The generated interpretation does more than recognize an event. Its resolution path can create or update problems, connect them with Impacts or Fixed_Onto, maintain Near lineage and propagate probability values used by potentialities.

if node.type != Sensor or node.status != ACTIVE: return
configuration = switch sensor subtype:
    quantitative | dedicated | media | knowledge -> build CEP
    messaging, message processing = LLM         -> build ML
    messaging, message processing = Keywords    -> build CEP (one rule per keywords row)
    messaging, message processing = None        -> nothing (messages stored)
deploy(configuration)
on detection:
    resolve affected model element
    update impact/fix relations and probability

• Probability ladder — when several thresholds give a probability to the same risk, they share one runtime node and its probability follows the measured value step by step; going back under an intermediate threshold lowers it to the previous step, only the first threshold removes the risk. The shared node carries the runtime key _potentiality_ladder and its probability is a cascade of ifThenElse on the measured value.
• Nature of the risk — a potentiality raised by a threshold keeps the nature of its reference (Emerging, Intrinsic…): without it, the objectives offered to the deduction would ignore it.
• Release with hysteresis — three optional columns delay the release: released below (MAX threshold) or released above (MIN threshold), and released after (h). What the threshold activates is then released only once the value has stayed past the release value for that many hours, with a SUSTAINED_THRESHOLD deactivation instead of the crossing back. A risk the same threshold gives a probability to is not removed with what it activates: at the release it is published again with the probability of the measure, since an actuality and the risk that generated it are never active together.
• SUSTAINED_THRESHOLD trigger — fires on a reading taken at least durationMillis after a first reading past the value (staysBelowValue or staysAboveValue, exactly one), with no reading of the same sensor on the other side in between. The duration is measured on the dates of the events, so a replayed scenario follows its own clock (CEPRuleDefinitionCompiler.compileSustainedThreshold).

MAX 2.50 m -> George-V bridge flood risk 0.2
MAX 3.50 m -> George-V bridge flood risk 0.5
MAX 4.00 m -> George-V bridge flood risk 0.8
MAX 4.50 m -> George-V bridge flood risk 1
MAX 4.50 m -> flood, released below 4.00 m after 6 h
# rule R5: activation THRESHOLD_TRANSITION crossesAboveValue 4.5
#          deactivation SUSTAINED_THRESHOLD staysBelowValue 4.0, durationMillis 21600000

The flood thus stays active while the Loire recedes from 4.50 m to 4.00 m and ends six hours after it went below 4.00 m; a reading back above 4.00 m in between starts the count again. The probabilities of the risk are covered in the 2L language reference, which sets the low tier of the same risk.

• Use-case clock — a connector sends nothing while the clock of the use case is not at the server time: values dated now would not belong to the time of the use case. The connection is then SUSPENDED and resumes when the clock is back at the server time.
• Connection state — every change is published as a dataSourceConnectionEvent (CONNECTED, FAILING, SUSPENDED or DISCONNECTED, with the date of the last value received). r-iose shows a Connect to … switch and the state under the selected sensor and its forecasts; r-iota raises an alert when a provider becomes unreachable and another one when it answers again. A wrong configuration (no station code, no hashtag) leaves the sensor disconnected with a FAILING state; it never prevents the sensor from starting.
• Message cap — a messaging sensor has a Message Processing section: message processing (LLM / Keywords / None), limit messages (default Yes) and messages per minute (default 5). With the limit on, beyond that number of messages in a sliding minute the messages received by the sensor are ignored, whatever channel they come from; a change of the value applies at once.