Applications · Public portfolio

One architecture. Four physical worlds.

LINUR connects geometry, fields, phase, propagation and energy flow to the domain-specific operators each system requires. Every result remains attached to its evidence boundary.

One evidence-controlled architecture

Shared coordinates. Domain-specific operators.

The current record supports a cross-domain scientific-computing and evidence architecture. It does not establish that every application comes from one equation without domain inputs.

GeometryFieldsPhasePropagationEnergy flow

Evidence-controlledLINURarchitecture

Matter & LifeEnergy & FieldsMaterials & DevicesCosmic Systems

How LINUR learns from reality

Evidence enters first. Equations do not rewrite themselves.

The Engine identifies the most informative next experiment. Laboratories provide physical evidence. Every result first enters an immutable evidence registry; some measurements are declared calibration data, while others remain unopened held-out tests.

01Frozen capability gap
02Predeclared experiment
03Immutable evidence
04Target-blind operator improvement
05Independent transfer
06Additive release

Connected, not interchangeable

Connected layers keep the physical claim honest.

The layers may couple, but one cannot silently stand in for another. Every connection carries units, uncertainty, provenance, authority and the exact operator version that produced it.

01
Nuclear state
Isotope composition, nuclear energy levels, admitted decay channels and intrinsic transition operators.
02
Atomic state
Electronic configuration, charge state and polarization.
03
Bond state
Chemical identity, connectivity, orientation, equilibrium scale and local stability.
04
Molecular state
Connected geometry, energy, conformational branches and environmental response.
05
Lattice state
Crystal organization, defects, collective geometry and phase.
06
Material state
Transport, thermal and electromagnetic response assembled from admitted lower layers.
07
Environment and device
Boundaries, loads, operating conditions and the physical decision being tested.
08
Laboratory evidence
Immutable measurements, uncertainty and provenance used for calibration or held-out comparison.

Illustrative molecular geometry and reaction field

Illustrative image · not evidence

01 · Application domain

Matter & Life

Nuclear, atomic and molecular states remain distinct while their measurable interfaces stay connected.

01Matter & LifePublic capability and evidence boundary

A14Strong showcase

Validated resultEngine-native result

Molecular coordination topology

Derived normalized geometry · coordinate-blind external transfer

Client problem: Can a frozen local-topology operator transfer across newly opened rubredoxin structures without target-specific fitting?
Public capability: From connectivity alone, LINUR derives the regular four-coordinate topology and transfers it across newly opened rubredoxin structures.
Public result: All 16 eligible external Zn–S₄ structures passed the frozen angular and normalized-topology panel limits.
Headline metrics: 16 / 16external structures passed4.97°median angular MAE0target-specific fits< 1 sforward evaluation on the local Mac
Evidence boundary: Local coordination topology only—not a complete protein fold, absolute bond length or X-ray electron-density reconstruction.

Comparison of the LINUR conditional Fe–S₄ reconstruction with the 9TA4 and 9TA6 regression references — Scale and topology frozen before regression scoring · experimental coordinates were not forward inputs

Reduced interpretable resultExtension in progress

Conditional Fe–S₄ Molecular Reconstruction

CALIBRATED CONDITIONAL RECONSTRUCTION / REGRESSION COMPARISON

Given Fe–S₄ connectivity, the sealed tetrahedral topology and one independently published ferric-tetrathiolate scale, LINUR reconstructed the dominant local geometry with zero target-specific fits. Across two regression structures, Fe–S bond-length MAE was 0.0218–0.0330 Å. The stricter complete-coordinate gate did not yet pass, identifying the protein-environment physics still required.

TopologyPASS

Connectivity-only sealed tetrahedral topology · 109.4712206° · zero target-specific fits.

Conditional bond scalePASS

One independent ferric-tetrathiolate scale · 2.284 ± 0.002 Å · frozen before scoring.

Complete-coordinate gateFAIL

9TA4 0.1684 Å · 9TA6 0.1661 Å aligned local RMSD · frozen gate 0.15 Å.

StructureBond MAEMaximum errorLocal RMSD

9TA40.0218 Å0.0364 Å0.1684 Å

9TA60.0330 Å0.0688 Å0.1661 Å

Numerical certificate. 1.13 × 10⁻¹⁶ relative build-bond residual. This is a coordinate numerical certificate, not a physical-accuracy claim.

Next sealed gap. R5BR-XW sealed the next missing layer: a protein-environment Hamiltonian and bounded continuous solver. This is the next target-blind development layer before independent transfer.

Evidence boundary. The scale is calibrated, and 9TA4/9TA6 are previously opened regression structures—not new blind targets. This does not establish measurement-free absolute scale, exact complete structures, a full protein fold or replacement of X-ray crystallography.

