Zentari

Product 04 · Quantum Materials

Autonomous discovery that cannot outrun its evidence.

Typed candidates, assays, selections and learning records form a loop in which evidence class controls what may be promoted.

Emulated two-batch discovery loop showing evidence-linked selection and a blocked physical claim
FIG 07 Emulated autonomous-loop wiringorigin/main · 3207d44 · placeholder-token toy oracle The 1.8 → 1.1 mM values are emulated fixture outputs—not a peptide, assay or material result.
Open evidence →
Engineering stateEnd-to-end typed discovery-loop wiring
Scientific evidenceEmulated L0 fixture on origin/main
Next gateMeasured assay or physical-material evidence

Traceable selection can remain valid when the physical claim is blocked.

Workflow

Every batch carries its evidence class forward.

The current public result exercises the wiring with placeholder tokens and a deterministic toy oracle.

  1. 01Define

    Programme, endpoint, controls and evidence requirements.

  2. 02Generate

    Candidate records carry identity and provenance.

  3. 03Evaluate

    Measured, simulated or emulated evidence is declared.

  4. 04Select

    Decision records cite exactly which evidence was weighed.

  5. 05Learn

    The next batch records model and evidence change.

  6. 06Gate

    Physical claims require physical provenance.

Actual result

The result is the evidence-aware loop.

It is not a peptide, assay or material discovery.

Emulated two-batch discovery loop showing evidence-linked selection and a blocked physical claim
FIG 07 Emulated autonomous-loop wiringorigin/main · 3207d44 · placeholder-token toy oracle The 1.8 → 1.1 mM values are emulated fixture outputs—not a peptide, assay or material result.
Open evidence →
Batch 1 best
1.8 mM
emulated toy-oracle output
Batch 2 best
1.1 mM
emulated refinement output
Control
9.0 mM
placeholder control in both batches
Evidence class
Emulated
explicitly retained
Physical L3
Blocked
promotion gate operates correctly
Measured records
0
in the public main-branch result

Methods

One construction path, three evidence classes.

Provenance is a first-class input to selection, learning and promotion.

Measured

Requires physical provenance, assay context, controls and retained observations.

No measured result is published here.

Simulated

Names the engine, inputs, model identity and numerical evidence.

Simulation does not inherit measurement authority.

Emulated

Supports deterministic workflow and governance testing.

Cannot support a physical L3 claim.

Branch-only discovery output is not public evidence.

Bring the next claim into contact with reality.