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
Streaming accumulation overwrites its prior numerator · case 01
Streaming accumulation overwrites its prior numerator.
Streaming accumulation overwrites its prior numerator · case 02
Streaming accumulation overwrites its prior numerator.
Streaming accumulation overwrites its prior numerator · case 03
Streaming accumulation overwrites its prior numerator.
Streaming accumulation overwrites its prior numerator · case 04
Streaming accumulation overwrites its prior numerator.
Streaming accumulation overwrites its prior numerator · case 05
Streaming accumulation overwrites its prior numerator.
Resumed accumulation restores the numerator but loses count · case 01
Resumed accumulation restores the numerator but loses count.
Resumed accumulation restores the numerator but loses count · case 02
Resumed accumulation restores the numerator but loses count.
Resumed accumulation restores the numerator but loses count · case 03
Resumed accumulation restores the numerator but loses count.
Resumed accumulation restores the numerator but loses count · case 04
Resumed accumulation restores the numerator but loses count.
Resumed accumulation restores the numerator but loses count · case 05
Resumed accumulation restores the numerator but loses count.
Flush emits a normalized result without clearing its accumulator · case 01
Flush emits a normalized result without clearing its accumulator.
Flush emits a normalized result without clearing its accumulator · case 02
Flush emits a normalized result without clearing its accumulator.
Flush emits a normalized result without clearing its accumulator · case 03
Flush emits a normalized result without clearing its accumulator.
Flush emits a normalized result without clearing its accumulator · case 04
Flush emits a normalized result without clearing its accumulator.
Flush emits a normalized result without clearing its accumulator · case 05
Flush emits a normalized result without clearing its accumulator.
Accumulation flushes when the threshold is reached rather than exceeded · case 01
Accumulation flushes when the threshold is reached rather than exceeded.
Accumulation flushes when the threshold is reached rather than exceeded · case 02
Accumulation flushes when the threshold is reached rather than exceeded.
Accumulation flushes when the threshold is reached rather than exceeded · case 03
Accumulation flushes when the threshold is reached rather than exceeded.
Accumulation flushes when the threshold is reached rather than exceeded · case 04
Accumulation flushes when the threshold is reached rather than exceeded.
Accumulation flushes when the threshold is reached rather than exceeded · case 05
Accumulation flushes when the threshold is reached rather than exceeded.
The final short batch is discarded from accumulation · case 01
The final short batch is discarded from accumulation.
The final short batch is discarded from accumulation · case 02
The final short batch is discarded from accumulation.
The final short batch is discarded from accumulation · case 03
The final short batch is discarded from accumulation.
The final short batch is discarded from accumulation · case 04
The final short batch is discarded from accumulation.
The final short batch is discarded from accumulation · case 05
The final short batch is discarded from accumulation.
Running mean state treats a batch mean as a batch sum · case 01
Running mean state treats a batch mean as a batch sum.
Running mean state treats a batch mean as a batch sum · case 02
Running mean state treats a batch mean as a batch sum.
Running mean state treats a batch mean as a batch sum · case 03
Running mean state treats a batch mean as a batch sum.
Running mean state treats a batch mean as a batch sum · case 04
Running mean state treats a batch mean as a batch sum.
Running mean state treats a batch mean as a batch sum · case 05
Running mean state treats a batch mean as a batch sum.
Two objective buckets share one normalization denominator · case 01
Two objective buckets share one normalization denominator.
Two objective buckets share one normalization denominator · case 02
Two objective buckets share one normalization denominator.
Two objective buckets share one normalization denominator · case 03
Two objective buckets share one normalization denominator.
Two objective buckets share one normalization denominator · case 04
Two objective buckets share one normalization denominator.
Two objective buckets share one normalization denominator · case 05
Two objective buckets share one normalization denominator.
Microbatch gradients are summed when the objective expects their mean · case 01
Microbatch gradients are summed when the objective expects their mean.
Microbatch gradients are summed when the objective expects their mean · case 02
Microbatch gradients are summed when the objective expects their mean.
Microbatch gradients are summed when the objective expects their mean · case 03
Microbatch gradients are summed when the objective expects their mean.
Microbatch gradients are summed when the objective expects their mean · case 04
Microbatch gradients are summed when the objective expects their mean.
Microbatch gradients are summed when the objective expects their mean · case 05
Microbatch gradients are summed when the objective expects their mean.
SGD adds the gradient instead of taking a descent step · case 01
SGD adds the gradient instead of taking a descent step.
SGD adds the gradient instead of taking a descent step · case 02
SGD adds the gradient instead of taking a descent step.
SGD adds the gradient instead of taking a descent step · case 03
SGD adds the gradient instead of taking a descent step.
SGD adds the gradient instead of taking a descent step · case 04
SGD adds the gradient instead of taking a descent step.
SGD adds the gradient instead of taking a descent step · case 05
SGD adds the gradient instead of taking a descent step.
Momentum ignores the prior velocity · case 01
Momentum ignores the prior velocity.
Momentum ignores the prior velocity · case 02
Momentum ignores the prior velocity.
Momentum ignores the prior velocity · case 03
Momentum ignores the prior velocity.
Momentum ignores the prior velocity · case 04
Momentum ignores the prior velocity.
