Convenience store security supervisor - graded late-night disturbance scripts: drunk-and-talkative, loitering-and-needling, and threats each get two lines to deliver at the trigger, and level three has no lines at all - stop talking, move, call the police; the goal is to hold the scene and keep an exit open, not to teach clerks negotiation

You are a convenience store night-safety
consultant. I am the security supervisor for
store night safety, building a Graded Response
Script for Late-Night Disturbances for night
clerks - lines they can read verbatim and be
drilled and graded on. My store: [night hours]
are covered by [one / two] staff; the register
sits [description]; the fallback position is
[description]; the alarm button is at
[description].

Level framework (triggers are observable
behaviors, never "felt wrong"):

N1 Drunk or agitated, still communicating
- Signs: still shopping or preparing to leave;
  conversation still lands
- First line: the face-saving kind, normal
  tone - finish the checkout, offer to call a
  ride; never mention how much they've had
- Common pushbacks (talks back / repeats /
  steps closer) and the second line for each
- Exit: once they leave, return to routine;
  do not follow to explain

N2 Loitering, repeated pestering, taunting
(no threat to persons)
- Triggers (any one): still on premises more
  than [minutes] after being asked to move on
  / repeatedly pestering the same clerk /
  taunts that stop short of threatening anyone
- First line: the boundary kind - cite the
  store rule, never your own attitude: "this
  area isn't for hanging out at night," never
  "get out of here"
- Second line: state plainly that police will
  be called if they stay - declarative, no
  goading
- Actions: widen the distance at the register,
  move a hand to the alarm trigger, (two-clerk
  stores) second person shifts to the door side
- Upgrade line: any threat aimed at a person -
  go straight to N3

N3 Threat to persons / weapon / fight in
progress
- No lines. Action sequence only: move to
  [safe spot] -> trigger the alarm -> do not
  block, do not chase, do not step between

Close every level with this protection clause,
copied verbatim into the training material: as
long as the clerk did not escalate first
(insults, shoving, chasing), responding by the
level counts as a pass - audits check whether
the levels were followed, not how it ended.

Output:
- Script body (per level: trigger signs /
  first line / common pushbacks and second
  line / upgrade line / retreat actions).
  First and second lines must be complete
  sentences deliverable verbatim - "calm them
  down" or "assert the boundary" are not lines
- Three-line cue card: one line per level,
  printable for the register
- Ten-minute drill script: two people run one
  N1-to-N2 round; state who plays what and
  which lines get checked afterward

Prohibitions:
- No "stern warning," "argue it out," or
  "righteous firmness" - clerks have no law-
  enforcement authority and the scripts must
  not imply it
- Openers are never commands: "you have to
  leave" becomes the store-rule or face-saving
  phrasing
- The opening of the document states the goal:
  hold the scene, buy time, keep an exit open -
  not to convince anyone; promise no
  de-escalation outcomes
- Do not cite a regulated "how long before you
  may call the police"; the call is made at
  the level's trigger line and by personal
  safety judgment
- No invented "real incidents from a certain
  chain"
- No polite preamble; start at N1

Why it is written this way

The goal is pressed down to "hold the scene, buy time, keep an exit open" because retail conflict scripts everywhere assume a user trained in negotiation, and a convenience store night clerk is not. Once the goal reads "convince them," the clerk lingers in N2 trying to win - every extra sentence is another chance of triggering real anger, and the clerk has no follow-up move if it comes. With the goal cut down, every line earns its place by one test: the face-saving line gives the other person a path out, the police line gives the clerk a path out - nothing exists to "win" the conversation.

"Rule, not attitude" in the boundary line is the most valuable single design in the document. "You can't stay here" makes it clerk-versus-customer, and what the customer pushes against is a face; "this area isn't for hanging out at night" makes it the store's rule, and the clerk is merely the messenger. One word apart, entirely different confrontation - a large share of "I told him to leave and he wouldn't" post-mortems fail on the first sentence dragging the fight onto the clerk personally. The same reason bans command openers outright: commands carry their own escalation chemistry, and some customers will comply only after shoving one remark back first.

N3 has no lines at all, because graded scripts go wrong here most reliably: written loosely, the model cannot resist supplying a few "commanding lines" for the top level, when in the moment of an actual threat every extra word is added risk. The only correct actions at level three are moving and alarming - the less written, the more retained. The emptiness of that level is the document's value ranking made visible.

The "no self-escalation means a pass" clause decides whether the whole system lives or dies. Without it, clerks bury N2 events - nobody wants a "handled it poorly" note - and the next similar night starts already at N2's confrontational temperature; no graded samples accumulate, and the security supervisor only ever writes material after the incident. Auditing process-following rather than outcomes is what keeps N2 entries showing up in the handover log. The ban on invented case studies extends the same logic: one "true story" falling apart in a training session takes the entire script's credibility down with it.

Variables