Home · Applications · Feedback delay
Long-tail guide · claim-safe
Feedback delay: when a soft thermostat becomes the wrong tool
Publishes a failure mode. Not a claim that GSRF prevents all runaway loops.
A soft thermostat that compresses peaks and fights mid-band ring can
hurt you if you put it on a command path that only sees measurements after a long delay.
Locked characterization made that explicit.
Locked result (command path)
MAD-to-measured under balanced GSRF Practical was about
+59% / +97% / +106% worse than EMA at delays of 3 / 6 / 10 minutes.
Full write-up: feedback delay note.
Practical prevention rules
- Measure plant/network delay before live command filtering
- Prefer measurement hygiene and shadow mode over aggressive command shaping
- Do not market GSRF as a delay-tolerant controller
- Keep safety interlocks outside the GSRF path
Autonomous control guide ·
When GSRF lost ·
Evidence safety cage
Questions people actually ask
Does GSRF prevent runaway feedback loops?
It is not sold as a universal runaway-prevention controller. The published lesson is the opposite for multi-minute delayed command paths: EMA was better on MAD-to-measured.
What delay breaks GSRF command smoothing?
Locked packs show large MAD-to-measured penalties at 3, 6, and 10 minute delays vs EMA.
What should I do instead under large delay?
Do not use balanced GSRF Practical as the command smoother; keep EMA/control-native tools; use GSRF for measurement hygiene if osc/peaks are the issue.
Request Audit Pack
Try snippet
Evidence
Research