A gold signal was already active when your archive began. Where does it belong?

Start with the question your clock is supposed to answer

A question about time since activation needs an activation clock and a definition of the outcome. When reliable coverage starts later, a signal observed at the boundary has already remained in its current state long enough to appear there. Signals that completed before coverage began may be absent. That boundary changes the group available for analysis even if every later record is accurate.

For a time-to-event study, record entry into observation separately from origin time and end time. Statistical methods that support delayed entry count each observation in the risk set only while it is under observation and still eligible for the event. A signal whose origin cannot be verified may belong in a separate descriptive group rather than a model that assumes a known age.

Source notes: [1] Terry Therneau · Methods in the survival package

A toy archive exposes the selection

Imagine coverage begins at noon. Two setups activated at 11:45. One closes at 11:50 and never appears in the captured feed. The other is visible at noon and closes at 12:15. The example is synthetic. Treating the surviving signal as representative of everything activated at 11:45 omits the shorter path. Setting its activation to noon changes the question and shortens its recorded age.

Keep the original activation if it can be established. Enter the setup into the observed risk set at noon for this example. If the study starts instead from the population active at noon, call that its own question and define follow-up from noon consistently. A clearly named landmark analysis can be useful for a terminal user deciding what to monitor next.

Compare with a clean incident cohort

Run a second view using only setups whose activation occurred during reliable capture. Document which identities enter each view, the start-of-coverage rule and the treatment of stop, expiry and missing terminal events. Where bars or alerts are missing, state the gap and avoid assigning a transition to it. A stop that prevents a subsequent TP3 within the original plan needs a distinct terminal status in the analysis.

The two views answer related questions about signals of different ages. A difference between them can arise because quick outcomes were unobservable at the boundary. Use full signal IDs and source timestamps from the terminal to reconstruct membership, and preserve enough detail for a reader to reproduce it.