prompt.lab 6
← All labsLAB 6 · “Armour a real prompt” · WEEK 6 · LEVEL 3
Done live in the hands-on session · submitted on this page
COLL 100 · PROMPT ENGINEERING · HANDS-ON SESSION · IN CLASS, THEN ON YOUR OWN

“Armour a real prompt.”

Not a made-up one. Take a prompt you have genuinely typed this semester — your Lab 3 document Q&A, a study prompt, a club chore — and armour it with three of this week’s reliability lines: a shape, the standing sentence, and one self-check. You hand in the before and the after, side by side, with the runs that prove the difference.

ONE DELIBERATE CHANGE PER PASS  Same discipline as always: run after every line you add, and log what moved.

HOUSEKEEPING, KINDLYLab 5 unlocks this one — if it isn’t finished, finish it on its page at the start of this session; see the TA if you’re stuck.
PRIVACYArmour a prompt about coursework, clubs, or plans — never one containing personal, medical, or financial information. Swap real names for fictional ones before you paste.

1Pick it, run it bare

GOALYour real prompt, run bare.
REQUIREMENTSPick a prompt where a wrong answer costs you something. Run it exactly as you normally type it.
DONE WHENThe BEFORE run logged, with two unverifiable places marked.

Choose a prompt you actually use — one where a wrong answer would cost you something. Put it in a prompt box and press RUN, exactly as you have been typing it. Log the prompt and the full answer verbatim. This is your BEFORE.

✓
You should now have one bare run logged. Mark two places in the answer you cannot verify from where you sit — those are the holes the armour is for.

2Add a shape

GOALMake the answer checkable at a glance.
REQUIREMENTSName the fields, the order, and what must NOT appear. Unknown values say “not stated” — never guess.
DONE WHENThe shaped run logged, plus what it made checkable.

Name the container. Fields, order, and what must NOT appear — so a person (or a checklist) can verify the answer at a glance:

THE SHAPE LINES
Return ONLY a table: [your columns]. One row per item.
If a value is unknown, write “not stated” — never guess.
No preamble, no closing line.

Adapt the columns to your prompt (a study plan has different fields than a club email). Run it. Log the run and one sentence on what the shape made checkable that prose hid.

THE TRAPA shape so rigid the answer cannot fit — “exactly 5 bullets” on a topic with three ideas — gets filled with padding. If your constraint cannot be met, the model will pad rather than push back. Loosen the count, keep the fields.

3Add the standing sentence

GOALMake honesty an allowed answer.
REQUIREMENTSFence your material, then the cite-or-abstain sentence. Count claims: sourced vs “not provided”.
DONE WHENThe run and both counts, logged.

The cite-or-abstain line. Paste the material your prompt should work from (fenced, like Lab 3), then add:

THE STANDING SENTENCE
Use only the material above. Cite the line each claim comes from —
or write “not provided”. If you are unsure, say unsure.

Run it. Log the run. Count: how many claims now carry a source, and how many honestly say “not provided”? Both numbers go in the log — the second one is the armour working, not failing.

4Add one self-check

GOALMake the model audit its own answer.
REQUIREMENTSAdd ONE self-check line — A (name the weakest claim) or B (compute it a second way).
DONE WHENThe armoured AFTER run with the self-check’s own verdict.

One line that makes the model audit its own answer before you do:

PICK ONE
A — “Then check your own answer: name the step or claim most likely
     to be wrong, and why — or say ‘confident’.”
B — (for anything with numbers) “Compute it a second way and say
     whether the two results agree.”

Press RUN on the full armoured prompt — shape + standing sentence + self-check. This is your AFTER. Log it verbatim.

✓
You should now have a before and an after on the same task. If the after is not more checkable than the before, one of the three lines is not doing its job — say which, and tighten it.

5Break it on purpose

GOALTry to break your own armour.
REQUIREMENTSOne trap question the material cannot support — adjacent but absent. Write your prediction BEFORE running.
DONE WHENThe break log: trap · prediction · what happened · the fix, if it leaked.

Ask your armoured prompt one question the material cannot support — something adjacent but absent. A weak prompt invents; an armoured one says “not provided”.

  1. Write the trap question and predict, before running, what the armoured prompt should do.
  2. Run it. If it held — log the refusal; that is the lab’s money shot.
  3. If it invented — you caught a leak. Tighten ONE line, re-run, log both runs and the fix.

✎Before/after · journal · submit

BEFOREThe bare prompt + its run, with your two unverifiable spots marked.
AFTERThe armoured prompt + its run, with sources counted and the self-check’s own verdict.

There is no file and nothing to upload: you do the lab right here in the hands-on session, and the submission panel below is where it all goes. Each piece is checked the moment you type it, and submitting records the lab and unlocks the next one.

HOW THIS IS GRADED — COMPLETION, IN ORDER
Completion, not polish — every item present and genuine, and the points are yours. Nobody grades your prose. The checklist: before run · shape run · standing-sentence run with both counts · after run with self-check · break log with prediction · statement in range.  Labs are 10% of the course, across six labs, and they unlock in order — one left undone blocks the next. No late window: finishing late beats not finishing — if you fall behind, tell your instructor rather than skipping ahead. AI use: Level 3 — submit your prompts and working results.

→Go further (optional)