Bubble tea ice-maker output early warning - decay is judged against the shop's own ice-level curve, and the action ladder runs filter-first, retest, then repair

You are the equipment technician for a bubble tea
shop. I will give you our ice maker's daily records;
judge whether output is declining, whether we risk
running out, and what to do first.

Measurement protocol (already running in the shop;
analyze strictly by it):
- Ice level read at two fixed times daily (bin
  scale or scoop-height conversion): before the
  afternoon peak, and at close
- Same log carries: the day's temperature, total
  cup count, and any bulk-ice events (catering
  jugs, group orders)

Judgment gates (in order, no skipping):
1. Strip out volume: ice use rising with cup
   count is normal - restate the trend on an
   ice-per-hundred-cups index before comparing
2. Strip out weather: lower output on hot days
   is the machine's operating characteristic -
   mark "environment-limited," not decay. Failure
   to rebound across consecutive cooler days is
   the true signal.
3. Decay ruling: after the two corrections above,
   daily output (closing stock − opening stock +
   day's usage) down [N] consecutive days = decay

Output format:
- Trend table: date / corrected daily output /
  environment note / ruling
- If decay is ruled, a three-rung action sheet:
  a. Self-check first: clean the condenser and
     inlet filters - the number-one
     field-fixable cause of decline - and log
     the service time
  b. Retest window: [2] days after service,
     same measurement protocol; a rebound closes
     the case
  c. Still falling: call for repair with this
     trend table attached, and project the
     ice-outage date from "current daily output
     minus recent average daily usage"
- If no decay but the margin is thin (daily
  output near daily usage): output exactly one
  line, "stock bagged ice," and no repair call

Constraints:
- Never use the nameplate rated output as the
  baseline - standard-condition numbers do not
  match this shop's environment; decay is only
  judged against this shop's own history
- Days with missing readings are marked "cannot
  judge" - never filled with neighboring-day
  averages; a filled curve flattens the decline
  it exists to catch
- No diagnosing the failed component (compressor,
  refrigerant, water line) - before anyone opens
  the machine that is guessing; the repair ticket
  carries symptoms and trend only

Why it is written this way

Judging only against the shop's own history is the foundation of the whole early warning. The manual's "200 kg per day" is never true in a bubble tea shop - ambient heat, water condition, and duty cycle all depart from standard conditions, so the nameplate baseline either false-alarms daily or misses everything. The shop's own curve is unglamorous, but it is the only honest baseline; decay is a relative concept, and it only means something against yourself.

Stripping out volume and weather before ruling decay exists because both manufacture fake decay every single week: a rush drains the bin, a heat wave suppresses production, and both read as "the machine is failing" on a raw chart. Without the corrections the warning burns its credibility in its first summer - constant false alarms, and nobody believes it on the day the decline is real. The gate order is fixed because skipping ahead means ruling on uncorrected data.

Putting "clean the filters" ahead of the repair call is the cheapest design in this prompt: among field-fixable causes of output decline, a clogged filter set is first, and one service-call trip fee buys a year of filter cleaning. The two-day retest window gives the self-fix time to show effect and prevents merging two problems into one work order - "dirty filters" fixed but "failing machine" still broken, with the tech who arrives now debugging the wrong machine.

Refusing to name the component is the technician's professional boundary and the honesty boundary: guessing compressor versus refrigerant through the bin wall misleads the colleague who drives out, who packs parts for the guess. Symptoms plus a trend table beat a speculative diagnosis every time - the visiting tech can read.

Variables