Outputs and ground truth¶

Real plant data is only what a historian recorded. Homeostat returns that view, the truth behind it, and the ground truth that explains both.

Three levels of values¶

Every signal has a level:

• true: sources, process variables, setpoints, modes, controller outputs, valve positions, promoted parameters;
• measured: the final output of an instrument, what a controller reads;
• internal: an intermediate stage of an instrument.

run.truth holds the true signals and run.measured the measured ones. run.observed is the historian view: the measured tags plus each controller's setpoint, mode and output. A signal without an instrument never appears in it, however much it matters, like the fraction of a reactant left in a reactor or a feed's impurity.

Ground truth¶

run.meta holds what real data never comes with:

• the causal graph: every signal and how it affects every other, within the step or through a state;
• the loops and roles: which signals each loop regulates, measures and manipulates;
• the event log: every change, planned or not, when it happened, and the value before and after;
• the timeline: the plan's regimes, transitions and maintenance;
• the trace: every value drawn from a distribution;
• the reproducibility record.

Labelers derive more: when a fault becomes visible and on which signal, how long each loop takes to settle, and which material a lab sample describes. See Labels.

Why it matters¶

Closed loops move the evidence. A transmitter bias leaves the transmitter reading on setpoint and shows on the controller output; a disturbance is suppressed on the regulated variable and shows on the valve; a product property shifts with conditions no instrument measures, and the lab sees it hours later. With the truth and the ground truth alongside the historian view, methods can be evaluated against what really happened, not against what the data seems to say.