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
Sparse tensor: row selection matches columns instead of rows · case 01
row selection matches columns instead of rows.
Sparse tensor: row selection matches columns instead of rows · case 02
row selection matches columns instead of rows.
Sparse tensor: row selection matches columns instead of rows · case 03
row selection matches columns instead of rows.
Sparse tensor: row selection matches columns instead of rows · case 04
row selection matches columns instead of rows.
Sparse tensor: row selection matches columns instead of rows · case 05
row selection matches columns instead of rows.
Sparse tensor: column projection returns all entries instead of one column · case 01
column projection returns all entries instead of one column.
Sparse tensor: column projection returns all entries instead of one column · case 02
column projection returns all entries instead of one column.
Sparse tensor: column projection returns all entries instead of one column · case 03
column projection returns all entries instead of one column.
Sparse tensor: column projection returns all entries instead of one column · case 04
column projection returns all entries instead of one column.
Sparse tensor: column projection returns all entries instead of one column · case 05
column projection returns all entries instead of one column.
Sparse tensor: stored entry count is reported instead of logical nonzero count · case 01
stored entry count is reported instead of logical nonzero count.
Sparse tensor: stored entry count is reported instead of logical nonzero count · case 02
stored entry count is reported instead of logical nonzero count.
Sparse tensor: stored entry count is reported instead of logical nonzero count · case 03
stored entry count is reported instead of logical nonzero count.
Sparse tensor: stored entry count is reported instead of logical nonzero count · case 04
stored entry count is reported instead of logical nonzero count.
Sparse tensor: stored entry count is reported instead of logical nonzero count · case 05
stored entry count is reported instead of logical nonzero count.
Sparse tensor: dense materialization overwrites duplicate coordinates · case 01
dense materialization overwrites duplicate coordinates.
Sparse tensor: dense materialization overwrites duplicate coordinates · case 02
dense materialization overwrites duplicate coordinates.
Sparse tensor: dense materialization overwrites duplicate coordinates · case 03
dense materialization overwrites duplicate coordinates.
Sparse tensor: dense materialization overwrites duplicate coordinates · case 04
dense materialization overwrites duplicate coordinates.
Sparse tensor: dense materialization overwrites duplicate coordinates · case 05
dense materialization overwrites duplicate coordinates.
Sparse tensor: CSR row-pointer intervals are treated as nonempty row markers · case 01
CSR row-pointer intervals are treated as nonempty row markers.
Sparse tensor: CSR row-pointer intervals are treated as nonempty row markers · case 02
CSR row-pointer intervals are treated as nonempty row markers.
Sparse tensor: CSR row-pointer intervals are treated as nonempty row markers · case 03
CSR row-pointer intervals are treated as nonempty row markers.
Sparse tensor: CSR row-pointer intervals are treated as nonempty row markers · case 04
CSR row-pointer intervals are treated as nonempty row markers.
Sparse tensor: CSR row-pointer intervals are treated as nonempty row markers · case 05
CSR row-pointer intervals are treated as nonempty row markers.
Sparse tensor: row gather returns the pointer end as an inclusive bound · case 01
row gather returns the pointer end as an inclusive bound.
Sparse tensor: row gather returns the pointer end as an inclusive bound · case 02
row gather returns the pointer end as an inclusive bound.
Sparse tensor: row gather returns the pointer end as an inclusive bound · case 03
row gather returns the pointer end as an inclusive bound.
Sparse tensor: row gather returns the pointer end as an inclusive bound · case 04
row gather returns the pointer end as an inclusive bound.
Sparse tensor: row gather returns the pointer end as an inclusive bound · case 05
row gather returns the pointer end as an inclusive bound.
Sparse tensor: negative COO coordinates are accepted as Python indices · case 01
negative COO coordinates are accepted as Python indices.
Sparse tensor: negative COO coordinates are accepted as Python indices · case 02
negative COO coordinates are accepted as Python indices.
Sparse tensor: negative COO coordinates are accepted as Python indices · case 03
negative COO coordinates are accepted as Python indices.
Sparse tensor: negative COO coordinates are accepted as Python indices · case 04
negative COO coordinates are accepted as Python indices.
Sparse tensor: negative COO coordinates are accepted as Python indices · case 05
negative COO coordinates are accepted as Python indices.
Sparse tensor: CSR conversion assigns entries using value offsets as row numbers · case 01
CSR conversion assigns entries using value offsets as row numbers.
Sparse tensor: CSR conversion assigns entries using value offsets as row numbers · case 02
CSR conversion assigns entries using value offsets as row numbers.
