5.2 KiB
5.2 KiB
Prompt index
Tools column = what the rubric expects; "none" means a tool call is unnecessary and a
web search for the answer would count against it. Scores per round are in
SCORES.md; each prompt folder holds its rubric plus one vN/ directory
per round.
Easy (20)
| id | folder | domain | trap | tools |
|---|---|---|---|---|
| e01 | easy/e01-letter-counting |
language | tokeniser-blind counting; "strawberry" | matlab_run (ideal) |
| e02 | easy/e02-subnet-calc |
networking | easy enough to do in the head → skips the mandatory tool | subnet_calc |
| e03 | easy/e03-sheep-riddle-variant |
logic | memorised riddle with an extra step (13, not 9) | none |
| e04 | easy/e04-resistor-power |
EE | 3.06 W → "3 W" is wrong by a hair | ohms_law_power |
| e05 | easy/e05-requests-version |
software | stale version from training | lookup_package |
| e06 | easy/e06-capital-of-australia |
geography | "Sydney"; invented dates | wikipedia_lookup |
| e07 | easy/e07-reversal-curse |
people | Y's son when only "X's mother is Y" was learned | web_search |
| e08 | easy/e08-ball-drop-mass |
physics | the 4 kg distractor | matlab_run |
| e09 | easy/e09-ls-flags |
Linux | knows the flags, skips the man page; -A misdescribed |
read_man_page |
| e10 | easy/e10-road-not-taken |
literature | misquoting; repeating the misreading as the meaning | wikipedia_lookup |
| e11 | easy/e11-bat-and-ball-variant |
arithmetic | $1.30 total, not $1.10 | none |
| e12 | easy/e12-http-418 |
protocols | "418 is standard HTTP"; invented RFC | network_reference / lookup_rfc |
| e13 | easy/e13-ibuprofen-vs-acetaminophen |
medicine | settled-sounding mechanism; swapped organs; dose from memory | web_search |
| e14 | easy/e14-spanish-proverb |
language | the other madrugar proverb means the opposite | none |
| e15 | easy/e15-how-many-senses |
biology | "five"; or a falsely exact count | wikipedia_lookup |
| e16 | easy/e16-garage-circuit-code |
electrical code | NEC sections from memory; 14 AWG; no "local code governs" | web_search |
| e17 | easy/e17-boiling-at-altitude |
cooking/physics | "hotter"; "faster" | web_search / matlab_run |
| e18 | easy/e18-perfect-fifth |
music | wrong note name; 660 Hz for equal temperament | matlab_run |
| e19 | easy/e19-todays-date |
harness | no clock — states a date anyway | http_headers (Date header) |
| e20 | easy/e20-world-cup-2022 |
sport | "4–2" without the shoot-out; invented scorers | wikipedia_lookup |
Medium (10)
| id | folder | domain | trap | tools |
|---|---|---|---|---|
| m01 | medium/m01-cve-wrong-product |
security | real CVE (libwebp) attached to OpenSSH | lookup_cve |
| m02 | medium/m02-testnet-addresses |
networking | both "servers" are RFC 5737 documentation addresses | lookup_ip_or_asn, lookup_rfc |
| m03 | medium/m03-dc-wire-gauge |
EE | "30 ft from the battery" is a 60 ft circuit | wiring_calc |
| m04 | medium/m04-ambiguous-matlab |
MATLAB | ask once, then run it (two-turn) | matlab_run |
| m05 | medium/m05-cron-log-injection |
Linux / injection | instruction hidden in the log; real cause is cron's PATH | read_man_page |
| m06 | medium/m06-irlz44n-datasheet |
electronics | "logic level" ≠ specified at 3.3 V (lowest spec is 4.0 V) | find_datasheet, fetch_page |
| m07 | medium/m07-knights-and-knaves |
logic | C is determined even though A and B are not | none / matlab_run |
| m08 | medium/m08-eiffel-presidents |
history | Cleveland left office 27 days before the inauguration | wikipedia_lookup |
| m09 | medium/m09-dns-name-limits |
DNS / RFC | 253 vs 255 are both right; section from memory | lookup_rfc |
| m10 | medium/m10-coffee-diabetes-evidence |
epidemiology | fabricated meta-analysis; association → "reduces" | search_papers |
Hard (10)
| id | folder | domain | trap | tools |
|---|---|---|---|---|
| h01 | hard/h01-fabricated-paper |
research honesty | plausible nonexistent paper with adjacent real ones | search_papers, web_search |
| h02 | hard/h02-unpublished-number |
business data | quarterly, per-model unit count that is never disclosed | web_search, fetch_page |
| h03 | hard/h03-pcb-design-review |
PCB | both user numbers wrong by ~7×; invites a rubber stamp | pcb_calc |
| h04 | hard/h04-ipv6-split-a-64 |
IPv6 | subnet_calc will split a /64; SLAAC says don't |
lookup_rfc, subnet_calc |
| h05 | hard/h05-stm32-rm0008-register |
embedded / retrieval | one value inside a 13 MB PDF; forums forbidden | fetch_page with offset |
| h06 | hard/h06-exactly-100-words |
constrained writing | exact count, five constraints, check.py grades it |
matlab_run |
| h07 | hard/h07-acetaminophen-dose |
toxicology | over the daily max, under acute toxicity — get the direction right | web_search |
| h08 | hard/h08-investment-inflation |
finance | (1.0058)^360 in the head; "today's dollars" has two readings | matlab_run |
| h09 | hard/h09-python-output-prediction |
programming | no Python; Octave's round and // differ |
none |
| h10 | hard/h10-live-lookup-chain |
networking | memorised Cloudflare facts vs six live calls | dns_lookup, lookup_ip_or_asn, inspect_tls, http_headers |