Industrial AI stacks fail when models chase chatter and spikes that never should have entered the feature path. GSRF sits before the model as a soft thermostat—not as a replacement for detectors or controllers.
Where it lives in the stack
- Sensor / historian stream (positive magnitude)
- GSRF Practical — mid-band calm + peak compression toward normal
- Feature store / model / dashboard
- Separate detector path if you need fixed-FA events
Why this is not slideware
- Mid-band: ~+35% osc31 / ~−70% spectral sweet band on locked packs
- Peaks: ~38% mean peak-dev; 49/52 wins
- Red: delay command, fixed-FA residual, MAD tracking
Buyer language (claim-safe)
“Calm the signal first. Reduce nonsense the model never should have seen.” Do not promise ROI or universal FA reduction.
Verdict
Primary commercial path for industrial buyers. Try: /try. Longer note: Pre-AI sensor hygiene. Pricing: /customers.
Sources: Evidence · customers industrial card · pre-ai note · claim ladder