feat(agents): replay-backed spatial goal-selection evals - #3307

Open

mc856 wants to merge 18 commits into

dimensionalOS:mainfrom

mc856:feat/agents-spatial-goal-eval

Commits on Jul 31, 2026

feat(agents): add spatial goal-selection eval contracts, harness and scorer
mc856
committed
feat(agents): add replay-derived reference generation for spatial evals
mc856
committed
test(agents): three-layer test suite for the spatial goal-selection evals
mc856
committed
fix(agents): tighten eval scoring semantics and regenerate references
mc856
committed
fix(agents): harden eval scoring semantics and observability
mc856
committed
fix(agents): drop wall-referenced question, sharpness-based review crops
mc856
committed
docs(agents): document the reference review protocol
mc856
committed
feat(agents): multi-dataset reference schema, add go2_short question set
mc856
committed
fix(agents): address self-review findings in docs, guards and fixture provenance
mc856
committed
feat(agents): let the spatial eval sweep caller-named model arms
mc856
committed
fix(agents): follow SpatialMemory move to perception/experimental
mc856
committed
docs(agents): review crops regenerate locally instead of shipping in-tree
mc856
committed
fix(agents): move eval test-support modules under testing/
mc856
committed
fix(agents): drop __init__.py — the tree uses namespace packages
mc856
committed