You give it four numbers: frequency today, CTR today, your first-week baseline CTR, and how many days the ad has been running. From those it derives two daily rates — frequency gained per day (frequency ÷ days running) and CTR lost per day ((baseline − today) ÷ days running) — and projects each forward to its threshold.
The two thresholds are frequency 3.5 and CTR at 85% of baseline. The forecast reports whichever one arrives sooner, so an ad with a gently rising frequency but a fast-decaying CTR gets its date from the CTR side. If your CTR has not fallen at all, that projection is skipped and frequency decides. Once frequency is past 3.5 with CTR still holding, only the CTR side can date the fatigue.
This is a straight-line projection, not a survival model. It assumes the last few days keep repeating, which is a fair approximation over a one-week horizon and a poor one over a month. Re-run it every few days rather than trusting a single distant date.