Version 1 record
This page describes software that has been retired. It is accurate about Version 1 and is not a description of what runs now — see the current site and the archive.
C-11 Monitor
Status: built
- Code:
services/monitor/— the residual, the persistence rule, the window, the warm-up and the coverage read port each in their own module - Delivered by:
specs/009-control-loop - Covered by:
services/monitor/tests/,tests/integration/test_monitor_warmup.py,tests/integration/test_monitor_scheduler_handoff.pyandtests/acceptance/test_at_02_threshold_breach_triggers_run.py. The threshold quoted below ismonitor.residual.threshold_m_per_sinconfig/local/monitor.json, which is 1.75 today
Responsibility: detect forecast divergence from observations. Owns the failure mode of: over-sensitivity.
What it does
It compares what the sensors measured with what the current forecast said would be measured, and raises a divergence event when the difference is large enough, for long enough, over enough space to mean something.
The residual is computed on sound speed, not on temperature. Sound speed is the derived quantity the system exists to get right, and it responds to temperature, salinity and pressure together; a temperature residual can be large while the quantity that matters is fine, and vice versa.
Why over-sensitivity is the failure it owns
A monitor that fires on every spike turns the control loop into a thrash generator. The rule is therefore explicit: a single spike shall never trigger a run. A divergence event requires the residual to exceed threshold with sustained spatial or temporal persistence.
There is also a declared warm-up period before any divergence event may be raised, because a system that has just started has no forecast worth diverging from, and the first minutes of a run would otherwise fire continuously.
The default threshold is around half a degree Celsius equivalent — order 1.5 to 2 m/s of sound speed — and is tunable per scenario.
What it must not do
It raises requests. It does not invoke the model. The decision about whether a run is warranted belongs to the scheduler, and keeping those two apart is what allows either to be reasoned about.
Requirements: FR-22 to FR-26. Feature: 009.