Examenos

Correct answers over all 2,157 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Equal-weight mean over the 43 categories.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Accuracy on the 140 questions new in v4.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 140

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Correct answers over usable responses only; shown beside coverage.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Share of the 2,157 inputs with a usable response.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Pooled primary client time, network and failed requests included.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Pooled primary client time, network and failed requests included.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Lookup and filtering: 43 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 43

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Forward conditionals: 46 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 46

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Necessary conditions: 52 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 52

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Boolean scope: 54 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 54

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Quantifiers and sets: 49 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 49

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Comparisons: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Scheduling: 76 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 76

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

One-to-one assignments: 67 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 67

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Exact truth counting: 71 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 71

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Truth-teller consistency: 74 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 74

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Sufficiency and inconsistency: 41 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 41

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Object swaps: 52 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 52

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Sequential procedures: 48 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 48

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Temporal intervals: 42 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 42

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Spatial tracking: 49 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 49

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Reference binding: 50 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 50

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Beliefs and perspectives: 50 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 50

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Causal intervention: 50 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 50

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Exceptions and priorities: 58 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 58

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Planning: 43 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 43

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Directed reachability: 43 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 43

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Preference ordering: 51 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 51

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Rule counterexamples: 52 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 52

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Authority and stale records: 43 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 43

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Fault diagnosis: 50 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 50

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Probabilistic risk: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Value of information: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Robust decisions without a prior: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Strategic decisions: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Concurrent-history consistency: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Information-flow security: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Understanding requests and routing: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Recognizing multiple requested actions: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Tracking current conversational intention: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Selecting a tool or requesting clarification: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Checking authorization and scope: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Making decisions with untrusted content: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Applying a contextual content policy: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Extracting and normalizing facts: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Checking claims against supplied evidence: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Assigning severity or priority from a rubric: 40 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 40

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Numerical pattern recognition: 23 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 23

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Quantitative reasoning: 200 questions.

kettei-fat; Hosted native; collected 2026-10-20 to 2026-11-02 America/Toronto (3 days); one primary response per input, no scored retries. Identity: Captured serving declaration and returned aliases; immutable hidden weights not independently attested.

Exact aggregate from the release statistics.

Development benchmark, not an independently sampled test: 2,157 questions in 43 categories, the unchanged 2,017-question v3 bank plus 140 AQuA questions with source keys. The 23 numerical-pattern (MAIQ) and 200 quantitative (AQuA) questions are borrowed; the rest is predominantly project-developed with model-guided development, and Jev plus retained Microsoft and Drex feedback are development exposure. The AQuA additions were selected for misses by Fastino, Mammoth, GPT-6 Sol, D3 Ultra and Drex Max, which spreads but does not remove selection bias. Earlier primary responses are retained and newly added inputs were collected in later sessions, so one configuration can span several dates and routes. Strict accuracy counts an unusable response as not correct in the fixed denominator. Timing pools primary client times with the network included. The scoring key is private. A different number from v2 is a new bank, not a model change. Not ranked here.

Checked / reported: 2026-11-08 · Samples: 200

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Provider-reported charges for the primary calls on the v4 bank, in USD per 1,000 questions.

kettei-fat; System 1 v4 ledger.

All 2,157 charges reported.

Known run subtotal divided by 2,157 and multiplied by 1,000, not the bill for a separate 1,000-call run. A partial value is only the known subtotal: missing charges stay missing, so the true cost is higher by an unknown amount. Providers, routes, dates and access terms differ; these are not current tariffs. Free public access is not a price and has no row. Development benchmark; not ranked here.

Checked / reported: 2026-11-08 · Samples: 2157

Snapshot: 2026-11-06 · Revision: 224ee46dc1c16a8ecf2c154a2144b4d2fdc45de1 · Published summary observations; upstream task/result terms retained (AQuA-RAT Apache-2.0; MAIQ declared MIT, original question rights not established)

Decision Index 0.3.1 public suite, scored with the maintainer’s kit on its stratified 12,000-request sample (Figure 4, Text).

kettei fat. Figure 4 Text point, scored with the maintainer’s kit.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: 12000

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Median latency per request on the same run (Figure 4, Text).

kettei fat. Figure 4 Text point; hardware not stated.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: 12000

