Communisme / CEI

CEI proposes many lightweight PRO agents with standardized neural perception and a generic symbolic rule interpreter. Agents load rules from a shared dynamic graph, select work by knowledge affinity and coordinate through graph changes. A planner posts hierarchical goals and priorities. A thesis–antithesis–synthesis cycle and dissident agents are intended to revise knowledge. The political framework supplies a historical design analogy, not validation of the mechanisms.

This is the first standalone documentary dossier for BP-0034. It bundles three complete historical files: the original, final accepted assessment and current catalog. Initial and pre-scope-correction full assessments are externally pinned corpus inputs, not bundled members. Current dossier review and root adoption follow the frozen submission in separate records. No CEI implementation, simulation, training or new primary/PDF/hardware investigation is supplied.

Read the evidence

Material	Role
Complete original	All 45,905 bytes, 241 lines, tables, examples and 34 bibliography entries.
Final full assessment	All 28 claims, complete passages, scopes, rationales and evidence references.
Current catalog	Exact accepted historical identity and lead decision.
Historical review guide	CEI versus ANIMA scope and the real B021B-CF01 correction.
Selected evidence	Exact historical values, three-stage comparison and the two CEI arguments.
Machine record	Separate statuses and routes to the actual source fields.
Provenance	Complete-copy identities, external versions and original path bases.
Rights	Unknown rights with copied materials and derivative guide identified separately.
Manifest	Ten member identities; its own identity is externally bound.
Placement	Eleven unique prospective repository targets and roles.

Read claims with their kinds, statuses, scopes, rationales and declared primary-read boundaries. Historical direct inspection of published text/code is not training, running an optimizer or executing CEI. Bibliography strings do not certify their linked works. Shared review counts cover 59 claims across CEI and ANIMA; CEI has 28 claims and two of the five assessor arguments.

Information, shared state and revision

What is supplied — C01, C03–C08, C16. The actual information_contract names the ontology/rules, initial knowledge, action/HTN methods, priorities, pretrained perceptors or teacher/distillation, support labels, synthetic-data generator, conflict cases and correctness oracle. Proposed learning includes upstream perceptual parameters, rule/category revisions and episodic/procedural records. Transformations still missing include symbolization, uncertainty/negation, revision preference, graph transactions, claiming/termination, planner methods and media generation. Those supplied priors and transformations are part of the information and cost budgets.

Shared state — C04, C06, C13–C16, C28. A generic rule engine may reuse explicit rules, and shared models/storage can reduce duplication. Perceptor weights still encode upstream learning. A graph assertion is not external truth. Common access, consistent versioned reads and actual processing of a fact are separate properties. Define identity, time, provenance, rule versions, concurrent reads/writes, retraction/dependency invalidation and arbitration. The source names affinity and stateless/identical agents but leaves distance, working-context location/lifecycle, task claiming, duplicate work, synchronization, fairness, deadlock and completion unspecified. Inspectable local steps do not by themselves predict global timing.

The planner needs grounded goals, operators, preconditions/effects, methods and a bounded search/resource policy. Strategic decomposition and later subplanners do not remove those obligations. Later planner, hardware and stagnation qualifications remain attached to earlier robustness/consistency/efficiency claims.

Revision — C02, C03, C11, C12, C27. Thesis/antithesis/synthesis leaves the update operator, candidate selection, revision criterion, evidence, provenance, validation, termination and recovery policy open. The early rule ends in an ellipsis and is preserved. A coherent change after one observation need not identify the intended rule or generalize. A challenge generator needs grounded observations, budgets, protected provenance, tests and rollback; manufactured contradictions are not observations and may instead warrant retaining a rule or abstaining.

C11 is the historical assessor's exact countermodel. Set IsObject(A), IsVogel(A), IsBlauw(A), KanVliegen(A) and OnderWater(A) true. The stated bird-implies-can-fly rule and observations hold together. No negated-flight premise or underwater incompatibility axiom is supplied. Flight capability differs from current activity. Fish reclassification or a waterbird category is not logically forced. Adding such an axiom would add knowledge; this dossier does not do so.

The later flying-penguin test could create a contradiction under properly grounded flight observations and compatible negation. Its 95% criterion concerns coherent hypotheses, such as manipulated image or incorrect rule. It does not establish a 95% externally correct diagnosis.

