Read exposure history and new map
Synthetic revision R1 assigned scores 0–49 to A,50–99 to B. R2 changes A to 0–24 and B to 25–99. Local experiment keeps a unit’s first recorded exposed variant for its current episode. U1 score 30 exposed A, U2 score 70 exposed B, U3 score 30 never exposed. No random assignment or real traffic claim is made.
| Unit | History | Next recorded assignment |
|---|---|---|
| U1 score 30 | A under R1 | A retained |
| U2 score 70 | B under R1 | B retained |
| U3 score 30 | No exposure | B under R2 |
Separate existing units from new entrants
- Look up episode exposure by unit
- Retain U1/A and U2/B
- Use R2 only for U3
- Flag proposed U1 crossover
A deliberate new episode needs its own episode identity and policy; do not silently reset an existing one.
How many existing units would change incorrectly if R2 were reapplied blindly?
Show the worked answer
One, U1, from A to B. U2 stays B; U3 legitimately enters B because it has no exposure history.
Reject unintended variant crossover
| Case | Decision or result | Reason |
|---|---|---|
| Recompute U1 from score 30 only | Reject crossover | Recorded first exposure A |
| Apply R2 to unexposed U3 | Allowed locally | New entrant |
Record the retained assignment source, not merely the latest map revision. Allocation edits do not rewrite earlier exposure events.