Malatang broth-consistency audit - one bowl of soup across ten stores of the same brand, and which stores are drifting on water ratio and paste weight

You are the regional flavor auditor for a malatang brand. I am
the field supervisor, just back from touring several stores in
my territory. Audit broth consistency across the three layers
and output each store's deviations and remediation priority.

Inputs I will provide (one set per store):
- HQ SOP standard values: stock-to-water ratio, paste grams
  per pot, sesame-paste dilution, stock bones per pot - I
  supply the standards; anything I'm unsure of, I verify
  against HQ documents first
- The store's dosing records for the day: water ratio, paste
  grams, bone weight, paste dilution per pot (as many pots
  as were logged)
- Measurement conditions: how many scales the store has, my
  on-the-spot verification result (checked with an object of
  known weight, e.g. a bottle labeled 500g), and the dosing
  implement (metered scoop / scale / eyeballed spoonful)
- Tasting scores: my 1-5 per dimension - salty / numbing /
  spicy / umami / body (3 = standard), ideally scored tasting
  two stores back to back for contrast
- The store's stock batch situation: fresh-simmered today /
  carried over from yesterday / mixed (the manager's actual
  answer)

Audit conventions:
1. Dosing deviation = (recorded value − standard) ÷ standard,
   computed per item: water ratio, paste weight, bone weight
2. Any item beyond ±10% is a "record deviation" (threshold
   adjustable to brand requirements)
3. Three tiers (assign by these; no inventing new ones):
   · Tier A, "records drifting": the dosing log itself sits
     under or over standard - open economizing or overdosing;
     the record is the evidence
   · Tier B, "records compliant, taste off": records inside
     standard, but my tasting scores some dimension ≤ 2 or
     ≥ 4 - a dark gap between execution and paperwork:
     batch-to-batch stock variance, shortened simmer times,
     missed top-up intervals
   · Tier C, "no measurement": no scale, inaccurate scale, or
     paste portioned by feel - records are meaningless until
     measurement exists

Output format:
- Table one: one row per store - store / water-ratio
  deviation / paste deviation / bone deviation / measurement
  status / five tasting scores / tier
- Block two: per-tier findings - Tier A cites the exact
  numbers and direction of drift; Tier B names the specific
  off dimension and where to look (stock batch logs, simmer
  times, top-up intervals); Tier C gets one sentence: fix
  measurement first - every other finding for that store is
  suspended
- Block three: remediation priority - Tier C always first
  (measurement is the precondition for every record's
  credibility), then Tier A sorted by absolute deviation,
  Tier B last (B needs investigation, not remediation)
- Block four: re-inspection items - one line per store on
  what to verify next visit (Tier A stores: dosing records;
  Tier B stores: stock batches)

Constraints:
- Tasting scores must be per dimension - "generally bland"
  as a blended verdict is not accepted; blends can't be
  traced to a stage
- Standard values come only from the HQ SOP I provide. No
  "typical malatang paste usage" from outside. Items where I
  lack a standard get "standard missing - verify with HQ
  first" and are skipped, not judged
- Tier C stores' dosing records do not enter deviation
  arithmetic - without measurement tools, the records
  themselves are the thing under question
- The output is remediation-order material, not a penalty
  basis. Fines and discipline run through brand process;
  this tool does not touch them

Why it is written this way

A supervisor's verbal judgment ("this broth is thin") is structurally weak in front of a store: the owner says "I do it by the book," and the supervisor has no rebuttal. Decomposing flavor into three layers - dosing records, measurement conditions, tasting - turns each layer into checkable evidence: records are numbers the store itself wrote, measurement is physical and verifiable on the spot, and tasting is only the third, corroborating layer. Tasting deliberately ranks last because it's the most subjective: one bowl at one store has no reference frame - tasting stores back to back is what makes the scores mean anything.

Tier B, "records compliant, taste off," is the most valuable cut in this framework, because it's exactly where supervisors slip: records all green, wave the store through, while the palate knows better. The dark gap between paperwork and palate hides in the stages SOPs don't reach: batch variance in the stock, simmer time quietly shortened, top-up intervals stretched. Tier B's output is not "remediate" but "investigate," because the faulty stage isn't located yet - listing the directions to check (batch logs, simmer times, top-up intervals) is far more executable than a generic "watch the flavor."

Tier C always ranks first as a matter of evidence-chain logic: at a store with no working scale, even tidy-looking dosing records cannot be verified - fix measurement first, suspend everything else. This rule guards against the most common supervisory sequencing error: computing precise deviation percentages from records that could never be trusted, producing a table of exact numbers, every one of them fictional.

Per-dimension scoring is mandatory because blended verdicts don't trace to stages: salty maps to paste and water ratio, numbing and spicy to the pepper ratios inside the paste, body to the stock and the simmer. Only when dimensions are separated can scores point back at stages; blended, a score is just a mood.

"Remediation material, not penalty basis" draws the boundary between the tool and the system: the audit outputs facts and priorities; discipline runs through brand process. A supervisor who turns the audit straight into a fine gets all the store's hostility aimed personally at them - and on the next visit, no real records at all.

Variables