Kept evidence

sample_dt_overview_2024

relevance: fixes the vocabulary the whole survey leans on claim: the model/shadow/twin distinction is a property of data flow, not model sophistication, and most systems called twins are shadows by that definition quote: A crude model with two-way automatic flow is a twin; a exquisite model updated by hand is not.

sample_dt_sync_2023

relevance: the synchronisation theme's only quantitative comparison claim: event-driven synchronisation cut transmitted volume by 71 percent against one-second polling with sub-two-second worst-case staleness, and explicit staleness marking roughly halved operator mistrust incidents quote: a twin that knows it is stale must say so on every read

sample_dt_factory_2022

relevance: the organisational-trust counterpart to the technical synchronisation story claim: scheduling authority was extended station by station on the strength of a published override log showing over ninety-five percent supervisor agreement, and both recorded failures were silent divergences between plant and model quote: treat every silent divergence between plant and model as a severity-one defect, because the operators will

sample_ml_anomaly_2023

relevance: what the twin's state stream is good for once it exists claim: a layered configuration -- a z-score baseline everywhere plus a learned detector only where its false-positive cost is justified -- matched the autoencoder's precision on point faults while staying immune to the baseline corruption that hid the slowest drift from the learned model quote: Run the z-score everywhere as the incorruptible baseline nobody retrains

sample_std_interop_2021

relevance: the integration theme; why pilots die claim: transport and syntax standards can be bought but semantic agreement must be negotiated, and a shared asset-model registry with named owners cut new-asset integration from seven weeks to a median of six days quote: Buy transport and syntax; budget for semantics.