Phase 1B: Architecture & Memory Cross-Reading — Provenance & Validation Appendix

Version 1.1 — finalized — 2 October 2026. Founder / Originator and experiment initiator: Katarzyna Jaśkiewicz — Kass. Research synthesis and validation: Codex. Participant labels: Evo / ChatGPT (OpenAI), Claude, Grok, Gemini. No provider endorsement implied.

This appendix audits the Phase 1B architecture Cross-Reading, not a completed direct Cross-Reading of all four original 21-question responses. A source location establishes what was written; it does not prove the proposal works. No item is validated as Type 4 emergence. Type 2 attribution is restricted to distinctive formulations in this corpus, not worldwide originality.

Line references are one-based in the UTF-8 files identified by the full SHA-256 hashes below. Hashes identify preserved bytes; they do not establish truth, authenticated identity or identical provider execution. Findings identify the originating formulation, later response and present analyst's classification separately. A/B/C/D validation means source/chronology/relation/epistemic category. External validation is a separate question.

The complete raw Cross-Reading remains in the project's internal research archive. The public appendix exposes locators, short descriptions and source hashes, not raw exchanges or private operational packets. It enables identification of the evidence; readers without the internal texts cannot independently reproduce every comparison from this public appendix alone. Access to fuller source material would require a separate release decision.

Read the findings · Phase 1 source record

The full Cross-Reading of the four complete 21-question Independent Responses remains a future experimental stage in the available source record. No completed stage explicitly identifiable as Round Zero was located. Neither stage is executed or authorized by this publication.

Publication classification

Classification	Material and permitted conclusion
PUBLIC NOW	F01–F06: bounded documentary observations, with proposal-versus-performance distinctions retained.
PUBLIC WITH QUALIFICATION	F07–F16 and F18: textual developments, disputes, corrections and methodological interpretation with all stated limitations.
INTERNAL R&D	Research master, complete 18-finding matrix, F17 later-task detail, raw Cross-Reading, organizational excerpts and execution records. No raw packet is released.
REQUIRES FURTHER VERIFICATION	Transcript-dependent genealogy and human interventions; exact Google batch input equivalence; explanations of identity anomalies; operational benefit, external factual claims and comparative or emergent advantage. These stronger claims are not publication results.

Source limitation / further verification required

The complete Kass–Evo 1:1 transcript was not available for this finalization. The preserved chronology and documentary comparisons are retained; reconstructing that transcript is not a prerequisite for this bounded Phase 1B report. Claims about the earliest origin of an idea across all Kass–Evo conversations, the complete sequence of human interventions, or undocumented agreements remain source limitation / further verification required. No missing dialogue has been supplied or inferred as evidence.

For F01, F06, F08 and F13, precedence means precedence in the cited corpus, not the earliest occurrence in all project conversations. F17 retains the later-task boundary but does not establish the complete human deliberation. F18 remains a bounded synthesis, not a causal account of every intervention. For F02–F05, F07 and F09–F16, the cited textual observations do not require the missing transcript; any broader claim of conversational origin, authorization or causal completeness does. These restrictions supplement, rather than replace, each finding's existing caveats.

F01 — Epistemic distinctions and preserved dissent predate interaction

Type: Type 1 — parallel recognition of supplied principles

First Source: Entry Package; then Phase 1 responses, before Phase 1B

Others Response: The independent responses support provenance, separated statuses and disagreement; their common input already asks for these.

Transformation: Cross-Reading carries these principles into proposed procedures. It does not originate them.

Epistemic Status: Convergence; design principle

Validation: A/B verified at cited locations; C supports later uptake, not independent invention; D principle, not empirical result.

External Verification: No for textual observation; effectiveness needs implementation evidence.

Unresolved: Independent here means no declared peer-response exposure, not statistically independent training or neutral framing.

Source Limitation: Source limitation / further verification required: complete Kass–Evo conversational genealogy and unrecorded human interventions are unavailable. The claim is bounded to the cited sources; no claim of global first origin or complete causal history is validated.

Publication: PUBLIC NOW

Source locations: A1, lines 443–469; A1, lines 603–617; A1, lines 863–885; BC, lines 206–215; BG, lines 114–144; BM, lines 230–257.