Read the evidence report ↗Independent scale source ↗

Technical evidence boundary

Expanded local-topology evidence

Frozen ideal angle: 109.4712206°. Sixteen of sixteen external Zn–S₄ structures were eligible and passed.
Median angular MAE: 4.9680°; range 4.4316° to 5.3013°.
Median normalized topology energy: 0.04808 against a frozen 0.15 panel limit.
Earlier Fe–S₄ comparisons remain preserved: 5.8312° MAE for 9TA4 and 5.6825° for 9TA6.

Reduced Fe–S scale · separate result

A general UFF radial adapter estimated the Fe–S bond scale with 0.0195 Å MAE in the earlier comparison. This is a reduced empirical result, not yet an Engine-derived absolute length.

Reduced interpretable result

Whole-fold and dose-response boundary

The earlier whole-fold model did not pass its conjunction gate. A single-drive elastic network improved the 8 Å dose-response RMS by only 1.52%; the 7 Å and 9 Å variants were worse than zero response. These remain negative or weak technical results, not success metrics.

Negative result

30OH / 30OR continuation

The 0.43 Å rubredoxin study identifies Fe–S₄ electronic structure as an unresolved modelling problem. LINUR’s frozen, coordinate-blind 30OH/30OR comparison remains pending because deposited coordinates were not publicly retrievable at freeze. No Max Planck electron-density result is claimed.

Extension in progress

Open the non-confidential evidence report

Challenge LINUR

A10Showcase with boundary

Validated result

Calibrated pH readout

Calibrated prediction

Client problem: Can a compact impedance readout transfer to closed pH labels after a declared calibration?
Public capability: A calibrated impedance-descriptor transfer under admitted sample and instrument conditions.
Public result: Five closed pH labels after four-record calibration: MAE 0.1199 pH and RMSE 0.1606 pH.
Evidence boundary: pH keeps its standard definition. This is not a proton-activity derivation, universal chemistry model or first-principles pH result.

Challenge LINUR

A12Research / mixed

Reduced interpretable result

3,030-nuclide isotope table

Calibrated reduced prediction

Client problem: Can a frozen reduced model compare mass, stability, lifetime and decay mode across a large admitted table?
Public capability: Element-blocked calibrated readouts across 3,030 nuclides and 92 elements.
Public result: Binding MAE 58.92 keV/A; stability AUC 0.9041; half-life MAE 2.4455 log₁₀ s; decay-mode accuracy 83.24% on 2,762 labels.
Evidence boundary: Mixed result. The zero-fit physical mass, stability, Q-value and lifetime lane remains unavailable; no claim of predicting isotope lifetimes from geometry is made.

Challenge LINUR

Illustrative antenna aperture and directional field topology

Illustrative image · not evidence

02 · Application domain

Energy & Fields

Field geometry, propagation, energy transfer and control are joined to the evidence each physical interface permits.

02Energy & FieldsPublic capability and evidence boundary

A04Strong showcase

Validated result

Passive wire antenna

Held-out computational comparison

Client problem: Can a frozen passive-antenna operator agree with an independent solver on held-out cases?
Public capability: Passive wire-antenna observables under a fixed computational comparison.
Public result: Against independent NEC2++: 0.799 dB mean S11-curve RMSE, 10.42 MHz resonance MAE, 0.0269 dB forward-gain MAE and 0.286 dB front/back MAE.
Evidence boundary: Zero target-specific fits. This is a computational cross-solver comparison—not antenna hardware validation.

Challenge LINUR

A02Showcase with boundary

Reduced interpretable result

Battery charging

Calibrated reconstruction

Client problem: Which declared charging protocol reduces simulated charge time while respecting the stated thermal constraint?
Public capability: Protocol screening in a conventional PyBaMM DFN/P2D battery model.
Public result: Eight held-out discharges: 68.86 mV voltage RMSE and 1.55 °C temperature RMSE. Selected protocol reduced simulated charge time 17.47% under the declared 45 °C constraint.
Evidence boundary: Conventional DFN/P2D model only; no new battery microscopic physics, hardware, charging-safety or lifetime authority.

Challenge LINUR

A08Showcase with boundary

Validated result

Phase-shifted DFB laser

Locked historical reconstruction

Client problem: Can a locked model reconstruct a published linear threshold without target-specific fitting?
Public capability: A certified linear-threshold reconstruction under declared prior knowledge.
Public result: QWS threshold 19.81556 mA versus the published 19.78 mA: 0.1798% relative difference.
Evidence boundary: Historical reconstruction—not a new blind experiment. No linewidth, thermal, spontaneous-emission, above-threshold power, fabrication or hardware authority.

