Reaction network model assurance plugin

Edition 9 October 2026 v2. Bounded exact model evidence; low commercial confidence.

Research finding

The newer source claims one eventual compact convex absorber per fixed positive stoichiometric class and fixed positive rates; entry time depends on initial state. Selected formal scope covers older initial-state-dependent all-time bounds. The source affine-family construction has not supplied practical numeric constants here.

Problem and buyer

A chemical-modeling team needs to know whether long simulations are consistent with structural guarantees and whether an edited reaction network still meets the assumptions underlying its analysis.

What the finding could enable

A runnable exact JSON model audit distinguishes effective reactions and rate conventions, checks source scope, and verifies supplied conservation vectors and positive invariant rectangles. Finite bounds come from conventional exact evidence, not from executing the new theorem construction.

Technical and commercial limits

At most eight species, 64 complexes, 128 reactions and degree eight. No SBML/Catalyst import bridge, unit validation, stochastic extension, arbitrary kinetic laws, source proof or industrial model is verified. Failed interval checks are inconclusive, and invariance does not certify entry from outside.

Minimal architecture

Explicit deterministic mass-action JSON and rate convention -> preserve and normalize effective equations -> complex graph and exact stoichiometry -> source-scope annotation -> checked conservation/face inequalities -> retained report. A real simulator or SBML bridge remains future work.

Existing alternatives and differentiation

Catalyst already provides network properties, conservation handling and simulation. NetworkX and SymPy agree with the finite graph/algebra controls, and SciPy matches analytic examples. Potential differentiation is a proven import/equation/witness review workflow, with buyer value still unmeasured.

Monetization hypothesis

Untested AUD 3,000-12,000 integration hypothesis. At an illustrative AUD 7,500 and thirty hours at AUD 180/hour, AUD 2,100 remains before support/overhead; fifty hours cost AUD 9,000 in labor. Demand, avoided-error value and delivery time remain unknown.

Validation experiment

713 passing controls: 702 graph/algebra/interval/schema checks with sixty stored generated models and eleven named fixtures, plus eleven SciPy/analytic diagnostics. Zero-rate arrows, factorial rate scaling, outside-box initial states and a coarse numerical failure demonstrate distinct evidence obligations.

Conditions to reject or defer

Defer numerical certification if effective constants cannot be extracted, and reject broad process-safety claims based solely on a qualitative theorem.

Next concrete action

Keep the specialist module below facility planning. Before a paid plugin, validate one real simulator/import bridge and a recurring permissioned model-review problem, including material errors missed by existing tooling.

Detailed evidence

Exact source scope, rate-convention controls, finite witnesses and conventional comparisons. The Facility Plan Auditor remains the first product experiment.