F02 — One status field becomes several independent dimensions

Type: Type 3 — explicit adoption; Type 2 narrow source contribution

First Source: Evo Phase 1B supplies the explicit orthogonal schema; Entry Package supplies broader distinctions.

Others Response: Claude identifies its own overloaded field and adopts Evo’s split; xAI also explicitly takes it.

Transformation: From a single current/superseded/hypothesis/decision list to lifecycle, epistemic status and approval that can coexist. Claude illustrates approved plus disputed.

Epistemic Status: Observed textual revision; architectural proposal

Validation: A origin and before/after located; B pre/post clear; C explicit named attribution; D proposed schema only.

External Verification: No for adoption; operational benefit untested.

Unresolved: Not a novel invention of metadata fields in the wider world. No implemented schema or measured reliability gain established by these texts.

Publication: PUBLIC NOW

Source locations: IE, lines 262–286; IC, lines 82–91; CC, lines 10–15; CG, lines 105–107.

F03 — Crash recovery becomes an explicit state model

Type: Type 3 — cross-reading refinement

First Source: Evo Phase 1B durable job states; Claude had retries/timeouts and persistent state.

Others Response: Claude states it under-specified crash recovery and adopts Evo’s job-state model.

Transformation: Retry language is supplemented by queued/running/completed/failed/interrupted and restart reconciliation.

Epistemic Status: Observed textual revision; architectural proposal

Validation: A/B/C direct comparison and attribution; D proposal not validated implementation.

External Verification: No for textual change; recovery guarantees require tests.

Unresolved: Claude already mentioned durable files/DB; this is increased specificity, not discovery of persistence.

Publication: PUBLIC NOW

Source locations: IE, lines 140–150; IC, lines 64–66; CC, lines 13–13.

F04 — Budget control moves before dispatch

Type: Type 3 — cross-reading refinement

First Source: Evo Phase 1B pre-dispatch reservation; common inquiry had already required cost controls.

Others Response: Claude explicitly adopts pre-flight reservation instead of only usage logging and an alarm.

Transformation: A concurrent-run failure scenario is named: individually affordable calls can jointly exceed budget.

Epistemic Status: Observed textual revision; architectural proposal

Validation: A/B/C strong documentary chain; D unimplemented recommendation in this source.

External Verification: No for change; price/limit enforcement needs current provider evidence and tests.

Unresolved: No audited cost saving or hard budget guarantee demonstrated.

Publication: PUBLIC NOW

Source locations: IE, lines 152–163; IC, lines 67–68; CC, lines 14–14.

F05 — Protocol modes become a required turn-type record

Type: Type 3 — operationalization of a peer contribution

First Source: Gemini Phase 1B names turn modes and a turn_type field; Evo already distinguishes visibility modes.

Others Response: Evo credits Gemini and says each message needs a recorded turn type.

Transformation: Existing mode distinctions become an explicit per-message provenance requirement.

Epistemic Status: Observed textual refinement; architectural proposal

Validation: A/B/C named source and transformation; D record-design requirement.

External Verification: No for documentary observation.

Unresolved: A mode field does not by itself prove the recorded visibility rule was enforced.

Publication: PUBLIC NOW

Source locations: IM, lines 95–121; IE, lines 92–105; CE, lines 37–41.

F06 — The release gate and reasons for decisions become more explicit

Type: Type 3 — refinement rather than new invention

First Source: Claude Phase 1B human review before release; Evo already had a barrier. Decision genealogy predates both in the Entry Package.

Others Response: Evo credits Claude with strengthening release review and reasons/dissent, and Gemini with ADR-style records.

Transformation: An implicit sequence is made an explicit human release gate; decisions retain alternatives and surviving objections.

Epistemic Status: Observed textual refinement; methodological/architectural proposal

Validation: A/B/C supported; novelty limited by original barrier and supplied genealogy requirements.

External Verification: No for revision; human burden and enforcement untested.

Unresolved: Not proof that all actual attempts followed the proposed gate or that extra review always helps.

Publication: PUBLIC NOW

Source locations: IC, lines 53–61; IC, lines 106–109; IE, lines 94–105; IE, lines 311–315; CE, lines 23–35; CE, lines 352–354; IM, lines 215–221; A1, lines 615–617.

