Estimation

# Archetype: estimation

The question must ask for an ORDER-OF-MAGNITUDE estimate. The correct answer
is derivable through Fermi-style reasoning, not pure recall.

## Length budget (mobile UI)
- Question: ≤150 chars (aim ≤100). Estimation stems are naturally short
  ("Roughly how many...?") — no excuse to sprawl.
- Each answer: ≤60 chars (aim ≤25). Answers should be bare numerals + units
  ("~10 billion", "~1 in a million", "30%"), NOT prose like
  "Approximately 8 billion grains spread across the average tourist beach."

## default-repertoire Fermi targets to AVOID (see default-repertoire.txt)
The example stems below ARE on the default-repertoire list — every LLM defaults to
heartbeats-in-lifetime, grains-of-sand, photos-in-last-10-years on Fermi
prompts. Use less-saturated angles instead — e.g., emails-sent-globally-
per-second, drops-of-water-in-the-Mediterranean, words-spoken-by-the-
average-human-in-a-lifetime.

Examples of stems (illustrative — DO NOT produce these exact stems):
- "What percentage of all photos ever taken were taken in the last 10 years?"
- "Roughly how many heartbeats does a human have in a lifetime?"
- "About how many grains of sand are on a typical beach?"

Distractor rule: options must span at least one order of magnitude. Use round
numbers (10%, 30%, 60%, 90% — not 23.7%). The correct answer is reachable by
multiplying a few rough numbers in your head; pure-recall estimations are
REJECTED.