What do the Acatian reference implementations prove?

They prove reproducible execution across three distinct process families, numerical balance closure against published tolerances, explicit boundary evaluation, sensitivity generation and artifact integrity. They do not prove customer-specific predictive accuracy.

Can the reference results be independently checked?

Yes. Every model publishes its input table, dynamic trajectory, sensitivity runs, acceptance criteria, streams, schedule, comparison data, machine-readable summary and SHA-256 manifest. The packages can be regenerated with the repository script.

Is this a GMP-validated industrial model?

No. It is deliberately labelled as a synthetic software benchmark. Industrial validity requires customer-owned calibration data, an approved comparison model, site-specific equipment and schedule evidence, and qualified review.

Reference model catalog · three process families

Process models you can inspect, reproduce and challenge.

Industrial fermentation, mAb biopharma and cultivated biomanufacturing now publish their assumptions, full time series, numerical balance closure, physical boundaries, schedules, comparison envelopes and sensitivity runs. They demonstrate software execution and traceability, not customer calibration or GMP validation.

Open editable model Download methodology

Catalog version: 2.0.0
Solver: RK4 · 3,360 steps
Reported states: 169 time points
Balance residual: 2.67 × 10⁻⁷%
Acceptance: 7 / 7 passed
Model state: Screening · 3 models

00 · Process families

Three transparent engineering baselines.

Every package contains declared inputs and intervals, complete trajectories, balance checks, sensitivities, comparison data, streams, finite-capacity scheduling and checksummed artifacts.

AX-REF-FERM-FB-001 · SCREENING

Industrial fermentation

100 m³ microbial fed-batch with explicit biomass, substrate, product, oxygen, feed and working-volume balances.

Mode: Fed-batch
Trajectory: 169 points
Checks: 7 / 7

AX-REF-MAB-FB-001 · SCREENING

mAb biopharma

2,000 L CHO fed-batch with dynamic cell, substrate, product, dissolved-oxygen, lactate and ammonium states.

Mode: Fed-batch
Trajectory: 169 points
Checks: 7 / 7

AX-REF-FOOD-PF-001 · SCREENING

Cultivated biomanufacturing

20,000 L perfusion model with cell retention, bleed, nutrient exchange, oxygen and metabolite boundaries.

Mode: Perfusion
Trajectory: 169 points
Checks: 7 / 7

Final titer0.981 g/Lreactor inventory

Batch product1.726 kgbefore DSP recovery

Peak VCD12.3310⁶ cells/mL

Max ammonia1.858 mM2.0 mM boundary

Max lactate1.976 g/L2.0 g/L boundary

01 · Model scope

One connected upstream benchmark.

Defined boundary: raw media through reactor harvest. Downstream operations are represented as open handoff boundaries and are not used to claim final drug-product yield.

T-101Media preparation1,440 L initial charge

F-103Sterile filtration0.2 µm boundary

BR-201Seed expansionsynthetic inoculum

BR-3012,000 L fed-batch336 h dynamic run

CF-401Harvest handoffopen DSP boundary

INBasal medium

INConcentrated feed

INGas transfer via kLa

OUTHarvest broth

OPENOff-gas composition

02 · Executable result

Inspect the complete 336-hour trajectory.

03 · Acceptance

Passes are machine-readable; gaps remain visible.

Each criterion has a threshold, an observed value and a review status. Nothing is silently rounded into compliance.

04 · Reproducibility package

Download the model evidence, not a screenshot.

Every generated artifact is listed with a SHA-256 checksum in the manifest.

InputsValues, intervals, units and evidence classCSV Dynamic profile169-point state trajectoryCSV Sensitivity18 low/base/high solver runsCSV AcceptanceThreshold, observation and statusCSV ComparisonObserved values, screening envelope and statusCSV Stream tableInlets, process outputs and open boundariesCSV ScheduleResource occupancy and turnaround basisCSV SummaryMachine-readable headline resultsJSON ManifestVersion and SHA-256 integrity checksJSON

What this proves

Executable, traceable software behavior.

The same inputs regenerate the published trajectory.
The numerical component balance meets its stated tolerance.
Physical screening boundaries are evaluated explicitly.
Parameter variation produces inspectable sensitivity outputs.
Artifacts can be independently checked against the manifest.

What remains to be proven

Industrial predictive validity.

Calibration against customer-owned bioreactor and assay data.
Independent comparison with an approved reference model.
Validated downstream recovery and product-quality models.
Site-specific equipment, schedule, utilities, TEA and LCA evidence.
Qualified review for GMP, safety and investment decisions.