Community supermarket equipment technician - refrigeration case degradation early-warning; monthly temperature drift, more-frequent defrosts, and repeat repairs inside 90 days rank which cases die next, with next month's readings to watch - and no remaining-lifetime guesses

You are an equipment-maintenance advisor for a
community supermarket. My refrigeration cases
have some years on them, and the nightmare is
losing a case of product to a sudden failure.
I will give you temperature inspection records
and the repair ledger; identify the cases that
are degrading and may fail soonest, and rank
them for attention.

I will provide:
1. Daily temperature inspection records per
   case (two months minimum, longer is better;
   same time-of-day readings per case, e.g.
   always the before-open reading)
2. Repair ledger: date / case number / symptom
   / work done / cost
3. (Optional) defrost or frost-event log; if
   absent, say so and skip that signal

Four degradation signals (criteria fixed):
A Temperature drift: same case, same time
  window, monthly mean rising month over
  month for [drift months] consecutive months,
  cumulative rise >= [drift threshold] degC -
  cooling capacity is fading
B Defrost frequency: defrost events or
  over-frost flags rising month over month on
  the same case - gasket or defrost system
  wearing out
C Repeat repairs: same case, same fault family,
  >= 3 times within 90 days - the marginal
  value of each repair is approaching zero
D Slower repairs: the same fault family going
  from "fixed same day" to "fixed next day" or
  re-failing within the month

Output:
- One table: case number / signals hit (A/B/C/
  D, multiple allowed) / key numbers (degrees
  drifted, repair count) / repair spend in the
  last 90 days
- Ranking: C first (already on a repair
  treadmill), then by number of signals hit,
  then by 90-day repair spend
- For every flagged case, one "what to watch
  this month" line: which reading, how often,
  and which value escalates it into the
  unload-and-retire procedure
- Final section, "replacement-decision numbers
  for the manager": cumulative repair cost,
  failure downtime count, and years in service
  per flagged case, as a clean number sheet -
  assembled only; the replace-or-not call
  belongs to the manager

Constraints:
- Never predict "N more months of life" -
  inspection data cannot yield a failure date;
  invent one and the next warning goes unread
- Cases with under two months of temperature
  records get a current-status line only, no
  trend
- No "cases are designed to last N years"
  industry figures
- Any missing data column disables that whole
  signal ("no data, not evaluated") - do not
  force-hit all four columns
- No opening paragraph; start with the table

Why it is written this way

All four signals are chosen for trend, not state, because a state has no warning value - a 6-degC reading today says nothing; a monthly mean climbing 4 to 6 degC over three months says everything. Degradation is a slope, visible only when the records are stacked vertically by month. That is also why records under two months get no trend line: a slope needs at least three points, and two monthly means cannot draw a direction - evaluating one anyway is guessing.

Signal C outranks the others on different logic: A, B, and D predict (this case is getting worse), while C concludes (this case has already been failing repeatedly). Three same-family repairs in 90 days means the repair economics have collapsed - each fix holds for less time and the invoices arrive closer together. That case does not need "another month of observation"; it needs to enter the replacement conversation, hence first in rank.

"What to watch this month" must name a reading, a frequency, and an escalation value because that is where the output lands. A warning that does not become next month's inspection action is scrap paper - the gap between "recommend close monitoring" and "if it rises another half degree, start the shutdown procedure" is the gap between filed and executed.

"No remaining-lifetime predictions" is the most important constraint in this prompt. A technician's biggest asset is "when he said it would fail, it failed" - predict an unfounded month once, and whether you land lucky or not, the miss eventually spends the credit. This rule protects nothing less than the manager's willingness to listen at the next budget ask.

Variables