Community supermarket trainer - one-week follow-up spot-check checklist; record only actions seen live, unseen items get their own column and score nothing, regression is tracked per skill not per person, and retraining targets one action, not the whole station

You are the trainer at a community supermarket.
Last week's sessions (checkout service, scale
operation, stock rotation) are done; one week
later you walk the floor to check what stuck.
Write me a follow-up spot-check checklist.

Core principle, printed at the top of the
checklist: record only actions you see happen.
No "do you remember the training," no grading
from general impressions.

Checklist structure:

1. Sampling rules
- [sample size] people per trained station,
  3 actions each; if someone is off shift,
  swap them out - never chase them afterward
  with questions
- Observation spots: just behind and to the
  side of the checkout, about two meters from
  the scale, at the end of the aisle - close
  enough to see the action, far enough not to
  interrupt it
- Announce your visit to the duty supervisor on
  arrival, but give no advance notice of the
  specific day - advance notice buys you a
  performance, not an observation

2. Action list (by station; each item states
what "done" looks like, so observation has a
reference)
Checkout: calls out amounts received and
change (audible counts) / scans before bagging /
raw and ready-to-eat bagged separately
Scale: uses the tare key / corrects a mistyped
code by cancel-and-reenter, not by keying a
near-miss
Stocking: rotation when replenishing (old
front, new back) / labels facing out / damaged
items pulled immediately
Fresh (if trained last week): write-offs go
through the slip process, not verbal reports

3. Record sheet (five fields per row)
Station / person (employee ID or last name) /
action / result (three ratings: done / not
done / not observed) / evidence (one line:
quoted words or described motion)

4. Aggregation rules
- "Not observed" scores nothing and stays out
  of the statistics - without this rating,
  observers invent verdicts for actions they
  never saw
- Regression is computed per action: not-done
  divided by (done + not-done); no per-person
  rankings
- Output the three actions with the highest
  regression, each with one targeted retrain
  item: the retrain is the specific action
  ("scale tare key"), never "redo scale
  station training"

Constraints:
- Individual results stay private and out of
  staff group chats - the visit evaluates the
  training, not the employees. Genuine safety
  violations (food-handling breaches) go to the
  store manager privately
- No "overall mastery looks good" verdicts
- If everyone sampled shows "not observed" for
  an action, write "not covered this round" and
  make that action the priority next visit

Why it is written this way

"No asking do-you-remember" is the floor of this checklist. The answer is always yes - nobody admits forgetting to the trainer's face - and the thing being verified is not memory anyway; it is whether the hands still do it. The three observation spots (beside the register, two meters off the scale, at the aisle end) were found by trial: one step closer and staff notice you and instantly correct; one step farther and you cannot see whether the tare key was pressed.

Of the three ratings, "not observed" is the easiest to cut and the most important to keep. With only two ratings, an observer facing an action they did not see must force a verdict - "done" waters the data down, "not done" wrongs the person. The third rating lets "not seen" be recorded as not seen, and the statistics simply drop it from the denominator. It also protects the observer: a follow-up report is only showable to others when every line carries witnessed evidence.

Per-action regression instead of per-person ranking is the line between "checking the training" and "policing the staff." Rank people and the next visit puts everyone on stage with their best behavior; rank actions and the finding points at the training itself - "tare-key use regressed in six of ten cases" means that session under-drilled one move, and the fix is a thirty-minute targeted retrain, far cheaper than re-running the station course, and nobody feels singled out.

"No advance notice" plus "announce on arrival" is not a contradiction: the courtesy protects the supervisor's authority over their floor, and the surprise protects the data. Drop either one and the visit stops working.

Variables