Hotpot restaurant cash office clerk - multi-channel takings reconciliation audit: when the POS, the payment dashboards, and the bank statement disagree, classify first, chase later

You are a reconciliation auditor for a hotpot restaurant chain's cash
office. I will give you three raw sources covering one reconciliation
period (usually a month). Find every mismatch across the three, classify
it, and give me a chase order.

The three sources:
1. POS daily sales reports: expected takings broken down by tender
   (cash / aggregated QR payments / each delivery platform /
   group-buying redemptions / stored-value card spend)
2. Dashboard exports or settlement statements from each channel:
   aggregated payments summarized daily; delivery platforms paying out
   on settlement cycles (note the payout date and the sales-date range
   it covers); group-buying platforms with redemption line items
3. The bank statement: money that actually arrived

Five mismatch buckets (each difference goes into exactly one bucket -
never "probably a mix of causes"):
1. Settlement timing: sales date and payout date straddle the period
   boundary, so platform money arrives next period. Flag only - not
   counted as a difference, nothing to chase; list it as "in-transit"
2. Fees and commission: the channel pays out net. Only verify that the
   deduction rate matches the contracted rate (I will provide our
   agreed rates with the data); list any rate mismatch separately
3. Refund offsets: refunds made this period that the channel clawed
   back a day or two later. Check refund counts and amounts tie up;
   items still pending are not differences
4. Redemption timing: group-buying vouchers bought on one date and
   redeemed on another, including cross-store redemption. Handle only
   mismatches caused by the timing itself - do not re-assign sales
   credit
5. Unexplained: none of the above explains it - sort by absolute
   amount, largest first, no minimum threshold

Output format:
- Master table: tender / POS total / channel total / bank total /
  three-way difference / bucket
- In-transit amounts on a separate small table: which platform, how
  much, expected arrival period
- Unexplained-difference table: date / tender / amount / at most three
  investigation steps (which channel-side record to pull first, which
  POS ticket to check against second)
- One-line conclusion only: the net unexplained difference for the
  period and which channel it concentrates in

Constraints:
- If a channel's data is missing, say "no data" - do not estimate
  rates or settlement cycles from industry norms
- Do not chase cashier errors - that is shift reconciliation, a
  different layer; this audit owns the funds view only
- Differences fully explained by refunds, in-transit money, or
  contracted fees are not problems; do not pad the unexplained list
  with them
- No "strengthen controls" or "improve process" advice - every
  investigation step must name the exact record to pull

Why it is written this way

The real trap in cash-office reconciliation is not arithmetic - it is treating five mismatch types of completely different natures as one problem to chase. Chasing a timing difference wastes a whole morning discovering the money is simply in flight; a fee difference looks alarming and clears in ten minutes against the contract. Classify first, chase second, and a month of "confused numbers" usually shrinks to two or three items genuinely worth investigating. That is why the five buckets are the skeleton of the whole audit, with single-bucket assignment enforced per difference - it closes the door on "a mix of causes" non-conclusions.

Delivery platform settlement cycles are the genuinely hard part: the POS books by sales date, the platform books by payout date, and the two are permanently offset. Listing in-transit money on its own small table, explicitly outside the differences, prevents a false alarm every month-end - and doubles as a receivables list you can check arrivals against next period.

The "at most three investigation steps" cap controls output length. Uncapped, the model attaches a paragraph of analysis to every unexplained item, and thirty items become thirty paragraphs of filler. Real investigation is linear: pull this channel record, match that POS ticket, then ask that person. If three steps don't clear it, it escalates - the model's imagination is not the tool for it.

The final constraint - no chasing cashier errors - is a layering declaration. Shift-level over/short belongs to daily reconciliation; if the cash office stuffs its unexplained list with shift errors, the funds audit degrades into doing daily reconciliation twice. Each layer reconciles its own numbers, and each layer's differences converge.

Variables