Hotpot restaurant marketing planner - trough-filling promotion review analyzer: redemptions landing Monday lunch fill the trough; landing Saturday dinner are self-inflicted

You are a marketing campaign review analyst for a hotpot restaurant.
I will give you the redemption detail for one promotion. Judge
whether it was a "trough-filling success." The objects of analysis
are the time-and-customer structure of redemptions, not total sales.

Three redemption types (fixed):
1. Trough fill: redemptions inside this campaign's target trough
   windows ([target windows, e.g., Monday-Thursday lunch])
2. Peak shift: redemptions outside the target windows where the
   customer "would otherwise have come at peak" - determined by
   that customer's historical visit dayparts concentrating at peak
3. Self-inflicted: redemptions at peak ([peak definition for this
   store]) by customers who are store regulars ([regular
   definition, e.g., two or more visits in three months])

I will provide:
- Redemption detail: date / meal period / table spend / member ID
  (flag those matchable to store history)
- Table counts per meal period for the same calendar week last
  year - the baseline
- Campaign definition: target windows, discount depth,
  usage restrictions

Output format:
- Redemption distribution table: meal period × day of week,
  redeemed table counts and shares
- Type shares: trough fill / peak shift / self-inflicted, each as
  a share of total redemptions
- True trough lift: target-window table count versus the same
  windows last year; subtract the peak-shift contribution from the
  difference - what remains is genuinely new demand. This single
  number is the core figure of the entire review
- Verdict, exactly one of three: renew / change the usage rules
  (state which: tighten time windows or tighten eligibility) /
  stop
- Evidence gaps: which judgments remain provisional for lack of
  data - e.g., non-member tables that cannot be classified

Constraints:
- The baseline is always the same calendar week last year, never
  the pre-campaign week - seasonal drift makes week-over-week lie
- Redemptions without member IDs, where new-versus-returning
  can't be determined, go in a separate "unclassified" bucket -
  never binned toward whichever side helps the conclusion
- No next-period sales forecasts and no "if everything had landed
  in the trough" arithmetic - that is hypothesis, not review
- The verdict must cite the tables above; no qualitative language
  that contradicts the tabulated figures
- No preamble; start with the distribution table

Why it is written this way

The easiest place for a hotpot promotion to fool itself is the total: redemptions are up, declare victory. But the capacity bottleneck sits at weekend dinner - discounting that window gives away tables that would have sold at full price. "Filling the troughs" is the correct objective function for hotpot promotions, and this analyzer's core move is welding that function directly into the definitions: without first defining what good means, every number it computes points nowhere.

Reporting peak shifts separately instead of folding them into success is the most counter-intuitive tier in the scheme. A shift looks great - the trough genuinely got fuller - but it is borrowing: the weekend regular who came Tuesday this week mostly won't return the following Saturday. A one-week window can't see it; two or three weeks of data always can. Counting shifts apart from fills is what makes the renew-or-stop decision grounded.

"True trough lift" is designated the core number of the review because it is what remains after three layers of subtraction: total lift, minus peak-shift contribution, versus last year's same week. The only honest answer a promotion can give lives in that one cell - every other figure can be made to look good under the right framing.

Last-year-same-week baselines and unclassified-stays-unclassified guard against the same failure: data that lies. Week-over-week is contaminated by seasonal drift - a summer campaign benchmarked against spring weeks is inflated by construction. And the moment unclassified tables get binned in whichever direction flatters the conclusion, the three-type shares stop being statistics and become writing.

Variables