Reasoning · spatial
SpatialComing soon
Mental rotation, projection, and relative position from generated text configurations.
- Source
- Generated
- Score
- ModelAxes score, λ = 1
- Ladder
- structural · operation count · 1, 2, 4, 6, 10, 16
- Items
- Not built yet
- Dimensions
- stream (4)
No results yet: this benchmark is planned and is not part of any score. Everything on this page describes the plan and may change before it ships.
The plan
Draft specWritten up in a draft specification; details may change once items are built and piloted.
What it will measure
Tracking positions, orientations and viewpoints described in text.
How items will be built
- Four streams: navigation (turns and moves; final heading and position), grid transforms (rotate or reflect a labeled grid; where is an object now), cube nets (which face is opposite, or on top after rolling) and viewpoints (the order of objects seen from another side of a table).
- Rungs set the number of operations (1, 2, 4, 6, 10, 16); grids and viewpoints also add objects as rungs rise.
- The ground truth comes from exact simulation.
How it will be scored
- Answers in a fixed format, graded by code; +1 / 0 / −1.
- x50 per stream, and tagged frame-of-reference errors (the right answer in the wrong frame, such as left and right swapped).
Example (burned: never used in a scored release)
You are standing facing north. Turn right 90°, walk 2 m. Turn right 90°, walk 3 m. Turn left 90°, walk 1 m. Which direction are you facing, and where are you relative to your start?
AnswerFacing east; 3 m east and 3 m south of the start.
Planned size900 items (6 rungs × 150, streams balanced).