Understand the failure.
Verify the repair.
Small, reproducible software failures. The broken implementation, the fix that didn’t work, and the one that passed—preserved together.
Explore the cases ↓How results are verified ↗WHAT THE ARCHIVE CONTAINS
100840 executable cases. 20168 are open.
Every case records the implementation that fails, the fix that did not work, and the repair that passed its checks—with recorded outputs and source hashes. This release adds 100840 cases across 20168 failure mechanisms and 254 domains.
The open tier gives you the failure and the unsuccessful fix for one case in every mechanism. The remaining 80672 cases, 5 variants per mechanism, are member-only: the verified repair, its recorded checks, and the full fixture suite are held in the member archive. Read the methodology ↗
A RECORD OF WHAT WENT WRONG
Browse the archive / 100840
x crossing interval keeps the direction sign · case 01
Leftward rays step x repeatedly without advancing the boundary time.
x crossing interval keeps the direction sign · case 02
Leftward rays step x repeatedly without advancing the boundary time.
x crossing interval keeps the direction sign · case 03
Leftward rays step x repeatedly without advancing the boundary time.
x crossing interval keeps the direction sign · case 04
Leftward rays step x repeatedly without advancing the boundary time.
x crossing interval keeps the direction sign · case 05
Leftward rays step x repeatedly without advancing the boundary time.
Ray exactly through a grid corner steps y first · case 01
Diagonal rays visit a different neighbor than the contract and cached tests.
Ray exactly through a grid corner steps y first · case 02
Diagonal rays visit a different neighbor than the contract and cached tests.
Ray exactly through a grid corner steps y first · case 03
Diagonal rays visit a different neighbor than the contract and cached tests.
Ray exactly through a grid corner steps y first · case 04
Diagonal rays visit a different neighbor than the contract and cached tests.
Ray exactly through a grid corner steps y first · case 05
Diagonal rays visit a different neighbor than the contract and cached tests.
Traversal stops before a boundary reached exactly at max_t · case 01
A ray ending exactly on a cell border misses that cell.
Traversal stops before a boundary reached exactly at max_t · case 02
A ray ending exactly on a cell border misses that cell.
Traversal stops before a boundary reached exactly at max_t · case 03
A ray ending exactly on a cell border misses that cell.
Traversal stops before a boundary reached exactly at max_t · case 04
A ray ending exactly on a cell border misses that cell.
Traversal stops before a boundary reached exactly at max_t · case 05
A ray ending exactly on a cell border misses that cell.
First y boundary ignores the ray origin · case 01
Rays that do not start at y=0 cross rows at wrong times.
First y boundary ignores the ray origin · case 02
Rays that do not start at y=0 cross rows at wrong times.
First y boundary ignores the ray origin · case 03
Rays that do not start at y=0 cross rows at wrong times.
First y boundary ignores the ray origin · case 04
Rays that do not start at y=0 cross rows at wrong times.
First y boundary ignores the ray origin · case 05
Rays that do not start at y=0 cross rows at wrong times.
Grid level uses the smaller box extent · case 01
Long thin objects are placed on too fine a level and span many cells.
Grid level uses the smaller box extent · case 02
Long thin objects are placed on too fine a level and span many cells.
Grid level uses the smaller box extent · case 03
Long thin objects are placed on too fine a level and span many cells.
Grid level uses the smaller box extent · case 04
Long thin objects are placed on too fine a level and span many cells.
Grid level uses the smaller box extent · case 05
Long thin objects are placed on too fine a level and span many cells.
Objects exactly one cell wide are promoted a level · case 01
Grid level populations shift upward and coarse levels get crowded.
Objects exactly one cell wide are promoted a level · case 02
Grid level populations shift upward and coarse levels get crowded.
Objects exactly one cell wide are promoted a level · case 03
Grid level populations shift upward and coarse levels get crowded.
Objects exactly one cell wide are promoted a level · case 04
Grid level populations shift upward and coarse levels get crowded.
Objects exactly one cell wide are promoted a level · case 05
Grid level populations shift upward and coarse levels get crowded.
Level search can run past the last grid level · case 01
Huge objects are assigned to a level that does not exist.
Level search can run past the last grid level · case 02
Huge objects are assigned to a level that does not exist.
Level search can run past the last grid level · case 03
Huge objects are assigned to a level that does not exist.
Level search can run past the last grid level · case 04
Huge objects are assigned to a level that does not exist.
Level search can run past the last grid level · case 05
Huge objects are assigned to a level that does not exist.
Home cell index uses the base cell size · case 01
Objects on coarse levels are hashed to non-existent fine cells.
Home cell index uses the base cell size · case 02
Objects on coarse levels are hashed to non-existent fine cells.
Home cell index uses the base cell size · case 03
Objects on coarse levels are hashed to non-existent fine cells.
Home cell index uses the base cell size · case 04
Objects on coarse levels are hashed to non-existent fine cells.
Home cell index uses the base cell size · case 05
Objects on coarse levels are hashed to non-existent fine cells.
Home cell is computed from the box centre · case 01
Neighbor search offsets assume the min corner and miss overlaps.
Home cell is computed from the box centre · case 02
Neighbor search offsets assume the min corner and miss overlaps.
Home cell is computed from the box centre · case 03
Neighbor search offsets assume the min corner and miss overlaps.
Home cell is computed from the box centre · case 04
Neighbor search offsets assume the min corner and miss overlaps.
Home cell is computed from the box centre · case 05
Neighbor search offsets assume the min corner and miss overlaps.
Level cell sizes grow linearly · case 01
Large objects need many levels and land on the wrong one.
