Hydrogels detailsHydrogels exposes five-model quality prediction, Class-4 reverse optimization, local sensitivity, reviewed feedback learning, staged retraining, promotion, rollback, and state inspection through the shared Module protocol for objective O4.

adapter-hydrogels

Hydrogels

Hydrogels exposes five-model quality prediction, Class-4 reverse optimization, local sensitivity, reviewed feedback learning, staged retraining, promotion, rollback, and state inspection through the shared Module protocol for objective O4.

Objective: O4
Workspace: Internal alpha
Operations: 7

Capabilities

Predicts one to 500 eight-material compositions with AUTO, CORAL, CORN, RF, HGB, or XGB and exports Parquet, CSV, and JSON results.
Optimizes feasible Class-4 mixture regions and reports an optimal composition plus model-agnostic local probability sensitivity.
Captures feedback as pending or held, requires explicit admin approval, and retrains all five models into immutable staged packs.
Lets authorized admins inspect metrics and fingerprints, promote a staged pack, or roll back by promoting a previous compatible pack.

Best used for

Evaluate candidate hydrogel compositions before running a physical experiment.
Batch-score formulations from browser-parsed eight-material CSV files without adding an upload endpoint.
Capture validated experiment outcomes while preserving a suspicious-review lane for follow-up analysis.

Constraints

The v1 runtime accepts the eight supported material components; the additional upstream training column remains outside the public contract.
Reverse optimization remains an audited sampled Monte Carlo Class-4 workflow, not a certified global optimizer.
Local sensitivity is a model-agnostic perturbation result and is neither SHAP nor causal attribution; model binaries remain private.

Admin features

review_hydrogel_feedback
retrain_hydrogel_models
promote_hydrogel_model_pack
inspect_hydrogels_state

Protocol surface

Operations

predict_hydrogel_quality
optimize_hydrogel_composition
submit_hydrogel_feedback
review_hydrogel_feedback
retrain_hydrogel_models
promote_hydrogel_model_pack
inspect_hydrogels_state

Source

This module ships with the mODa platform and is maintained by the platform team. Turn on developer mode in Settings to see the source locations behind it.

Capabilities

Predicts one to 500 eight-material compositions with AUTO, CORAL, CORN, RF, HGB, or XGB and exports Parquet, CSV, and JSON results.

Optimizes feasible Class-4 mixture regions and reports an optimal composition plus model-agnostic local probability sensitivity.

Captures feedback as pending or held, requires explicit admin approval, and retrains all five models into immutable staged packs.

Lets authorized admins inspect metrics and fingerprints, promote a staged pack, or roll back by promoting a previous compatible pack.

Constraints

The v1 runtime accepts the eight supported material components; the additional upstream training column remains outside the public contract.

Reverse optimization remains an audited sampled Monte Carlo Class-4 workflow, not a certified global optimizer.

Local sensitivity is a model-agnostic perturbation result and is neither SHAP nor causal attribution; model binaries remain private.

Protocol surface

Operations

predict_hydrogel_quality
optimize_hydrogel_composition
submit_hydrogel_feedback
review_hydrogel_feedback
retrain_hydrogel_models
promote_hydrogel_model_pack
inspect_hydrogels_state

Source

This module ships with the mODa platform and is maintained by the platform team. Turn on developer mode in Settings to see the source locations behind it.