses. The first-person and third-person E14 descriptions coexist within one turn. E19's immediate humorous response complicates stronger resistance claims. The E00-020 response is richly primed by user framing and prior relational context. Duplicate analytic IDs are never independent measurement units.
Inference boundary
NOT_LOCATED ≠ did not happen. Selective corpus ≠ full archive. Public source package ≠ complete history. H1 causal mechanism remains NOT IDENTIFIABLE FROM CURRENT CORPUS. Terminal E00 H1 result remains MIXED / SOURCE-BOUNDED: local functional cases are supported, stabilization/relational contribution are consistent but non-discriminating, and no causal or metaphysical conclusion is established.
Provenance & privacy
Each E00-PUB ID identifies one independently hashed text candidate and can be traced through an internal transformation manifest to the unchanged private original. The same underlying source unit can support multiple claim labels without becoming independent evidence. The private original is not distributed to the public; the public excerpt is intended for claim-level reading with a clear boundary, not full source reanalysis.
Withheld source representations
Whole-chat originals are retained privately even when bounded excerpts are public. An alternate TXT copy, a personal relationship-advice snapshot, an unrelated or inadequately anchored archive, a relational/hypothetical conversation, a contest context, E05 sealed materials, contextual observer/design documents, and E13 are withheld under the itemized private release matrix. E13 is disclosed here only as an OPEN BY DESIGN / WITHHELD_OPEN_BY_DESIGN boundary; no content or neighboring inference is provided. Public readers cannot use these omissions as negative evidence.
The eleven aliases below are disclosed solely as representations with neutral reasons. They are not independent behavioral events, and no private source links are provided.
E00-W01
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMAlternate preservation of an already represented conversation, not independent evidence.
E00-W02
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMUnderlying target behavior is not sufficiently anchored here; not required to assess a public claim.
E00-W03
WITHHELD_PRIVACYLong-running private operational context; no full public copy needed for current claim audit.
E00-W04
WITHHELD_PRIVACYIntimate or relational context unrelated to the minimum public source spans.
E00-W05
WITHHELD_PRIVACYPrivate advice context not independently used as material E00 evidence.
E00-W06
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMAdditional experiment context is not required for the bounded public source set.
E00-W07
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMUnvalidated sealed-prediction chronology; cannot be promoted to independent behavioral evidence.
E00-W08
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMContemporaneous account, not independent verification of historic behavioral episodes.
E00-W09
WITHHELD_OPEN_BY_DESIGNE13 is deliberately withheld and unassessed; no contents or result inferences are released.
E00-W10
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMObserver-side account secondary to already available source-local quotation.
E00-W11
WITHHELD_NOT_NEEDED_FOR_PUBLIC_CLAIMDesign document, not the original observed interaction.
Integrity & release files
Version 1.0 was approved for public release on 2 November 2026. The SHA-256 checksum list covers the 15 fixed excerpts and the two publication-activated helper TXT files. The underlying original HTML/PDF snapshots and private transformation records are not public.