Visibility versus dissemination — C19, C20. C20 is the second CEI assessor argument. With coherent shared storage and immediate reads, a committed fact may be available to every permitted reader; its actual use still depends on queries and scheduling. Under the different model of one initially informed agent, with each informed agent contacting at most one new agent per synchronous round, at most 2^t agents are informed after t rounds. Reaching m therefore needs at least ceil(log2(m)) rounds. This is a conditional lower bound, not a CEI guarantee. The source does not specify either communication model. Its greater-than-50% goal after X iterations leaves X, topology, rates and update schedule open. A task impossible for a single agent due to missing observations/actions also needs a different comparison from a task merely slower under a matched resource budget.

Perception, compilation, media and resources

Component scope — C07–C10. The accepted CLIP abstract provides image-text pretraining and downstream transfer context, including its 400 million pretraining pairs. It does not establish a compact distilled checkpoint or one-example CEI invariance. Charge labels, teacher access, distillation and validated synthetic data. Object identity/localization, thresholds, calibration, unknowns and continuous-to- predicate translation remain unspecified.

Lobster supplies a bounded relational-compilation/semi-naive evaluation precedent and selected provenance modes. The evaluated top-1 proof cap does not provide exhaustive proofs or a dynamic rule-revision swarm. The concept-variable diffusion reading is narrower than audiovisual generation and does not implement the CEI synthesis operator. None of these component results validates the whole design.

Generation — C17, C21. The early perceptor role is perception-only, while the animation example later introduces generation-perceptors. Source line 125's empty subgoal placeholders remain empty. Generators/renderers, asset/version/dependency contracts, temporal consistency and independent media-quality evaluation are missing. Shared assets and a logged process do not themselves produce correct SVG, 3D geometry or film output.

Resources — C25, C26. Lightweight threads are execution units, not automatically complete neural-symbolic agents. Count shared weights, per-agent activations/state, graph and join/index buffers, provenance, synchronization, scheduling and host/GPU memory. Symbolic compression needs explicit equivalence semantics and algorithms. The historical manufacturer's 32 GB statement does not certify an agent count. Lobster's recorded CPU/GPU transfers are amortized rather than universally eliminated; its high-memory evaluated transitive-closure case is not evidence that CEI fits a consumer device. No current hardware specification check or CEI workload measurement is made here.

Targets and the six existing comparisons

The source proposes 100% bounded planning success within a declared time limit, one-example recognition significantly above a baseline, 95% coherent conflict output, 100+ then 1000+ agents, greater-than-50% knowledge after unspecified X and TI above 90%. Keep those surviving qualifications: the source does request a baseline and time limit, while their identities, suites and budgets remain open. TI needs a decision-step denominator covering perception/generation/planning. EC lacks an estimator/encoding/input and denominator convention; no new compression surrogate is supplied. DER counts synthesis events, which have no stable data or compute unit. Performance improvement and total upstream/adaptation costs must remain separate.

The actual simpler_comparisons field contains six reading routes:

1. A single exact rule engine with a versioned shared work queue versus many PRO threads, using identical predicates/rules and total compute.
2. The same pretrained encoder and labels with a prototype classifier and exact rule checks, measuring new instances, calibration and abstention.
3. Explicit minimal-change belief revision or truth maintenance, with the same candidate repairs, evidence, validation and costs.
4. A*/BFS/HTN using identical actions, methods and the union of observations.
5. An event-driven dependency scheduler with identical generation tools/renderers, assets and full upstream/tool costs.
6. CPU and GPU semi-naive Datalog with matched semantics/provenance, transfers, indexes and join-memory costs.

The assessment has no research_questions field. No optional question or new experiment is invented. Existing comparison routes are unexecuted study material.

Historical correction and delivery

The supplied initial CEI assessment first changed only review_status and review_records. The pre-scope-correction version still contained an independent- review denial in every claim scope. Coordinator B021B-CF01 identified this after an earlier clean readback missed it. The final CEI version substitutes exactly 28 scopes with separate review_ref attribution while retaining empirical and competence/safety limits. Statements, kinds, statuses, rationales, evidence refs and quotations remain unchanged. Exact reversal reconstructs both available older byte stages. The joint correction is 59 scopes: 28 CEI and 31 ANIMA. The previous review and its clean finding list remain preserved history. Independent correction closure and lead acceptance are distinct steps.

The proposed private destination is Flynn/bp-0034, owned by WR. Rights are unknown, actual repository/commit/platform IDs null, and no public release is authorized. Known answers are exposed study material. The lead decides adoption and sealing and retains the exact original owner handoff locally under the WR communication hold. This documentary edition does not implement or run CEI.