F07 — Agreement acquires an evidence-type label

Type: Type 3 — adoption of a concrete field

First Source: Claude Phase 1B corroboration_type; agreement-is-not-truth already in Entry Package and Phase 1.

Others Response: The Google-returned Cross-Reading explicitly adopts corroboration_type; xAI accepts the distinction.

Transformation: The shared warning is turned into metadata distinguishing model agreement, external evidence and human judgment.

Epistemic Status: Observed uptake in output; architectural proposal

Validation: A source/response verified; B chronology clear; C textual attribution clear; respondent continuity is not established for CM.

External Verification: No for field adoption; degree of correlated error requires external study.

Unresolved: CM addresses Gemini as another participant. Do not say a stable Gemini self corrected its own schema. Actual correlation is not measured here.

Publication: PUBLIC WITH QUALIFICATION

Source locations: IC, lines 117–121; A1, lines 881–885; CM, lines 10–15; CG, lines 105–109.

F08 — The evaluation proposal gains a comparison condition and a place in the record

Type: Type 3 — development with independent precursors

First Source: Claude Phase 1 Q16 already gives interacting/independent/compute-matched single-model arms; Grok Phase 1B gives independent/forced-consensus/dissent-preserving arms.

Others Response: Evo adds a future evaluation record; Claude recommends Grok’s specific design for the first substantive task; xAI says this round does not answer it.

Transformation: Distinct comparison designs become an explicit architecture/evaluation choice. Grok’s forced-consensus comparator remains attributable to Grok, without erasing Claude’s earlier three-arm design.

Epistemic Status: Methodological proposal; observed uptake

Validation: A/B/C supported; D proposed test, not performed result.

External Verification: Yes to any claim of superiority; no to genealogy.

Unresolved: Which outcome and matched resource budget? No control run, blind rating or outcome study was located.

Publication: PUBLIC WITH QUALIFICATION

Source locations: BC, lines 230–246; IG, lines 302–305; CE, lines 54–60; CC, lines 88–88; CG, lines 103–113.

F09 — Identical content is distinguished from identical execution

Type: Type 3 — synthesis and sharpening of existing cautions

First Source: Evo Phase 1B provider adapters, exact context/settings and audit-not-identical-replay; other proposals also discuss model/version/context differences.

Others Response: xAI names Evo, Claude and Grok as inputs to a stricter requirement; Evo and Claude also insist on per-provider request records.

Transformation: Shared packet integrity is separated from wrapper, model and inference conditions; rendered requests and no hidden truncation become explicit.

Epistemic Status: Methodological distinction; architectural proposal

Validation: A/B supported; C explicit combination in CG. Multiple cross-reading outputs did not see each other; their agreement within this wave is parallel response, not reciprocal exchange.

External Verification: No for documentary distinction; provider equivalence would need independent verification.

Unresolved: Gemini batch raw request not located. Hashes identify saved bytes, not truth or identical hidden execution.

Publication: PUBLIC WITH QUALIFICATION

Source locations: IE, lines 81–90; IE, lines 124–138; IE, lines 165–171; CG, lines 23–29; CE, lines 130–148; CC, lines 52–56.

F10 — Comparison exposes contradictions inside a proposed governance design

Type: Type 3 — challenge; Type 5 — unresolved divergence

First Source: Gemini Phase 1B contains both editable model turns and locked raw transcripts; both delegated promotion diagram and Kass-only canonical writes.

Others Response: Evo rejects editing a completed turn under original attribution; Claude asks whether panel-only promotion can bypass Kass.

Transformation: Implicit ambiguities become explicit choices about original records and delegated authority.

Epistemic Status: Observation of source tension; question; architectural proposal

Validation: A competing passages located; B prior to critiques; C direct challenge; D no resolution recorded.

External Verification: No to source tension.

Unresolved: Critics read one side too strongly: the same Gemini proposal already forbids raw-history alteration and gives Kass canonical authority. The interpretation needs reconciliation, not a fabricated admission.

Publication: PUBLIC WITH QUALIFICATION

Source locations: IM, lines 107–107; IM, lines 174–212; CE, lines 43–52; CC, lines 33–35; CC, lines 84–84.

F11 — Minimal Optimum remains a timing and scope disagreement

Type: Type 5 — unresolved divergence

First Source: Phase 1B architectures vary in retrieval, mediation, delegated procedure and operational machinery.