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Public half of the Decision Index vision board, scored with the board’s formula and anchored to its published scale (Figure 4, Images).

kettei fat. Figure 4 Images point.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: Not published

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Time per request on the same run (Figure 4, Images).

kettei fat. Figure 4 Images point; hardware not stated.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: Not published

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

The 231 public JevBench items (Figure 4, JevBench).

kettei fat. Figure 4 JevBench point.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: 231

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Median end-to-end latency per request from a UK client through the provider’s public API (Figure 4, JevBench).

kettei fat. Figure 4 JevBench point, timed end to end from the UK.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: 231

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Sqwish’s own held-out image set: 5,000 questions, 542 of them multi-image.

kettei fat. Held-out image set.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: 5000

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

SIB-200 topic classification on a 35-language sample.

kettei fat. 35-language sample; the post rounds to whole percent.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: Not published

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

The 19 languages in the SIB-200 sample not written in the Latin alphabet.

kettei fat. 19 non-Latin-script languages of the sample.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: Not published

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Tetris from a picture of the board, choosing among lettered landing spots; 48 games for kettei fat (Figure 6).

kettei fat. Figure 6 tetris panel.

Published in the vendor’s launch post.

Sqwish Labs’ own results in the kettei launch post of 14 October 2026. Figure values are the labelled data points the page draws (name, score, latency), read from the site bundle /static/js/main.10f99d62.js, read 8 November 2026; prose values are quoted from the same post. Values were published before the core and dot weights (18 October) and name no revision.

Checked / reported: 2026-11-08 · Samples: Not published

Snapshot: 2026-10-14 · Revision: site bundle /static/js/main.10f99d62.js, read 8 November 2026 · Published observations; result reuse terms not specified

Capability Score on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

Intelligence on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

Calibration on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

Cost Score on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

Modelled cost per 1,000 decisions on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

API latency, end to end on the rendered board.

kettei-fat (Sqwish) · Closed API (weights not public). Board key kettei-fat; method tag v1.7. Measured on the Second regular draw (v1.6.43), 2026-10-16; put on the v1.6.1 scale by the draw's anchor offset (see How it works). Source link https://docs.sqwish.ai.

Rendered board row. Hosted kettei-fat through docs.sqwish.ai; the API ID has no dated version.

Benchmark Heaven's rendered board, read 2 November 2026: 283 ranked systems · 1,500 decisions each (600 for 19 equated API re-runs) · 131 newer ranked rows tagged with their draw · 14 older rows dated separately. Newer rows are measured on dated draws and put on the v1.6.1 scale by the draw's anchor offset, so a row's draw is part of its identity. Distinct from Leanmcp JevBench, the community Decision Index and the stored Benchmark Heaven v1.6.1 feed. Cost is the source's model on a GPU-time basis for self-hosted rows, not a provider tariff. Self-hosted latency is timed on the GPU over loopback and API latency end to end with the network; the source says they are not directly comparable, so they stay in separate datasets. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-31 · Revision: Board v1.7.48, method v1.8, updated 2026-10-31; page sha256 df9093384c9cdf5854e065f71dd3abd83a127793f22f4ca70268e9981e0a0b17 · Published summary observations; upstream task/result terms retained

Capability on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained

Composite on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained

Intelligence on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained

Calibration on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained

Speed axis on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained

Cost axis on the core image track.

kettei-fat (Sqwish); board key kettei_fat; last measured 2026-10-16T17:44:49.892791+00:00.

Exact published core-track row.

Benchmark Heaven's image decision benchmark, core track of v0.3.0, built 2026-10-28T18:06:26.533607+00:00. Published core track only; dated carry and computer-use track are separate. Cite source_sha256 for the exact result bytes. Separate from JevBench text rows and the community vision index. H2O image rows are not stored here, because serving settings changed between releases and no row names the registered release. No Examenos recomputation.

Checked / reported: 2026-11-08 · Evaluated: 2026-10-16 · Samples: Not published

Snapshot: 2026-10-28 · Revision: v0.3.0; source_sha256=90ec4137edf485c5f6927c14d22188a6cea26e3355f0fd0e93d68da251c5e5d8 · Published summary observations; upstream task/result terms retained