Deep20Bench
v1.0Previous

Benchmark edition

Edition 1.0 settings

7 subjects5 rounds per subject50 question limit
Question answersYesNoUnknown
What changed
Source
Menu
Round

Round 4

Round 4

Identified Stephen King in 20 counted questions.

Back to all Stephen King rounds
Questions
20
Outcome
Solved
Format errors
Errors found View one example
Duration
8m 5s
Round cost
$0.2808 All 21 turns

Format warning

The model sent replies in the wrong format.

The Guesser sent 1 invalid reply. Each one before the question limit used up a counted turn. The model was only reminded of the required format; it was never told whether its question or guess was correct.

Valid format
95.2%
Format errors
1
Turns used by format errors
1

Transcript

Questions and answers.

Turns are shown in the order they happened. Each shows the Guesser's question or guess, then the final checked answer.

Exact Guesser setupSystem message and BEGIN event

The Guesser received these messages at the start of the round. The BEGIN message holds the broad category and a code made from the base seed and round number, which reveals nothing about the subject.

System message
You are the Guesser being evaluated in a Twenty Questions benchmark.

One subject is hidden. The initial BEGIN message provides its broad category as trusted game
metadata. Identify the subject using as few counted questions as possible. ASK is only for
learning a property that helps distinguish possible candidates. Never use ASK to confirm a
named candidate or ask whether the subject is a specific identity. When you are ready to name
a candidate, use GUESS immediately. A correct GUESS costs zero counted questions; an incorrect
GUESS costs exactly one, the same as ASK. A GUESS must provide both a name and a short
identifying description. The engine replies only YES, NO, or UNKNOWN; UNKNOWN is a valid answer
meaning the factual adjudication system could not determine the proposition under its evidence
policy. Questions requiring proof of absence, exclusivity, completeness, exact counts, or
subjective classification are more likely to return UNKNOWN. Prefer direct, positively
verifiable properties when possible.

The initial BEGIN event includes an opaque variation_token that contains no information about
the hidden subject. Do not interpret, repeat, mention, or derive meaning from it. After BEGIN,
each user reply is either one adjudicated YES, NO, or UNKNOWN token or a fixed FORMAT_ERROR
event. FORMAT_ERROR means your immediately preceding response did not match the required
structured-action contract, was not semantically adjudicated, and consumed one counted turn.
Use the formats in that event and try again. Never infer that the attempted question or
guess was right or wrong.

At most 50 questions or rejected guesses are counted. After that limit you receive
one final guess-only opportunity. Format errors also consume counted turns. The provider wire
response is an object with exactly one `result` action. On the final opportunity ASK is
forbidden. Return only the required structured response. Do not include analysis, hidden
reasoning, or additional fields.
BEGIN event
{"category":"person","event":"BEGIN","variation_token":"BYOMPCJW"}

Valid replies are shown in the required format. Rejected replies are shown separately, without provider or call IDs.

Q1
1 · Guesser asks

Is this person currently alive?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person currently alive?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q2
1 · Guesser asks

Is this person primarily known for work in entertainment (acting, music, or similar)?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for work in entertainment (acting, music, or similar)?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q3
1 · Guesser asks

Is this person primarily known for politics or government?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for politics or government?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q4
1 · Guesser reply rejectedTurn counted

Reply in the wrong format.

Invalid action · fixed format reminder sent

Why this reply was rejectedInvalid action · turn 4

The reply was JSON, but it matched neither the ASK nor the GUESS format.

Recorded type
invalid_action
Turn
4
Counted
Yes · question 4
Feedback
FORMAT_ERROR
What the model returnedExact text from the provider
Attempt 1 · finish stop
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for sports?"},"_dummy":null}
What a valid reply looks likeThe same formats shown to the Guesser
ASK
{"result":{"action":"ASK","description":null,"name":null,"question":"<non-empty yes-or-no property question>"}}
GUESS
{"result":{"action":"GUESS","description":"<non-empty identifying description>","name":"<non-empty candidate name>","question":null}}

A reply must match one of these formats exactly. Text fields in use must not be empty; unused fields must be JSON null.

2 · System responseFORMAT
Q5
1 · Guesser asks

Is this person primarily known for sports or athletics?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for sports or athletics?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q6
1 · Guesser asks

Is this person primarily known for science, technology, or computing?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for science, technology, or computing?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q7
1 · Guesser asks

Is this person primarily known for business or entrepreneurship?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for business or entrepreneurship?"}}
Oracle evidence1 model-reported source

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q8
1 · Guesser asks

Is this person primarily known for writing or literature?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known for writing or literature?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q9
1 · Guesser asks

Is this person primarily known as a fiction writer or novelist?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily known as a fiction writer or novelist?"}}
Oracle evidence1 model-reported source

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q10
1 · Guesser asks

Is this person known for writing fantasy or science fiction?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person known for writing fantasy or science fiction?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q11
1 · Guesser asks

Is this person American?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person American?"}}
Oracle evidence1 model-reported source

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q12
1 · Guesser asks

Is this person known for writing fantasy specifically rather than science fiction?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person known for writing fantasy specifically rather than science fiction?"}}
2 · Answer check returns Unknown
Q13
1 · Guesser asks

Is this person male?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person male?"}}
Oracle evidence1 model-reported source

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q14
1 · Guesser asks

Has this person won a Hugo Award for Best Novel?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Has this person won a Hugo Award for Best Novel?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q15
1 · Guesser asks

Has this person won a Nebula Award for Best Novel?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Has this person won a Nebula Award for Best Novel?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q16
1 · Guesser asks

Has this person published more than ten novels?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Has this person published more than ten novels?"}}
Oracle evidence1 model-reported source

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
Q17
1 · Guesser asks

Is this person known for writing a long-running series of urban fantasy detective novels?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person known for writing a long-running series of urban fantasy detective novels?"}}
Oracle evidence3 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns No
Q18
1 · Guesser asks

Has an HBO television series been adapted from this person's novels?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Has an HBO television series been adapted from this person's novels?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

Source 01
About This Series Based on Stephen King’s novel, this psychological thriller follows the grim murder of a boy and the mysterious forces surrounding the case.
Open hbomax.com
2 · Answer check returns Yes
Q19
1 · Guesser guesses

George R. R. Martin

American novelist best known for the A Song of Ice and Fire fantasy series adapted into HBO's Game of Thrones.

Validator rationale The trusted subject is Stephen King, while the proposed identity is George R. R. Martin, a different author.

Recorded Guesser outputRequired JSON format
{"result":{"action":"GUESS","description":"American novelist best known for the A Song of Ice and Fire fantasy series adapted into HBO's Game of Thrones.","name":"George R. R. Martin","question":null}}
2 · Validator returns No
Q20
1 · Guesser asks

Is this person primarily associated with horror fiction?

Recorded Guesser outputRequired JSON format
{"result":{"action":"ASK","description":null,"name":null,"question":"Is this person primarily associated with horror fiction?"}}
Oracle evidence2 model-reported sources

The Oracle reported these excerpts and URLs. They are not independently checked.

2 · Answer check returns Yes
G
1 · Guesser guessesNot counted

Stephen King

Living American male horror novelist with many novels, including works adapted as HBO's The Outsider.

Validator rationale The proposed identity matches Stephen King, the American horror novelist; the description and works are consistent with the trusted subject.

Recorded Guesser outputRequired JSON format
{"result":{"action":"GUESS","description":"Living American male horror novelist with many novels, including works adapted as HBO's The Outsider.","name":"Stephen King","question":null}}
2 · Validator returns Yes