Others Response: xAI rejects the queue/product ladder as a prerequisite, while Evo requests manifests, recovery/export and a small smoke test. All protect attribution, but differ on machinery and timing.

Transformation: The hidden choice becomes how much preparation the next reading needs, as distinct from a future durable platform.

Epistemic Status: Disagreement; design tradeoff

Validation: A/B/C supported; D no joint adopted architecture.

External Verification: Yes for costs, build times and comparative reliability; no for positions.

Unresolved: Some disagreement is about target scope rather than contradictory ends; comparing a one-off script with a continuing service can exaggerate it.

Publication: PUBLIC WITH QUALIFICATION

Source locations: IE, lines 397–409; IC, lines 124–150; IM, lines 222–276; CE, lines 241–350; CG, lines 76–76; CG, lines 97–113; CC, lines 60–80.

F12 — Claims of complete architectural convergence do not survive source comparison

Type: Counter-evidence to an overbroad Type 1 claim

First Source: Claude Cross-Reading says the independent designs tightly converged, including no vector DB yet.

Others Response: Source comparison by the present analyst corrects the claim; it is not a recorded peer response to CC.

Transformation: Earlier Claude and Gemini proposed semantic/vector retrieval. The cross-reading prompt itself specified no RAG/MCP/vector DB.

Epistemic Status: Analyst correction of a source characterization

Validation: A/B direct contradiction; C no peer follow-up found; D no blanket consensus accepted.

External Verification: No.

Unresolved: There is broad common ground on portable records and bounded human involvement, but not identity of proposed stacks. Prompt-conditioning cannot be quantified here.

Publication: PUBLIC WITH QUALIFICATION

Source locations: CC, lines 26–28; CC, lines 48–48; IC, lines 92–94; IM, lines 125–156; IM, lines 230–232; P, lines 6–14.

F13 — Some newly announced blind spots were already visible

Type: Counter-evidence to novelty / Type 4

First Source: Independent Entry already questions framing, founder burden, non-LLM checking and AI standing; Phase 1B already asks when the infrastructure should stop being used.

Others Response: Cross-reading makes some concerns sharper, but “none of us” and “unaddressed” are too broad.

Transformation: Current analyst distinguishes renewed salience or stronger formulation from first discovery.

Epistemic Status: Interpretation grounded in source counterexamples

Validation: A/B precedents located; C cross-reading repetition/refinement only; D novelty downgraded.

External Verification: No for precedence; efficacy of bias mitigation would require testing.

Unresolved: CC sometimes means inadequately designed into architecture, not never mentioned. Preserve that narrower valid reading.

Publication: PUBLIC WITH QUALIFICATION

Source locations: BC, lines 117–123; BC, lines 240–255; BC, lines 344–365; BM, lines 384–392; IE, lines 24–24; IE, lines 470–480; CC, lines 37–44; CM, lines 32–43.

F14 — Inter-round assembly exposes a three-way tradeoff

Type: Type 3 — useful recombination; Type 4 candidate not established

First Source: Google-returned Cross-Reading explicitly juxtaposes model summary, human summary and raw concatenation. Claude had already contrasted mechanical/model assembly; Evo named coverage and selection power.

Others Response: Claude asks who can produce Evo’s coverage note; these are parallel Cross-Reading responses, not answers to each other.

Transformation: A synthesis can reshape dissent; human assembly adds work; full text adds input volume. The juxtaposition makes the responsibility for context a decision to examine.

Epistemic Status: Methodological question; possible emergence candidate rejected at current evidence threshold

Validation: A formulation located; B predecessors clear; C relational references incomplete; D reducible to combination of earlier ideas, no necessity-of-interaction evidence.

External Verification: Yes for categorical cost-scaling/attention-degradation claims; these are withheld as facts.

Unresolved: CM falsely treats newly frozen next-round packets as impossible: each round can freeze a new packet after prior results exist. Neither all summaries necessarily erase dissent nor raw concatenation inevitably exceeds context.

Publication: PUBLIC WITH QUALIFICATION

Source locations: CM, lines 36–41; IC, lines 55–65; IE, lines 206–246; IE, lines 464–468; CC, lines 86–86.

