The Research Corner collects documented findings, rejections, open questions and historical development tracks from M.E.M., M.E.M.I. and ETOS. A status always applies only within the declared test, version and access to reality.
The same relational method has been used across domains. The domains are not identical, and a result does not transfer automatically between them. The purpose is to ask whether the structure can meet resistance without being rescued afterwards.
Knowledge becomes stronger when test conditions, countercases, negative results and conclusion boundaries can be inspected and challenged.
Openness does not mean releasing operational beta architectures without the constitutional and relational conditions governing their use. M.E.M.I.'s complete runtime, the minimum structure and reconstructable implementation details remain protected.
The evidence must be open to inspection. The operational core is not released without the conditions that make responsible use possible.
An outcome is recorded; an explanation is not established.
Supported, but insufficiently discriminated or replicated.
Survived the declared frozen discriminator — only within its scope.
Version, question or test conditions are locked. It does not mean ‘true’.
The frozen test did not pass. The outcome is retained.
The test did not reach the claim and therefore cannot count as a pass.
The available basis cannot distinguish the outcomes.
Retained as provenance, but not current runtime or current evidence status.
A potential field or necessary conditions must be present; a relation or transition carries a difference; a manifestation becomes observable. The manifestation then changes the conditions of the next course.
When a manifestation meets what is actually there, a residual Δ appears. If that difference cannot constrain the next state, a system can remain internally coherent while drifting away from reality.
Registration and integration may be delayed. Drift can accumulate between event, observation and correction. Timing must not be mistaken for mechanism.
Observed may become plausible and earned, but a finding may also be suspended, demoted or withdrawn when new discriminators or countercases require it.
Preserves the boundary between what the system knows, lacks, or is not entitled to infer.
Governs when a decision may continue, stop, abstain or require human review.
Preserves meaning, provenance, history and use boundaries over time. It is a condition of use, not an optional appendix.
K1 function; K2 attractor/recovery; K3 history; K4 active response revision based on history.
100,000+ iterations, 14 sessions, 30/30 in the declared open-world stress test and no recorded governance breach within the stated tests. This is a historical result — not a claim of universal robustness.
Persistent runtime, memory, continuity bridges, governed language, sensors and external-memory ports. Classification: V19_LAPTOP_BODY_EXTERNAL_MEMORY_PORT_PASS_TARGET_HARDWARE_PENDING.
World interaction, causal structure, counterfactual planning, goals, self-model, identity, values, emergent concepts and transfer learning exist as a later development line. They are not documented as integrated and re-frozen into v1.9.
In the v0.8 run, recovery_clustering_cv = 0.000: recovery occurred in contiguous episodes. This is consistent with measured continuity being carried by temporal organisation; it does not by itself establish identity or consciousness.
M.E.M.I. is not documented as AGI. The results show bounded architectural properties under declared tests.
A/B design using the same local language model alone and through M.E.M.I. S0 → answer → Δ → S1 → revised answer. Pilot 01 produced no functional discrimination: both arms had 20/20 decision_match=false and 0 lookup calls. The pilot therefore tested the language access floor and the test itself first.
Tests LANGUAGE_IN and LANGUAGE_OUT as two stateless one-way lanes. No result claim is allowed before the authoritative run. A pass would speak only to language carriage in the frozen micro-benchmark.
All seven technical gates passed, but central substantive gates failed. Frozen classification: NO_CALIBRATED_BOUNDED_READ. The model was not rescued through new thresholds or post-run reinterpretation.
The core 2×2 logic could be routed through the Access Map, while the graded partial-access curve produced schema stress. Strict answer: NO; the result shows constructed mechanistic sufficiency and is not a physical measurement or universal proof.
The mathematical audit separates state spaces from operators: μ organises potential, π manifests, G returns consequence as new potential, and χ makes access history-dependent. Batch 05 retains C-COND as a candidate family; its boundary with C-STR remains open.
AO-01 EXTEND_WINDOW, AO-02 DENSIFY_ACCESS, AO-03 REPOSITION_ACCESS, AO-04 ENABLE_CONTINUATION, AO-05 SELECT_ACCESS and AO-06 MEDIATE_ACCESS. AO-05 and AO-06 were canonised after independent replication; evidence maturity differs across the six.
