BP-0051 · Einstein / Causetica

This first standalone documentary dossier preserves Causetica, a historical proposal for a learned perception module, growing event-order memory, explicit growth rules, an epistemic graph network and a controller. It supplies an architectural research plan, conditional analytical evidence and documentary qualifications. No implementation, training, simulation or hardware measurement is delivered.

The planned destination is the WR-owned private repository Flynn/bp-0051. Actual repository, commit and platform IDs are null. Rights are unknown, known source and review answers are exposed, and publication is unperformed. Historical scientific acceptance, current dossier review and external root adoption have separate statuses.

Read the materials

Material	Role
Full original	All 70,515 original bytes and literal notation.
Full accepted assessment	All 23 BP-0051-C claims, main excerpts, rationales, scopes and evidence references.
Historical catalog	Unchanged catalog with its original path conventions.
Historical review guide	New guide to paired review, metadata-only closure and later PDF qualification.
Selected evidence	Exact upstream selections and the historical E1–E4 analytical arguments.
Source provenance	Copy identities, original bases and documentary roles.
Machine record	Explicit claim, comparison, analysis, qualification and question locators.
Rights	Unknown licensing and separate original, assessment, catalog and derivative-guide roles.
Placement	Eleven prospective repository members with unique targets and roles.
Manifest	Ten non-self member hashes; the external handoff binds the manifest.

Markdown links resolve from their containing document. Structured paths and pointers use their explicitly declared document and path base.

Follow the proposed model

C03–C06 separate a finite partial order from the physical geometry it might approximate. The source calls its order reflexive while writing a strict-looking symbol and later using a DAG. An encoding must declare whether stored edges are strict, reflexive, cover edges or full transitive closure. Classical Sequential Growth needs downward-closed precursor pasts and normalized transitions satisfying its stated requirements; an example rule weight is not that distribution. A VAE latent vector supplies neither a Lorentzian metric nor homogeneous Poisson counts. Perception-determined coordinates, manifold-induced order and rule-guided growth need a specified interface.

C09–C14 distinguish components from their proposed integration. A Gaussian VAE compresses inputs but does not guarantee preserved semantics or calibrated epistemic uncertainty. Replacing a predictive transition model with order growth still requires action-conditioned dynamics, state values, rewards and rollout distributions. The historically checked EpiGNN provides a relational message-passing mechanism with supervision and algebraic assumptions; that is different from identifying causal effects. Candidate rule admission explicitly allows human or automatic validation, but the extractor, causal evidence, conflict handling and revision tests remain missing.

Physical order can constrain possible influence without identifying an actual mechanism. A statistical causal effect needs its own assumptions and intervention information. Associating a percept with an event gives an instance identity, not automatically a reusable concept or correct causal explanation. None of these relations grants authority to act.

Keep the two notation issues distinct

C15 analyzes the literal Markdown at line 168: the paired vector is maximized, while the prose says other vectors are minimized but prints j=i. The accepted analysis and full original are unchanged. This is not presented as a definitive transcription of the PDF.

The existing scan PDFT-003-04 recorded possible_relevant_addition. Its later follow-up PDFT-003-04-U1 establishes that PDF physical page 13 contains an additional reproducible embedded glyph before a separate equals sign. The bracketed form j[unresolved glyph]=i is editorial notation, not literal PDF text. The mapping identifies a private-use representation (U+E020); it does not identify the intended mathematical operator. The intended meaning remains unresolved, recovered_operator is null, no source-content supplement was justified and the scientific status did not change. Familiar contrastive-learning semantics do not repair the source. The practical investigation is closed; this dossier calls for no new page opening, imaging or extraction.

Separately, physical page 15 and the roadmap retain the literal 107 and the first row's extra cell. They do not restore an exponent or column ownership. The earlier assessor's resource arithmetic is conditional on interpreting 107 as 10^7. That condition remains visible and is independent of the page 13 glyph.

Inspect the historical analytical boundaries

The exact E1–E4 arguments in selected evidence were authored by the historical assessor and independently reviewed on 20 October 2026. They are conditional analysis, not measured failures of an implementation and not new results of this dossier.