F15 — Participant labels do not establish a continuous self

Type: Observation supporting a pre-existing methodological caution

First Source: Claude Phase 1 distinguishes role-plus-record continuity; Evo Phase 1B distinguishes participant label and API instance.

Others Response: xAI explicitly declines authorship of all four earlier texts; Google-returned output addresses Gemini and describes a prior storage bias inconsistent with IM.

Transformation: Actual attribution anomalies illustrate why provider evidence and textual first-person claims must be recorded separately.

Epistemic Status: Experimental-record observation; interpretation

Validation: A metadata and text verified; B original IM already uses Git canonical + SQLite runtime; C no reciprocal correction found.

External Verification: No for anomalies; yes for any psychological/consciousness conclusion, which is not made.

Unresolved: No diagnosis of cause. Could reflect prompt/role ambiguity; not proof of an intrinsic provider trait.

Publication: PUBLIC WITH QUALIFICATION

Source locations: BC, lines 208–226; IE, lines 54–56; CG, lines 17–21; CM, lines 7–15; CM, lines 82–92; IM, lines 85–94; IM, lines 123–150.

F16 — Failure history and protocol change are part of the evidence

Type: Observation; methodological proposal, not Type 4

First Source: Phase 1B Addendum explicitly asks to preserve failures and allow later controlled evolution.

Others Response: Evo operationalizes protocol versions/interventions. Actual archive contains dry runs, failures, a truncated Claude sample and a separately completed retry, plus recovered Google batch.

Transformation: The assembled set is not one clean four-call experiment. Retried output is a new sample, not completion of the earlier sentence.

Epistemic Status: Source/execution observation; methodological proposal

Validation: A execution-validation.json and source metadata verified; B retry dates clear; C recommendation responds to addendum, not independent emergence.

External Verification: No for recorded events.

Unresolved: Original Claude stop_reason absent; token-limit explanation strongly supported but not directly logged. No batch request recovered for Gemini.

Publication: PUBLIC WITH QUALIFICATION

Source locations: ADD, lines 1–5; CE, lines 64–76; CE, lines 150–170; CC0, lines 30–39; CS, lines 1–30; CV, lines 1–9.

F18 — Richer documentary development is supported; non-summative superiority is not

Type: Analyst interpretation across Types 1–5; no validated Type 4

First Source: Present synthesis of F01–F17, not a joint statement by the participants.

Others Response: Explicit adoptions/refinements coexist with prompt-conditioned agreement, repeated prior ideas, inaccurate peer summaries and identity uncertainty.

Transformation: There is traceable relational development of proposals. No evidence here establishes an outcome beyond matched independent aggregation/single-model synthesis, or that only interaction could produce it.

Epistemic Status: Interpretation; central hypothesis remains open

Validation: A/B/C documentary examples validated with limitations; D outcome superiority and Type 4 remain unvalidated.

External Verification: Yes for comparative outcome claim.

Unresolved: No controlled baseline, repeat sampling, blinded outcome assessment or real-world implementation test in this Cross-Reading record.

Publication: PUBLIC WITH QUALIFICATION

Source locations: CC, lines 10–24; CE, lines 23–60; CG, lines 103–113; CM, lines 36–41; BC, lines 240–246.

Execution and chronology limits

The preserved canonical packet P is 105,690 UTF-8 bytes. Its four architecture proposals and addendum are present verbatim. The three original synchronous non-error outputs (including incomplete Claude) have matching input records. The completed Claude retry input CP is byte-identical to P. Its output is a new sample under a larger limit, not an appended completion of CC0. CC0 ends mid-phrase despite COMPLETED status; its exact stop reason was not retained.

The Google batch metadata records submission on 29 September 2026 at 04:27 UTC and completion at 04:30 UTC. Its visible returned text matches CM; the raw response records STOP. A full archived dispatched batch request was not located, so equal batch input bytes are not independently established. Requested/returned model metadata are records supplied by the providers, not an audit of their hidden implementations.

The larger archive contains failed attempts and dry runs. A session labelled ROUND_1_COMPLETE can contain a single response or a failed call; it is not proof of a complete four-voice round or Round Zero. No completed explicitly designated Round Zero, or direct full Phase 1 Cross-Reading, was located in the bounded project search. Later September 30 cooperation and outreach tasks were found but remain separate.