P1 was retrospectively classified CONTINUATION_PRESERVED. P2 is frozen before a qualifying long-term MESOLAND outcome and may resolve positive, negative, reverse or ABSTAIN. Prospective confirmation does not yet exist.
Composition Test 01 found non-commutativity for SELECT↔REPOSITION and MEDIATE↔DENSIFY. The P3 rerun found EXTEND↔DENSIFY non-commutative under current-window semantics but commutative as a persistent acquisition policy.
O4 produces structure; countercases produce discrimination; discrimination enables promotion from plausible to earned.
Recoverability concerns not only return, but which relations, options and historical conditions can still be carried after disturbance.
Promoted after Audit 01 as a classification principle — not as a causal mechanism. Representation and integration remain separate.
K70–K74 pointed to state-dependent override and Level 3. The status does not transfer to a universal developmental mechanism without a new discriminator.
Energy, health and education yielded different mismatch types. This is consistent with a generative rather than generic structure, but external replication is missing.
Timestamped document chains preserve which questions, stop rules and boundaries arose before later tests. Provenance is evidence of sequence — not automatically of theoretical truth.
RPS is not an alternative arrangement of manifested elements. It examines the conditions, operators and transitions before a periodic system of elements can appear in Layer 3. The established table remains the description of manifested form.
Bridge Test 03 showed that Access Map v0.1.2 could reconstruct the frozen chemical discriminators without schema change or rewriting source outcomes. It does not identify the unknown carrier or replace a prospective Population 04 test.
Five generic chemical cases are identity-frozen around threshold, selectivity/cross-reactivity, interface compatibility, calibration history and valid negative readout. A final population result has not yet been located.
The frozen 0010110 motif and derived 011101 were tested in human mtDNA, yeast and E. coli. The E. coli candidate survived simple nulls but was explained by local sequence structure at Markov order 5. No universal DNA code was established.
SFRD and BHAR co-peak near z≈2. The preregistered VESPA direction falsified H1, while the Firefly signal was stable but sensitive to SSP-library choice. The Firefly MaStar probe remains the decisive open next step.
Pre-registered audit framework. Five tracks were rejected; periodicity ambiguity remains open and may reflect measurement method rather than physical change.
Prediction registered against future DESI data. No outcome exists yet.
Seven registrations produced no positive quantitative match. Five specific laboratory gaps were identified; the feature identity remains open.
The Burren/Magnesian track shows a signal near selected boundaries, while the South Clare candidate was rejected because the proposed zone was not a discrete transition. Category is not function.
A/B asymmetry was 0/50,000 in the declared shuffle test and structural elements were identified. This does not establish a linguistic translation.
The Windfarm simulator supports recoverability under hard knockout, but early breadth did not predict generativity. Generalisation beyond the simulator remains open.
HER2, KRAS, PD-L1 and Alzheimer's are used as cases for recoverability and discrimination. The framework is a research hypothesis, not a diagnostic or clinical tool.
Sprint 1 and 2A were rejected. Their outcomes were retained and contributed to the distinction between Recoverability v1 and v2. Completed does not mean passed.
An environment-sensitive developmental system with three regimes: Stable, Unstable and Overload. Key finding: “Stabilisation is not the same as recovery.” The system does not choose strategies yet; recovery is mechanical, not agentic. Read about AGI-BABY →
Medicine examined through the relation between substance, the female body, internal exposure, hormones, life phase, history, time and recoverability. Methodological backbone: substance ≠ exposure · proxy ≠ function · transition ≠ stable state · absence ≠ sameness. Not treatment or discontinuation guidance. Open the audit → · DOI 10.5281/zenodo.22129841
Relations among light, opening, bearing, landscape and geometry are examined as operator questions — not as established intention or universal code.
Relational mediators, ṛta/aša and administrative substrates are compared with pre-registered control arms. Meaning and historical function must not be conflated.
The theory shapes conditions in which children can try, be wrong, meet resistance and revise. Pedagogical effects require real readers, teachers and learning trials.
Round churches, folklore, the Dodekalitten and other sites are examined as relational spaces. Pattern is not automatically historical cause.
A finding becomes useful only when its boundary can also be seen.