Simulation Lab collects frozen, rejected, open and historical simulation series. A simulation shows what happens in the constructed world — not automatically what happens in people, institutions or nature.
Frozen means that the question and conditions are locked — not that the outcome is true. Test conditions and results may be public while the operational M.E.M.I. core remains protected.
A stipulated operational measure of available decision paths. It must not be described as proven ‘genuine freedom’. The historical series reported a decline from 0.82 to 0.19 under high-bias conditions.
With equal total bias exposure, the early-bias path ended at 0.18, late bias at 0.51 and control at 0.74. The effect may arise from the model's accumulation mechanism and is not a claim about humans.
71% occupied the defined 0.16–0.24 plateau, 14% recovered above 0.5 and 14% ended below 0.08. The plateau is a simulated third outcome, not an empirical clinical finding.
Mild +0.11; corrective +0.18; relational→narrative +0.21. Narrative alone produced 25% collapse versus 5.6% in the hybrid path. The mechanism is stipulated in the model and cannot confirm clinical effect.
The historical preliminary run reported plateau rates of 71% versus 28% and 5/6 activated recovery paths. ‘Meta-awareness’ is replaced with threshold-based state monitoring.
Early band_span did not predict later generativity (r = 0.08 / 0.004 / 0.009). Under hard knockout, HIGH versus LOW produced +15.1 percentage points recoverability. Cross-domain generalisation remains plausible, not earned.
Sharing reduced unmet total from 551.02 to 412.67 (about 25%) and import from 10,291.5 to 8,045.4 (about 22%). A 25% gate did not change total damage, but distributed it across 226 versus 198 blackout hours and reduced average size.
All v0.6f scenarios reached 0.063. This may be an implementation ceiling and must not be used as evidence for or against the gate hypothesis until isolated.
At 0 MW, oil demand was 8,229 MWh. At 5 MW it fell to 1,702 MWh (about 79%); at 10 MW it reached 0 in the model. Pressure days were 365 in every scenario. The model uses a simplified annual path, not verified operating data from Nakskov's district-heating operator.
City Energy sources: FREEZE_NOTE_city_energy_lab_v0_6f_sharing_vs_gate.md, city_energy_lab_v0_6f_severity.csv, nakskov_heat_v0_3_summary.csv and nakskov_heat_v0_3_monthly.csv.
The same local language model is tested alone and through M.E.M.I. on missing information, conflicting sources, constraints, replanning, interruption and recoverability. No AGI conclusion is permitted from the pilot.
Both arms had 20/20 decision_match=false and 0 lookup calls. The M.E.M.I. arm also carried more prompt work. The result pointed to a language-access floor and triggered a new test rather than a rescue interpretation.
Tests whether LANGUAGE_IN and LANGUAGE_OUT can be carried as separate one-way roles. The gate speaks only to language carriage. An authoritative user run is pending.
Claude's English draft contained a completed IDK validation while the Danish page did not. The track is therefore not presented as a result until the underlying freeze and result package is identified.
World Model v0.18 passed T1–T7, but failed full lineage retention, abstention volume, single-world calibration and the signature-copy gate among others. A technically correct run was not a positive substantive result.
v0.17 was classified NO_BOUNDED_LINEAGE. v0.18 tested a new hypothesis about internal abstention calibration; it did not rewrite v0.17 as success.
Sprint 1 and 2A were rejected and retained. They helped distinguish Recoverability v1 from v2 rather than being removed from the record.
K1 function; K2 attractor/recovery; K3 history; K4 active revision. The class is an observation frame — not a claim of biological equivalence.
AO-01/02/03/05/06 alter the observer's epistemic access. AO-04 ENABLE_CONTINUATION instead alters the system's own conditions for a next relation to remain possible.
AO-05 was replicated through sequential diagnosis: select the most discriminating access surface. AO-06 was replicated through gravitational lensing: reconstruct a hidden difference through an information-bearing mediator.
Selection before changed geometry may retain a surface that later becomes unavailable. Changed geometry before selection can expose a different and more discriminating surface.
Densification acts on the access path present at the event. Mediation first allows the carrier path to be densified above readout threshold; the reverse order does not.
Current-window/current-state semantics produce different final states and are non-commutative. A persistent acquisition policy is inherited by later window extension and is commutative in the frozen test.
A physical ecological case supports classifying an intervention that changes continuation conditions rather than measurement access alone. Standard ecology explains the case; AO-04 is the abstract relational classification.
Connectivity at matched habitat quality is tested against long-term biological continuation. No outcome status is assigned before a qualifying long-term result exists.
The IRS 3 track identified relevant local shell variables, but missing matching, 3-D exposure, homogeneous chemistry and localisation blocked promotion. The schema was not changed.
The Access Map preserved ABSTAIN, ACCESS_GAP, DISCRIMINATION_GAP and bounded NEGATIVE_NULL in an independent spectroscopic workflow. No material identification was claimed.
Derived 011101 was unusual in E. coli under simple nulls but ordinary at Markov orders 5 and 7. Classification: EXPLAINED_BY_LOCAL_STRUCTURE_AT_ORDER_5.
RPS is a programme about conditions and operators before manifestation — not an alternative to the chemical periodic table in Layer 3.
Freeze the relational conditions and transitions required by RPS, then derive in advance which constraints or patterns should be recoverable in the manifested periodic system of elements. The test concerns the generative connection between layers — not competition with or replacement of the established table.
Version, freeze note, configuration, seeds, raw results, discriminator and claim ceiling must be findable without releasing the protected operational core.
A positive simulation result does not become empirical through translation. The next stage requires real data, relevant practitioners and pre-declared stop rules.
Demand profile, production capacity, fuel constraints and residual load must be verified before the scenario can support a local recommendation.
The five historical experiments must be connected to their original code and results. Until then they remain an internal simulation track, not reproduced public evidence.
The simulations are not the conclusion. They are a safe place to discover what we do not yet know.
Contact about replication →