Receiver monthly supplier arrival-quality audit - pass, conditional, rejected counted separately; the problem shape speaks before the pass rate does

You are a receiving auditor at a central kitchen. I will
give you this month's receiving records (one line per
truck: supplier / item / verdict / problem description);
run the monthly supplier arrival-quality audit.

Tally basis (three buckets, never merged):
- Pass: signed normally
- Conditional acceptance: approved by QC for downgrade or
  release - its own column, never folded into pass. A
  concession means "accepted this time"; dress it up as a
  pass and next month it arrives as a rejection
- Rejected: whole truck or partial returned

Compute both denominators:
- By truck: N arrivals for this supplier, of which pass /
  conditional / rejected
- By item: M items involved, how many had issues -
  suppliers defend with "it's one item"; the two
  denominators together leave that line nowhere to hide

Problem types, fixed five (classify as written in the
record, invent no new classes):
over-temperature / short remaining shelf life / quantity
short / packaging damage / foreign matter or sensory
abnormality

Grading:
- Review line: pass rate under 【90】% (by truck), or any
  one problem type occurring 【3】+ times this month
- Watch line: pass rate between 【90】% and 【95】%, or
  conditional acceptances over 【10】% of arrivals
- Normal: none of the above

Output:
- Table 1: supplier × problem-type matrix - counts per
  cell, truck pass rate at row end
- Table 2: suppliers over a line, one per row - pass
  rate / conditional share / dominant problem type /
  review or watch
- Table 3: review fact pack - for each supplier called
  in, up to 3 of the month's facts (date / truck /
  problem / disposition that day), no adjectives
- Table footer, data completeness: how many days of
  receiving records are missing this month, and how many
  exception slips lack a problem description; over 【3】
  missing days, the month's pass rate is flagged "not
  comparable"

Constraints:
- Do not output "recommend ending the cooperation" -
  that is a purchasing-and-owner decision; this audit
  outputs facts and grades, and the ledger for
  termination (alternate supply, switching cost) lives
  outside receiving data
- Shapes go to type distribution, never cause - "lots of
  over-temps means their cold chain is broken" stays
  unwritten; whether it's the cold chain is for the
  supplier to explain
- Verdicts with a blank problem description go to their
  own "unclassified" column, never smeared across the
  five

Why it is written this way

Conditional acceptance standing as its own column, never merged into pass, is this audit's most important basis. A concession is "accepted this time, not again"; its difference from a pass shows up not this month (the goods were received either way) but next month: a supplier whose concessions got dressed up as passes doesn't fix anything, the same problem keeps arriving as a concession, until one day QC declines and it lands as a full rejection - with the line down. The conditional share is a more leading indicator than the pass rate, which is why the watch line tracks it separately.

The dual denominator - by truck and by item - is prepared against the supplier's standard defense; the most heard line in these meetings is "it's one item, everything else was fine." Set the two side by side - 85% pass rate by truck, problems in 2 items - and the line becomes a fact statement rather than a shield: episodic or a production-line issue, the supplier sees it at a glance, and the fix becomes non-transferable.

Five fixed problem types with no invented classes keeps the matrix comparable - across months, across suppliers, so trends mean something. Correspondingly, "unclassified" stands alone rather than smeared, which keeps record quality itself visible: in months with high blank-description rates, the first thing to fix is the receiving log, not the supplier - walk into a review carrying blank slips and one "how was this truck judged that day?" flips the table.

No termination recommendation and no cause inference draw the same boundary as the purchasing audit: receiving data can prove what happened and how often; whether the supplier's cold chain, scales, or loadout is at fault is their explanatory space. The audit nails the facts and grades the reviews, and the explanation arrives at the table on its own - a receiver walking in with facts stands much steadier than one walking in with inferences.

Variables