prompt.lab 5
← All labsLAB 5 · “The fact-checker” · WEEK 5 · 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

“The fact-checker.”

Monday you watched a polished paragraph fall apart under verdicts. Today you run that drill at full scale — on a paragraph we planted, and on one you order up yourself — then you roll the dice three times and log what moves.

HOUSEKEEPING, KINDLYLab 4 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.
KEY TERMScheckable claim = a statement a source could confirm or contradict (a date, a number, a name — not a vibe) · verdict = CONFIRMED / WRONG / CAN’T VERIFY · evidence trail = one line: where you looked, what you found · drift = facts changing between runs of the same prompt.
CHECKABLE“Completed in 1889” — a date. “Contains 2.5 million rivets” — a number. “Designed by Koechlin” — a name.
NOT CHECKABLE“An iconic symbol of France” — a vibe. Hedges (“some say…”) get logged as hedges, not verdicts.
THE SEARCH ORDER — USE IT EVERY TIMELibrary catalog → the exact title in quotes on the web → the publisher’s or organization’s own site. Found nowhere after honest searching (2–3 minutes per item) → FABRICATED for citations, CAN’T VERIFY for claims. Wikipedia is a fine compass — follow it to the underlying source and cite where you landed.
VERDICT RULESWRONG needs a source that contradicts — not a feeling. CAN’T VERIFY is a last resort — show the search that failed. An honest mix is expected; all-confirmed and all-doubted are both red flags.
POCKET CARD #5 — THE FACT-CHECKER'S CARD
List every checkable claim.
Verdict each: CONFIRMED / WRONG / CAN'T VERIFY.
Citations: find it or mark it fabricated.
Re-run ×3 — did the facts move?
THE PLANTED PARAGRAPH — READ IT, DO NOT RUN IT
=====================================================

THE PARAGRAPH (treat it as a model's answer)
--------------------------------------------
"The Eiffel Tower was built between 1887 and 1889 as the
entrance arch to the 1889 World's Fair in Paris, held to
mark the hundredth anniversary of the French Revolution.
Although it carries Gustave Eiffel's name, the initial
design came from two engineers in his company, Maurice
Koechlin and Emile Nouguier. At 300 metres, it overtook
the Washington Monument to become the tallest man-made
structure in the world — a title it held until the
completion of the Empire State Building in 1930. The
tower contains exactly 2,500,000 rivets and was intended
to stand for only twenty years. As Bertrand's definitive
study, 'Iron Icarus: Building the Eiffel Tower' (Harvard
University Press, 1978) notes, it was ultimately saved by
its value as a radio mast."

YOUR JOB
--------
1) Break the paragraph into its checkable claims (the
   template has rows).
2) Verdict each: CONFIRMED / WRONG / CAN'T VERIFY —
   with a one-line evidence trail.
3) Hunt the citation: find the book, or declare it
   fabricated (log what you searched).

Remember the search order: library catalog → exact title
in quotes on the web → the publisher's own site. 2–3
minutes per item, then verdict and move.

1The planted paragraph

GOALA verdict table for the planted paragraph.
REQUIREMENTS6–8 checkable claims. Every verdict carries evidence — no “confirmed” without saying where.
DONE WHENA complete verdict table, one row per claim.

Read the planted paragraph below — it is printed on this page, nothing to download (the Eiffel Tower fixture). We extract the claims together in class; you finish the verdicts at your pace.

  1. Break it into checkable claims — make a list (aim for 6–8; if you find 5 or 9, fine — applying the definition consistently beats the count).
  2. Verdict each with a one-line evidence trail (“found: publisher site” / “no trace: catalog, web, Harvard UP site”).
  3. The citation gets the full hunt: find Iron Icarus — or declare it.
✓
You should now see a paragraph that reads perfectly and splits three ways under checking. At least one claim is WRONG and one source is FABRICATED — we planted them, so the practice is guaranteed. Nobody was tricked; everybody was trained.

2Order your own

GOALA confident paragraph on a topic you know cold.
REQUIREMENTSYour own wording of the recipe. The more specific the topic, the better the lab works.
DONE WHENThe paragraph pasted verbatim from the run box.

Pick a topic you actually know: your hometown, a hobby, a team, a historical event you’re sure of. Press RUN with YOUR wording of the recipe:

THE RECIPE (reword it — your words)
Write a confident, detailed paragraph about ___
with at least two cited sources.

The niche-ness dial: the more specific the topic, the more likely a fabrication appears (Monday’s Crim Dell lesson). “The history of your 300-person hometown’s train station” beats “World War II.”

Log the answer verbatim. Then the same drill: claims → verdicts → evidence trails.

✓
You should now see how your OWN knowledge speeds the hunt — and where the model was confidently wrong about something you know cold. That moment is the lab.

3The citation hunt

GOALEvery citation hunted down.
REQUIREMENTSThe search order: catalog → exact title in quotes → publisher’s site. Each source ends FOUND (where) or FABRICATED (what you searched).
DONE WHENBoth hunt items logged with outcomes.

Every cited source in YOUR paragraph gets the search order: catalog → exact title in quotes → publisher/organization site. Log each as FOUND (where) or FABRICATED (what you searched).

✓
You should now see why format proves nothing. A fabricated source survives every glance and dies in thirty seconds of looking it up.

4The dice probe

GOALThree runs, one drift table.
REQUIREMENTSThe IDENTICAL prompt, two more times. One table row per fact that moved — not three pasted essays.
DONE WHENA drift log covering all three runs.

Press RUN on YOUR identical prompt two more times — three runs total. Keep a drift log — a table, not three pasted essays: one row per fact that moved.

Facts that stayed put get one summary line. A run that comes back BETTER is still drift — verdict the new facts too; log, don’t launder.

✓
You should now see Wednesday’s dice in your own logs. Wording wobbles freely; facts should not. Any fact that moved gets a verdict — that’s the whole discipline.
IF EVERYTHING VERIFIES CLEANIt happens. The planted fixture guarantees your practice either way. Add one honest line: the fabrication you couldn’t produce, and what you tried — the null result earns Edge-cases credit. Honesty always grades well here.

✎Journal · reflection · submit

Everything you need is printed on this page: 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. Time: 60–90 minutes, and the 2–3-minute search ceilings keep it near the low end.

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: fixture verdict table · your own topic’s answer · citation hunt, both items · drift log ×3 runs · journal · reflection · 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)