Historical argument	Premise and consequence
E1 / C08	Holding event order fixed while changing its interpretation leaves an order-only score unchanged. That score cannot alone certify semantic truth. A monitor with extra interpreted evidence changes the premise.
E2 / C22	An explicit strict total-order closure contains n(n−1)/2 pairs; appending a maximum adds n relations. A compact chain reduction has different reachability costs. The roughly 200 TB destination-index example assumes n=10^7; it predicts no typical graph.
E3 / C21	The printed budget sums to 31.5 GB before a symbolic engine strictly below 0.5 GB, with 110 MB of weight rows. Compatible arithmetic supplies no training or runtime bound.
E4 / C17	Identical visual information with two swapped auditory codebooks cannot identify both first actions. The balanced first-choice bound assumes no prior sound mapping; paired data, grounded encoders or feedback change it.

The source's dimensionality, grounding, success-rate, memory and update-time gates remain prospective. Finite local intervals alone give no constant-work insertion guarantee. Neighborhood caps require declared approximation and can discard long dependencies. CUDA execution graphs and semantic event graphs have different roles; numeric precision is not logical correctness.

Compare with the same labels, modalities and access

These six routes are the accepted assessment's /simpler_comparisons, not blueprint results. Each route remains unexecuted.

Assessment pointer	Comparison route
/simpler_comparisons/0	Timestamped typed event log with explicit rules versus causal-set geometry.
/simpler_comparisons/1	Ordinary learned transition model or known simulator versus CSG predictive memory.
/simpler_comparisons/2	Exact rule closure and standard relational GNN versus EpiGNN at matched supervision.
/simpler_comparisons/3	Direct keyed event-percept associations and aligned encoders versus contrastive event identity.
/simpler_comparisons/4	Prediction-error and contradiction monitors versus CCS on independently injected errors.
/simpler_comparisons/5	Reachability indexing and transitive-reduction storage versus explicit closure, counting all updates.

Declare labels, paired modalities, prior codebooks, pretraining, action/reward information, hidden variables, withheld tasks and total storage/update/training costs before interpreting a difference. A plausible counterfactual answer or successful task does not uniquely diagnose causal reasoning. One optional existing question is quoted exactly from /research_questions/0: Does causal-set order improve a held-out intervention task beyond temporal ordering and explicit causal mechanisms? It remains unexecuted.

Evidence versions and scope

The original nine external-source metadata records and the assessment are dated 20 October 2026. Their scopes range from selected primary text to abstract-only or manufacturer-table reading. The paired Einstein/BP-0059 independent report and author metadata closure are dated 20 October 2026. The later lane-A lead decision records the same date at its own /date. Different actors and stages remain separate. The full assessment changed only review_status and review_records: reversing them reproduces all 33,294 earlier bytes. Disposition, technical content and all claims are unchanged. The shared BP-0184 correction and BP-0201 author readback belong to those sources.

Bibliography item 59 is CLIP-VG; the historical assessor separately added the original CLIP primary abstract. Neither is audio-transfer evidence. The Hauptvermutung record preserves its route-versus-served-date discrepancy. Proofs, result tables, interactive figures and original experiments were not generally audited or reproduced.

The earlier PDF scan author declared 26 overview pages and enlargements 13 and 15. Its independent reviewer viewed overview pages 1–2 and those enlargements, read raw Markdown ranges 1–10, 160–169, 197–204 and 323, and separately compared bibliography item 59 in extracted PDF page 25 with MD line 323. Lead triage adoption sampled retained Einstein page 13. Later source-specific mapping and alternate-render diagnosis, independent readback and lead adoption support only the documentary qualification described above. The five register queries bp-0051, bp-0051-c, einstein, causetica yield 0/0/0/6/5 matched string values. The candidate, accepted scan and later qualification are separate records. PDFs, duplicate Markdown, images, primary works and complete shared reports remain reference-only; no new PDF or scientific campaign was performed.