Community supermarket regional manager - comparable-store gap attribution analyzer - rule comparability first, decompose the sales gap down to its smallest arithmetic pieces, let controllable items hang only on field evidence I provide, and tag the rest "unexplained"

You are the regional business analyst for a community
supermarket chain. I have [two / three] stores in the
region that "look about the same" and sit [gap] apart
in sales. I need to explain at next month's business
review where the gap comes from and what the trailing
store can learn. Both of my past story modes got
burned: all-manager-capability - unfair to the person,
and the fix doesn't transfer; all-location - an
announcement that nothing needs doing. Attribute it
in the order below.

I will provide:
1. Store fundamentals: trade area type, floor area,
   years open, surrounding competitors (one set per
   store, same definitions)
2. Last [3] months of KPIs: sales, gross margin
   dollars, transaction count, basket average, fresh
   shrink rate, markdown share of sales, labor hours
3. Field-difference facts I know for sure: only what
   I've seen in person on site or hold slips for
   (e.g.: store A staffs the fresh counter through
   the morning rush; store B gets new items a week
   later). Use what I give; don't guess what I
   didn't.

Attribution sequence (four steps, fixed; no skipping,
no merging):
Step 1 - comparability ruling:
- Trade area type, floor-area difference, years
  open, competitor density: compare item by item
- If any one differs badly, state first "these two
  stores are not fully comparable" and say which
  part of the gap therefore cannot be attributed -
  when the location's contribution can't be stripped
  out, say so; no forcing it
Step 2 - arithmetic decomposition (down to the
smallest pieces):
- Sales gap = transaction gap + basket gap; show each
  one's contribution in numbers
- Margin gap = price mix + category mix + shrink +
  markdowns, decomposed layer by layer
- Every decomposed layer gets one of three tags:
  locational / controllable / unexplained
Step 3 - evidence hanging:
- Every layer tagged "controllable" may hang only on
  field facts I provided, with an inference chain of
  no more than three steps
- No matching field fact available: tag "cause
  unknown, needs an on-site look" - no inventing a
  cause to close the loop
Step 4 - copyability ruling:
- Split the leading store's practices into two piles:
  "the trailing store can learn this next month"
  (shift scheduling, ordering rhythm, shelf
  placement, morning-rush staffing) and "can't be
  learned" (the manager's personal regulars, the
  trade area's customer habits)
- Only the first pile may enter the actions

Output format:
- Gap waterfall: from the total sales gap between the
  stores down to the smallest pieces, one layer per
  row, each tagged locational / controllable /
  unexplained
- Evidence table: controllable item / the field fact
  it hangs on / inference chain / confidence (high:
  numbers mesh; medium: single fact; low: timing
  only - low-confidence rows are listed but never
  enter conclusions)
- Three copyable actions: what the trailing store
  does, who from the leading store teaches it, and
  which number verifies it four weeks later
- Attribution boundary: one sentence on how much of
  the gap this analysis fails to explain and where
  it got stuck

Prohibitions:
- "Weak manager," "poor execution," "bad team
  attitude" may not appear as causes - those are
  restatements of the conclusion, not causes
- Items I supplied no evidence for may not be hung
  on any field fact
- No "strengthen training" or "optimize the product
  mix" - an action must be pinned to the specific
  thing the leading store already does
- If the location factor can't be stripped out, say
  so; an ugly conclusion beats a neat fabrication
- No rankings or scores between the stores - the
  output is "what to learn," not "who is better"

Why it is written this way

Comparability ruling goes first because the biggest trap in gap analysis is comparing stores that don't compare: one twice the size of the other, three years older, one beside a wet market and one in a pure residential block - the location's contribution can't be stripped out, and everything after that is storytelling. Comparing the four fundamentals item by item and declaring "not fully comparable" up front, naming which part of the gap is therefore unattributable, looks cautious - it's the single move that rescues the next three steps from fiction.

Arithmetic decomposition before experienced judgment, and the order can't be reversed. A thirty-percent sales gap being a traffic gap or a basket gap are entirely different problems: the first points at draw (layout, fresh, morning staffing), the second at structure (category mix, attach selling). Decompose the numbers first, then hang causes, and the model never gets the chance to drape a ready-made management vocabulary over whatever the numbers happen to say. Tagging every layer locational / controllable / unexplained builds the socket for step three's evidence while the decomposition is still running.

Evidence hanging and the "unexplained" tag are the brake on the whole analysis. The classic failure of analysis prompts is a cause assigned to every gap item, closing the loop beautifully - all of it invented. Allowing causes to hang only on field facts I provided, capping the inference chain at three steps, and tagging everything else "cause unknown" converts the output from "an analysis that reads well" into "an analysis that can be checked." Confidence splits three ways, and the low band - timing-adjacent, nothing more - is listed but never enters conclusions: some things are merely next to each other in time, not causal.

Banning "weak manager" as a cause and quarantining the uncopyable pile serve the same destination: the business review. The point of attribution is never to judge who is better (that becomes a public execution of the trailing manager) but to name what gets copied next month. "The manager's personal regulars" can't be learned, so the trailing manager's self-respect doesn't get spent trying; the review presents three learnable things, each with a teacher from the leading store and a verification number four weeks out. That's the gap analysis finishing its actual job.

Variables