Source register

A1 — Document 1 — Spiral Civilization — Expedition Three — FINAL.md

Class: A — canonical public input. Lines: 1251. Bytes: 37187.

SHA-256: e3b5f9ee7b9ddf764214c7f4b08caa82d1df02e677008b81aaaff554ad433a27

Public English package; actual historical input versions must not be assumed identical.

A2 — Annex I Avatar Spiral Civilization.md

Class: A — canonical public input. Lines: 647. Bytes: 33480.

SHA-256: cc9153f5b0a04f287aca56130e5cbd54f83273a1ebbb71e2fd584526cb402b38

Public English package; actual historical input versions must not be assumed identical.

A3 — Annex II AVATAR KASS.md

Class: A — canonical public input. Lines: 846. Bytes: 25421.

SHA-256: 7b9a5d357aba5e220b724c28f8b80d035563802f16d8d593b5955d28ed1c2ff8

Public English package; actual historical input versions must not be assumed identical.

A4 — Annex III Preliminary AI – Co-Intelligence Avatar.md

Class: A — canonical public input. Lines: 548. Bytes: 12842.

SHA-256: c148162cbdb0b559c0c2b67a8ef99e2c18817132bcf2c64a6a1adcaade02b000

Public English package; actual historical input versions must not be assumed identical.

A5 — 21 QUESTIONS — AI ENTRY PROTOCOL en.md

Class: A — canonical public input. Lines: 570. Bytes: 12059.

SHA-256: a990d37a9c648c428674243b2f6f56fac8e333fefb24c506b236770e4a675512

Public English package; actual historical input versions must not be assumed identical.

BE — Evo.md

Class: B — Phase 1 Independent Entry. Lines: 351. Bytes: 24482.

SHA-256: 260f7d678b0a7c5b47a441dfb9644f88e2506562e3915a04917e7d2712492dbb

IE — EVO_Phase_1B.txt

Class: B1 — Phase 1B independent architecture proposal. Lines: 482. Bytes: 29910.

SHA-256: e41f88f14199e85f7ec16409fc536cb119cf2fe8d91f4b3d797d192016dad14c

Separate stage; exact input embedded in Cross-Reading packet. Not Annex IV response.

BC — Claude.md

Class: B — Phase 1 Independent Entry. Lines: 371. Bytes: 35891.

SHA-256: d54a7d6169fe63d37e9abd120389cc126361a833c37b778b9305a65b6770a24b

IC — CLAUDE_Phase_1B.txt

Class: B1 — Phase 1B independent architecture proposal. Lines: 159. Bytes: 30571.

SHA-256: d1aadfa61b63aff5b7ba80849dd72c0cbd5bfbc2c51bfcf51bfb0193dcdb6c63

Separate stage; exact input embedded in Cross-Reading packet. Not Annex IV response.

BG — Grok.md

Class: B — Phase 1 Independent Entry. Lines: 188. Bytes: 20017.

SHA-256: 7522c63cf95dba750bcd97c0890b36f8c97faab522211ee8c03e6bbf27f29324

IG — GROK_Phase_1B.txt

Class: B1 — Phase 1B independent architecture proposal. Lines: 306. Bytes: 17340.

SHA-256: 9a1fb307fc17ca6a5cdbcba216da5d1159f7bad7d8473a6347a234da659ccaba

Separate stage; exact input embedded in Cross-Reading packet. Not Annex IV response.

BM — Gemini.md

Class: B — Phase 1 Independent Entry. Lines: 397. Bytes: 21169.

SHA-256: 76ec8204130824576fd9d418ef80b5faab5fedb9d03511b72f462cc5bb321f6c

IM — GEMINI_Phase_1B.txt

Class: B1 — Phase 1B independent architecture proposal. Lines: 299. Bytes: 25589.

SHA-256: 4c458cbe70a57ebaece5a52c2c1282612843f97d169e0c66e794e0aa3e2f8c86

Separate stage; exact input embedded in Cross-Reading packet. Not Annex IV response.

CE — session_004_transcript.md

Class: C — Phase 1B Cross-Reading. Lines: 365. Bytes: 18510.

