A changed screen can invalidate an old visual reference
A moved control, notification badge, revised panel or different lighting can alter the pixels used to recognise a screen. Small references are useful for stable interface controls, but can become ambiguous when the surrounding content changes.
In one VRA development case, a notification badge reduced the match for a City report control. The workflow and completion checker then disagreed about whether the farm had returned to City. A focused badge-aware reference and consistent interpretation addressed that specific case; it was not evidence that every navigation failure was fixed.
A correct screenshot can still lead to stale input
The camera may be moving after Go, an animal may walk, or a panel may still be closing. Between observing a location and the game receiving an input, the target can change. Shorter fixed waits do not by themselves make an action faster or more reliable.
VRA testing exposed a nominal blank-map tap during camera motion followed by a resource collection marker. The observation-only validation mode was tightened to prohibit map-focus and target inputs. The lesson is to validate the action boundary, not trust the name assigned to a tap.
Different farms reveal different game states
A reward can be ready on one account and already claimed on another. Training can be waiting for collection, a queue can already be occupied, or a building can have a prerequisite. The visual route may look similar while the valid action is different.
Compare the selected options and starting state before attributing every difference to image recognition. No-work, game-blocked, failed and uncertain outcomes should remain distinct; none is interchangeable with verified productive work.
An evidence-first troubleshooting checklist
Stop competing input before investigating, preserve the first unexpected screenshot and identify whether any effect may already have occurred. Then retest the smallest relevant action only when repeating it is safe.
- Record the farm, game version, action and selected settings.
- Find the first screen that differs from the expected route.
- Compare a before-and-after capture with the recorded input.
- Check existing queues, pending collection, active marches and interruptions.
- Distinguish an unreadable counter from an action that actually failed.
- Do not replay an uncertain purchase, claim or dispatch merely to get a cleaner log.