Challenge LINUR

A11Showcase with boundary

Reduced interpretable result

Eight-port ECRH control

Reduced open-loop control

Client problem: Can an inverse controller expose an infeasible request before returning the largest admissible action?
Public capability: Headroom-aware global scaling in a declared reduced open-loop control problem.
Public result: A 10% request is infeasible inside 0.8% symmetric headroom; exact global saturation returns 0.8%. A declared 20% authority envelope delivers the full 10% request.
Evidence boundary: Reduced open-loop result only. No plasma deposition, MHD, fast-particle, divertor, hardware, delay/noise or closed-loop robustness validation.

Challenge LINUR

A06Research / mixed

Reduced interpretable result

Wireless power transfer

Reduced numerical optimization

Client problem: Can a reduced optimization identify promising geometry before a higher-fidelity model or physical test?
Public capability: A reduced numerical comparison across 120 declared cases.
Public result: Median reduced-model efficiency rose from 2.52% to 6.16%, a median relative uplift of 77.8%.
Evidence boundary: Research result only. Segment convergence is insufficient for an absolute continuum-efficiency claim; no skin/proximity, ferrite, shield, tissue, thermal, EMC, power-electronics or hardware authority.

Challenge LINUR

Illustrative graphene lattice and material-scale paths

Illustrative image · not evidence

03 · Application domain

Materials & Devices

Conventional physical models are computed under frozen certification, with discovery and device claims kept separate.

03Materials & DevicesPublic capability and evidence boundary

A01Showcase with boundary

Reduced interpretable result

Graphene / Floquet materials

Certified conventional-model computation

Client problem: Can a large conventional lattice calculation remain governed across a declared case set?
Public capability: Exact-grid Dirac/Floquet/Haldane cases under LINUR certification.
Public result: 375 exact-grid cases: 276 certified and 99 failed closed; largest admitted lattice 99,900 sites.
Evidence boundary: Conventional Dirac/Floquet/Haldane physics under LINUR certification—not experimental validation, a new microscopic law, a quantum-computer replacement or a comparative speed claim.

Challenge LINUR

Deep field of a galaxy cluster with gravitational-lensing arcs

Illustrative image · not evidence

04 · Application domain

Cosmic Systems

Reduced cross-survey comparisons are kept distinct from new-gravity, lensing and strong-field authority.

04Cosmic SystemsPublic capability and evidence boundary

A13Research / mixed

Reduced interpretable result

Galaxy, cluster and expansion data

Calibrated reduced comparison

Client problem: Can one frozen reduced comparison be audited across galaxy, cluster and expansion cohorts?
Public capability: One-scale calibrated mappings across 175 galaxies, 47 clusters and 32 chronometers.
Public result: Mixed science: conventional RAR slightly beat the closure-shaped mapping in both radial cohorts, and the expansion deformation did not beat flat ΛCDM.
Evidence boundary: No new-gravity claim. Weak lensing, BAO/SN distance observables, strong-field and black-hole predictions remain unfinished.

Challenge LINUR

Technical evidence boundary

Useful, unfinished and negative results stay distinct.

Reduced results remain visible as reduced results. Missing operators remain extensions in progress. Only results that missed their declared physical gate are labelled negative.

A03Materials & Devices

Negative result

Ni-63 betavoltaic

Physical-prediction attempt

Recorded outcome

The measurement-free device prediction overestimated measured output by 8.018×.

Public boundary

No admitted physical device-performance or full-3D transport claim. An earlier exact match depended on one calibrated coefficient and is not presented as a prediction.

A05Energy & Fields

Negative result

Steerable antenna array

Held-out computational comparison

Recorded outcome

32 cases: beam error 19.25° versus 0.25° in NEC; pattern RMSE 6.32 dB. Only the front/back gate passed.

Public boundary

No array or hardware claim. The result is retained as an explicit missing-operator and problem-definition finding.

A07Matter & Life

Extension in progress

CO₂ photoreduction

Capability audit

Recorded outcome

Execution stopped before target data were opened because required adsorption, reaction-coordinate, solvent, photochemical-transition, kinetic, selectivity, yield and stability operators were unavailable.

Public boundary

No CO₂ conversion claim. This is a transparent no-go result, not a successful application.

A09Matter & Life

Reduced interpretable result

Six-anchor electrochemical impedance

Reduced calibrated reconstruction

Recorded outcome

Best admitted lane reached median held-out complex NRMSE 0.3196; the stricter lane rejected all 519 spectra.

Public boundary

Six measured anchors per spectrum are required. No measurement-free chemistry, microscopic uniqueness or pH authority.