AutoDIDACTS integration

The repo is the public scaffold. Funding scales it into domain use.

Mech Invariants supports a proposal posture where the evidence protocol already exists as open-source infrastructure, while AutoDIDACTS work extends it into domain-specific turbulent-flow and aerodynamic validation workflows.

Role in the program architecture

A discovery engine may generate candidate invariants using symbolic regression, persistent topology, equivariant operators, reduced-order models, neural surrogates, or future model classes. Mech Invariants provides the common verification interface.

Discovery engine proposes candidate invariant

Mechanistic Invariant claim object is emitted

Admissible transformations are declared

Cycle-basis or loop tests are executed

Evidence bundle records defects, persistence, and provenance

Claim is passed, failed, replicated, or retired

Public versus private boundary

Public scaffold	Private or controlled adapters
Schemas	Partner data
CLI validation	Controlled aerodynamic benchmarks
Synthetic toy examples	Specialized simulator adapters
Evidence bundle format	Proposal-specific scoring workflows
Claim lifecycle documentation	Export-sensitive or proprietary datasets

Proposal language

The evidence protocol is already instantiated as open-source infrastructure.
AutoDIDACTS funding scales it into domain-specific use.

The open-source scaffold does not claim that an invariant is physically true by
itself. It records the claim, admissible transformations, evidence, defects,
persistence, provenance, and falsification status so candidate laws can be
compared, rerun, challenged, and retired.

Reviewer-risk reduction

Generalizability

Candidate laws are tested across declared transformation loops rather than judged by isolated fit.

Interpretability

Each claim has an explicit quantity, scope, transformation family, and evidence record.

Risk mitigation

High-defect claims become obstruction records instead of disappearing as failed experiments.