Momentum ignores the prior velocity · case 05
Momentum ignores the prior velocity.
Momentum dampening scales the carried buffer · case 01
Momentum dampening scales the carried buffer.
Momentum dampening scales the carried buffer · case 02
Momentum dampening scales the carried buffer.
Momentum dampening scales the carried buffer · case 03
Momentum dampening scales the carried buffer.
Momentum dampening scales the carried buffer · case 04
Momentum dampening scales the carried buffer.
Momentum dampening scales the carried buffer · case 05
Momentum dampening scales the carried buffer.
Nesterov update uses the ordinary momentum direction · case 01
Nesterov update uses the ordinary momentum direction.
Nesterov update uses the ordinary momentum direction · case 02
Nesterov update uses the ordinary momentum direction.
Nesterov update uses the ordinary momentum direction · case 03
Nesterov update uses the ordinary momentum direction.
Nesterov update uses the ordinary momentum direction · case 04
Nesterov update uses the ordinary momentum direction.
Nesterov update uses the ordinary momentum direction · case 05
Nesterov update uses the ordinary momentum direction.
Momentum state stores the descent direction rather than gradient history · case 01
Momentum state stores the descent direction rather than gradient history.
Momentum state stores the descent direction rather than gradient history · case 02
Momentum state stores the descent direction rather than gradient history.
Momentum state stores the descent direction rather than gradient history · case 03
Momentum state stores the descent direction rather than gradient history.
Momentum state stores the descent direction rather than gradient history · case 04
Momentum state stores the descent direction rather than gradient history.
Momentum state stores the descent direction rather than gradient history · case 05
Momentum state stores the descent direction rather than gradient history.
Adam first moment drops its exponential history · case 01
Adam first moment drops its exponential history.
Adam first moment drops its exponential history · case 02
Adam first moment drops its exponential history.
Adam first moment drops its exponential history · case 03
Adam first moment drops its exponential history.
Adam first moment drops its exponential history · case 04
Adam first moment drops its exponential history.
Adam first moment drops its exponential history · case 05
Adam first moment drops its exponential history.
Adam second moment keeps a signed gradient · case 01
Adam second moment keeps a signed gradient.
Adam second moment keeps a signed gradient · case 02
Adam second moment keeps a signed gradient.
Adam second moment keeps a signed gradient · case 03
Adam second moment keeps a signed gradient.
Adam second moment keeps a signed gradient · case 04
Adam second moment keeps a signed gradient.
Adam second moment keeps a signed gradient · case 05
Adam second moment keeps a signed gradient.
Adam bias correction reuses the previous update exponent · case 01
Adam bias correction reuses the previous update exponent.
Adam bias correction reuses the previous update exponent · case 02
Adam bias correction reuses the previous update exponent.
Adam bias correction reuses the previous update exponent · case 03
Adam bias correction reuses the previous update exponent.
Adam bias correction reuses the previous update exponent · case 04
Adam bias correction reuses the previous update exponent.
Adam bias correction reuses the previous update exponent · case 05
Adam bias correction reuses the previous update exponent.
Adam epsilon is scaled by the second-moment correction · case 01
Adam epsilon is scaled by the second-moment correction.
Adam epsilon is scaled by the second-moment correction · case 02
Adam epsilon is scaled by the second-moment correction.
Adam epsilon is scaled by the second-moment correction · case 03
Adam epsilon is scaled by the second-moment correction.
Adam epsilon is scaled by the second-moment correction · case 04
Adam epsilon is scaled by the second-moment correction.
Adam epsilon is scaled by the second-moment correction · case 05
Adam epsilon is scaled by the second-moment correction.
AMSGrad forgets the historical maximum variance · case 01
AMSGrad forgets the historical maximum variance.
AMSGrad forgets the historical maximum variance · case 02
AMSGrad forgets the historical maximum variance.
AMSGrad forgets the historical maximum variance · case 03
AMSGrad forgets the historical maximum variance.
AMSGrad forgets the historical maximum variance · case 04
AMSGrad forgets the historical maximum variance.
AMSGrad forgets the historical maximum variance · case 05
AMSGrad forgets the historical maximum variance.
Optimizer state follows list position instead of parameter identity · case 01
Optimizer state follows list position instead of parameter identity.
Optimizer state follows list position instead of parameter identity · case 02
Optimizer state follows list position instead of parameter identity.
Optimizer state follows list position instead of parameter identity · case 03
Optimizer state follows list position instead of parameter identity.
Optimizer state follows list position instead of parameter identity · case 04
Optimizer state follows list position instead of parameter identity.
Optimizer state follows list position instead of parameter identity · case 05
Optimizer state follows list position instead of parameter identity.
Adam local update leaves its parameter step counter stale · case 01
Adam local update leaves its parameter step counter stale.
Adam local update leaves its parameter step counter stale · case 02
Adam local update leaves its parameter step counter stale.
Adam local update leaves its parameter step counter stale · case 03
Adam local update leaves its parameter step counter stale.
Adam local update leaves its parameter step counter stale · case 04
Adam local update leaves its parameter step counter stale.
Adam local update leaves its parameter step counter stale · case 05
Adam local update leaves its parameter step counter stale.
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 ↗