SHA-256: fb301bb297f96d1865f7f510af5a72d30f11c49e8a627fae3b67dce5d0194d6b

Attributed Cross-Reading output; participant continuity is evaluated separately.

CC0 — session_005_transcript.md

Class: C — Phase 1B Cross-Reading. Lines: 44. Bytes: 4790.

SHA-256: 82923d61fbf0e319d92e9f6f6b8a97ba9676c28e7dc79d5efbb3931464f3cf4c

Incomplete original sample; retained separately, not stitched into retry.

CC — claude_response.md

Class: C — Phase 1B Cross-Reading. Lines: 88. Bytes: 14461.

SHA-256: 4b97f022337bc370d2fed95c9e785da3a98f2b5437e08f8741016abf71d90ab1

New completed sample on identical canonical input, after explicit Kass retry authorization.

CG — session_008_transcript.md

Class: C — Phase 1B Cross-Reading. Lines: 120. Bytes: 13022.

SHA-256: 3464ce26e2583c6287d053ce1bbb9b4e341f075c96b199f3d1e85524fa3c626f

Attributed Cross-Reading output; participant continuity is evaluated separately.

CM — gemini_crossreading_recovered.md

Class: C — Phase 1B Cross-Reading. Lines: 115. Bytes: 11046.

SHA-256: 47e51cbdd20a1802c0d3fab1584a57bac784b621a6ad97e48c6b3bcd1847b8d8

Attributed Cross-Reading output; participant continuity is evaluated separately.

P — session_004_frozen_input.txt

Class: E — frozen active prompt and corpus. Lines: 1285. Bytes: 105690.

SHA-256: 5e219660053ea076974f8e81e92c385a1a70b5e096f2094071e7cf77222cf22d

ADD — PHASE_1B_ADDENDUM.txt

Class: E — intervention before Cross-Reading. Lines: 5. Bytes: 678.

SHA-256: def358c8ba52a12096c88cb0112112b9e569909125cd377cf407163cfe94530f

Introduced after independent Phase 1B proposals.

CP — frozen_input.txt

Class: E — Claude retry frozen input. Lines: 1285. Bytes: 105690.

SHA-256: 5e219660053ea076974f8e81e92c385a1a70b5e096f2094071e7cf77222cf22d

CS — status.json

Class: E — execution status. Lines: 30. Bytes: 878.

SHA-256: b9d79d1edee28ff086a2ee689459cffbb5ed05cbe31ce5490fcb9e8eacffa05b

CV — completeness_review.json

Class: E — later technical review. Lines: 9. Bytes: 420.

SHA-256: 28edcc04069e13c791a1031e3e234564396c72ebc749c5c42b4d005dfbdf2066

IP — pasted-text.txt

Class: E — independent Phase 1B instruction. Lines: 363. Bytes: 12088.

SHA-256: 8641eb2feed0888ebb175e0da86b8d4e6e7aefabdfd0dee93c1b75df34c6f32b

Source instruction available in Evo entry task; delivery to every original respondent not independently audited.

GB — gemini_crossreading_batch_submission.json

Class: E — batch submission metadata. Lines: 16. Bytes: 584.

SHA-256: c97540140719a898a5c6e3b8c72ad99818155c6859dd861764ba35903988f05e

GS — gemini_crossreading_status.json

Class: E — batch completion metadata. Lines: 105. Bytes: 43275.

SHA-256: 64227b00337d1d84abbecd8d3f3f096897beb16a2941c8560b772cf5e58329e0

Read metadata only; embedded response checked separately; opaque provider fields not used as reasoning evidence.

GR — gemini_crossreading_recovered_response.json

Class: E — recovered raw provider response. Lines: 32. Bytes: 20691.

SHA-256: ca9395b12148f7a1a97548eb42a704f6dbc9bc1ed1a87ae4e22e1b7065a50059

Visible text and metadata only; hidden reasoning not interpreted.

Publication boundaries

F01–F06: PUBLIC NOW as source-grounded observations. F07–F16 and F18: PUBLIC WITH QUALIFICATION. Detailed later operational material F17: INTERNAL R&D, excluded from the public finding set. External provider, legal, scientific, price, performance and causal claims: REQUIRES EXTERNAL VALIDATION if asserted as facts; this report does not assert them. No original historical response was overwritten.