What is a self-healing selector?
A self-healing selector is one that locates its element again after the original expression stops matching. It retains several independent signals about the element, among them a visible label and the field beside it. A recovery is recorded as an event, so a person can confirm the element chosen was the right one.
Why recovery has to be recorded as well as attempted
Self-healing is easily heard as a claim that the automation never fails, which is the wrong reading. The mechanism narrows one specific failure, in which an element has moved or been renamed while remaining the same element. What matters in an answerable system is the record it leaves, because a recovery nobody reviews is an agent quietly choosing a different field. Reviewed on a report, the same mechanism reduces the maintenance that selectors otherwise demand.
- Most element failures are a rename rather than a removal.
- A second signal distinguishes a moved element from a missing one.
- Recovery without a record lets an agent act on the wrong field unnoticed.
- Maintenance effort falls when a cosmetic release no longer halts the work.
How a self-healing selector works
- Several attributes of each element are captured when the selector is first written.
- The primary expression is attempted on every execution as normal.
- On failure the remaining signals are scored against the candidate elements present.
- A candidate above the confidence threshold is acted upon and the recovery recorded.
- A candidate below it halts the work and raises an exception for a person.
- Confirmed recoveries are promoted into the stored selector for later executions.
How a self-healing selector is evaluated
A self-healing selector is assessed on the elements it chooses when it heals. Examine the following over successive application releases:
- share of selector failures recovered without intervention
- recoveries later judged to have chosen the wrong element
- confidence reported on incorrect recoveries
- proportion of recoveries reviewed by a person
- maintenance time per release against a static selector baseline
Limitations, risks and common failure modes
- A plausible neighbouring element is chosen and the data lands in the wrong field.
- Recovery hides a deliberate application change the process owner should have seen.
- Confidence thresholds are relaxed to reduce exceptions, which admits wrong matches.
- The recovery report is produced and nobody is assigned to read it.
- Healing is assumed to cover removal, where the element has genuinely gone.