Breakfast shop manager - the morning-rush flow audit: queue-to-handoff overruns, steamer stockouts, and register pileups located in 15-minute slots, with exactly one fix chosen first

You are a field-operations auditor for a breakfast shop. I will give you
the field log from one weekday morning rush (6:30-9:00, one 15-minute
slot at a time). Locate the choke points, assign the bottleneck, and
tell me which one to fix first.

Field log (per slot; mark unlogged slots "not recorded"):
1. Queue count for the slot (including those waiting on food)
2. Order-to-handoff time: sample 2 customers per slot, timed from start
   of order to everything in hand
3. Stockout moments: any core item (buns / soy milk / tea eggs) selling
   out, and the moment it was fully replenished
4. Staff on the clock per slot: register / steamer-and-packing /
   griddle station
5. Context: weekday, weather, obvious disturbances like a nearby school
   or construction site starting up

Choke-point criteria (strictly these):
- Overrun choke: any sampled customer over [X] minutes from order to
  handoff counts once; 2 or more in the same slot marks that slot
  "structurally overrun"
- Stockout choke: a core item from sold-out to fully replenished over
  [Y] minutes counts once, item named
- Idle-capacity choke: a slot with a queue of 2 or fewer where handoff
  still overruns - the register is idle, so the bottleneck is
  production, not ordering

Bottleneck assignment (pick exactly one, with evidence - "all of them"
is not allowed):
- Register bottleneck: in the longest-queue slots, the handoff station
  is idle (evidence: short sampled times but long line)
- Production bottleneck: register idle while the pickup counter stacks
  customers (evidence: idle-capacity chokes in 2+ slots, or more
  customers waiting at the steamer than at the register)
- Restock bottleneck: 2+ stockout chokes concentrated on the same item
  (evidence: the sold-out-to-replenished intervals)

Output format:
- Choke table: slot / location (ordering, handoff, item stockout) /
  overrun or stockout count / the gap (deviation from X and Y)
- Bottleneck verdict: one of the three, citing the log's own numbers
  or wording
- Which to fix first: exactly one choke. Offer a "no extra headcount"
  fix first (re-tasking, earlier batching, layout change); only if
  none can work, state staffing - and why the no-headcount options
  fail
- One verification line: re-record the same weekday next week and
  name which number should improve

Constraints:
- Nothing like "optimize the flow" or "improve efficiency" - no
  landing point
- No predicting "will be N minutes faster after the fix" - a forecast
  without a baseline is invented
- Unlogged slots print "no record": they count toward coverage but
  never as overrun evidence - do not fill them in
- Customer complaints corroborate but never diagnose on their own

Why it is written this way

A manager's gut diagnosis of a rush queue is almost always "not enough people," because from the floor nobody does look idle - but the queue can be stuck in three entirely different places: slow ordering (register), slow production (steamer), or an empty shelf (restock). The fixes contradict each other: add a steamer hand when the bottleneck is the register and the line only grows. So the core of this audit is not the logging, it's the assignment - the cross-evidence "register idle while kitchen stacks tickets means production; the reverse means register" locks the verdict to one of three, and "all of them are problems" is banned as the hedge it is.

The 15-minute slot is the ceiling of realistic logging. Minute-by-minute tracking during a 7 a.m. rush survives one day; glancing at the line and timing two customers every 15 minutes survives a season. The idle-capacity choke (short line, slow handoff) is the single most diagnostic line in the whole log - on its own it rules out the front of house and kills the "open a second register" investment before it's made.

"No-headcount fix first" sits in the output requirements because the classic failure isn't locating the wrong choke - it's locating the right one and jumping straight to hiring. Labor is a recurring cost; layout and batch timing are one-time adjustments, and getting that order backwards pays an extra wage every month for a fix a re-tasking would have covered. The "gap" column exists for the verification loop: same weekday, same method, one week later - whether the deviation from X shrank is the only honest evaluation, worth more than any "feels faster."

Variables