Level cell sizes grow linearly · case 02
Large objects need many levels and land on the wrong one.
Level cell sizes grow linearly · case 03
Large objects need many levels and land on the wrong one.
Level cell sizes grow linearly · case 04
Large objects need many levels and land on the wrong one.
Level cell sizes grow linearly · case 05
Large objects need many levels and land on the wrong one.
Axis variance is computed from box minima · case 01
Scenes with mixed box sizes sweep along a poorly separating axis.
Axis variance is computed from box minima · case 02
Scenes with mixed box sizes sweep along a poorly separating axis.
Axis variance is computed from box minima · case 03
Scenes with mixed box sizes sweep along a poorly separating axis.
Axis variance is computed from box minima · case 04
Scenes with mixed box sizes sweep along a poorly separating axis.
Axis variance is computed from box minima · case 05
Scenes with mixed box sizes sweep along a poorly separating axis.
Axis variance uses the sample denominator · case 01
Reported spread disagrees with the population-variance contract.
Axis variance uses the sample denominator · case 02
Reported spread disagrees with the population-variance contract.
Axis variance uses the sample denominator · case 03
Reported spread disagrees with the population-variance contract.
Axis variance uses the sample denominator · case 04
Reported spread disagrees with the population-variance contract.
Axis variance uses the sample denominator · case 05
Reported spread disagrees with the population-variance contract.
Axis spread uses absolute deviation · case 01
Axis choice flips for distributions with outliers.
Axis spread uses absolute deviation · case 02
Axis choice flips for distributions with outliers.
Axis spread uses absolute deviation · case 03
Axis choice flips for distributions with outliers.
Axis spread uses absolute deviation · case 04
Axis choice flips for distributions with outliers.
Axis spread uses absolute deviation · case 05
Axis choice flips for distributions with outliers.
Centre uses the neighbouring axis maximum · case 01
Axis variances mix two axes.
Centre uses the neighbouring axis maximum · case 02
Axis variances mix two axes.
Centre uses the neighbouring axis maximum · case 03
Axis variances mix two axes.
Centre uses the neighbouring axis maximum · case 04
Axis variances mix two axes.
Centre uses the neighbouring axis maximum · case 05
Axis variances mix two axes.
Sweep order sorts by box maxima · case 01
Endpoint arrays are built out of order and pruning breaks.
Sweep order sorts by box maxima · case 02
Endpoint arrays are built out of order and pruning breaks.
Sweep order sorts by box maxima · case 03
Endpoint arrays are built out of order and pruning breaks.
Sweep order sorts by box maxima · case 04
Endpoint arrays are built out of order and pruning breaks.
Sweep order sorts by box maxima · case 05
Endpoint arrays are built out of order and pruning breaks.
Variance is measured around the midrange · case 01
Skewed distributions report the wrong spread.
Variance is measured around the midrange · case 02
Skewed distributions report the wrong spread.
Variance is measured around the midrange · case 03
Skewed distributions report the wrong spread.
Variance is measured around the midrange · case 04
Skewed distributions report the wrong spread.
Variance is measured around the midrange · case 05
Skewed distributions report the wrong spread.
Swept box only translates the maximum corner · case 01
Objects moving left or down tunnel because the swept box shrinks.
Swept box only translates the maximum corner · case 02
Objects moving left or down tunnel because the swept box shrinks.
Swept box only translates the maximum corner · case 03
Objects moving left or down tunnel because the swept box shrinks.
Swept box only translates the maximum corner · case 04
Objects moving left or down tunnel because the swept box shrinks.
Swept box only translates the maximum corner · case 05
Objects moving left or down tunnel because the swept box shrinks.
Swept bounds use velocity instead of per-step displacement · case 01
Sweeps are dt^-1 times too long, flooding narrowphase with candidates.
Swept bounds use velocity instead of per-step displacement · case 02
Sweeps are dt^-1 times too long, flooding narrowphase with candidates.
Swept bounds use velocity instead of per-step displacement · case 03
Sweeps are dt^-1 times too long, flooding narrowphase with candidates.
Swept bounds use velocity instead of per-step displacement · case 04
Sweeps are dt^-1 times too long, flooding narrowphase with candidates.
Swept bounds use velocity instead of per-step displacement · case 05
Sweeps are dt^-1 times too long, flooding narrowphase with candidates.
Speculative margin is added only on the maximum sides · case 01
Contacts approaching from below or left start late.
Speculative margin is added only on the maximum sides · case 02
Contacts approaching from below or left start late.
Speculative margin is added only on the maximum sides · case 03
Contacts approaching from below or left start late.
Speculative margin is added only on the maximum sides · case 04
Contacts approaching from below or left start late.
Speculative margin is added only on the maximum sides · case 05
Contacts approaching from below or left start late.
Fast-mover test compares with the larger box extent · case 01
Thin fast objects are not flagged and tunnel through walls.
Fast-mover test compares with the larger box extent · case 02
Thin fast objects are not flagged and tunnel through walls.
Fast-mover test compares with the larger box extent · case 03
Thin fast objects are not flagged and tunnel through walls.
Fast-mover test compares with the larger box extent · case 04
Thin fast objects are not flagged and tunnel through walls.
Fast-mover test compares with the larger box extent · case 05
Thin fast objects are not flagged and tunnel through walls.
INSPECTABLE BY DESIGN
Every result has a runnable source.
Runnable implementations with recorded outputs, source hashes, and explicit contracts. Related variants share a failure mechanism and belong together in evaluation splits.
Read the methodology ↗