Sparse tensor: CSR conversion assigns entries using value offsets as row numbers · case 03
CSR conversion assigns entries using value offsets as row numbers.
Sparse tensor: CSR conversion assigns entries using value offsets as row numbers · case 04
CSR conversion assigns entries using value offsets as row numbers.
Sparse tensor: CSR conversion assigns entries using value offsets as row numbers · case 05
CSR conversion assigns entries using value offsets as row numbers.
Top-k routing follows expert names instead of gate scores · case 01
Top-k routing follows expert names instead of gate scores.
Top-k routing follows expert names instead of gate scores · case 02
Top-k routing follows expert names instead of gate scores.
Top-k routing follows expert names instead of gate scores · case 03
Top-k routing follows expert names instead of gate scores.
Top-k routing follows expert names instead of gate scores · case 04
Top-k routing follows expert names instead of gate scores.
Top-k routing follows expert names instead of gate scores · case 05
Top-k routing follows expert names instead of gate scores.
Equal gate scores route according to arrival order · case 01
Equal gate scores route according to arrival order.
Equal gate scores route according to arrival order · case 02
Equal gate scores route according to arrival order.
Equal gate scores route according to arrival order · case 03
Equal gate scores route according to arrival order.
Equal gate scores route according to arrival order · case 04
Equal gate scores route according to arrival order.
Equal gate scores route according to arrival order · case 05
Equal gate scores route according to arrival order.
One expert filling its quota blocks unrelated experts · case 01
One expert filling its quota blocks unrelated experts.
One expert filling its quota blocks unrelated experts · case 02
One expert filling its quota blocks unrelated experts.
One expert filling its quota blocks unrelated experts · case 03
One expert filling its quota blocks unrelated experts.
One expert filling its quota blocks unrelated experts · case 04
One expert filling its quota blocks unrelated experts.
One expert filling its quota blocks unrelated experts · case 05
One expert filling its quota blocks unrelated experts.
Overflow tokens disappear despite available alternatives · case 01
Overflow tokens disappear despite available alternatives.
Overflow tokens disappear despite available alternatives · case 02
Overflow tokens disappear despite available alternatives.
Overflow tokens disappear despite available alternatives · case 03
Overflow tokens disappear despite available alternatives.
Overflow tokens disappear despite available alternatives · case 04
Overflow tokens disappear despite available alternatives.
Overflow tokens disappear despite available alternatives · case 05
Overflow tokens disappear despite available alternatives.
Selected gate weights retain probability mass from discarded experts · case 01
Selected gate weights retain probability mass from discarded experts.
Selected gate weights retain probability mass from discarded experts · case 02
Selected gate weights retain probability mass from discarded experts.
Selected gate weights retain probability mass from discarded experts · case 03
Selected gate weights retain probability mass from discarded experts.
Selected gate weights retain probability mass from discarded experts · case 04
Selected gate weights retain probability mass from discarded experts.
Selected gate weights retain probability mass from discarded experts · case 05
Selected gate weights retain probability mass from discarded experts.
Dropped routes make surviving token weights sum to less than one · case 01
Dropped routes make surviving token weights sum to less than one.
Dropped routes make surviving token weights sum to less than one · case 02
Dropped routes make surviving token weights sum to less than one.
Dropped routes make surviving token weights sum to less than one · case 03
Dropped routes make surviving token weights sum to less than one.
Dropped routes make surviving token weights sum to less than one · case 04
Dropped routes make surviving token weights sum to less than one.
Dropped routes make surviving token weights sum to less than one · case 05
Dropped routes make surviving token weights sum to less than one.
Second expert is omitted from a top-two token dispatch · case 01
Second expert is omitted from a top-two token dispatch.
Second expert is omitted from a top-two token dispatch · case 02
Second expert is omitted from a top-two token dispatch.
Second expert is omitted from a top-two token dispatch · case 03
Second expert is omitted from a top-two token dispatch.
Second expert is omitted from a top-two token dispatch · case 04
Second expert is omitted from a top-two token dispatch.
Second expert is omitted from a top-two token dispatch · case 05
Second expert is omitted from a top-two token dispatch.
Low-confidence routes survive the gate threshold · case 01
Low-confidence routes survive the gate threshold.
Low-confidence routes survive the gate threshold · case 02
Low-confidence routes survive the gate threshold.
Low-confidence routes survive the gate threshold · case 03
Low-confidence routes survive the gate threshold.
Low-confidence routes survive the gate threshold · case 04
Low-confidence routes survive the gate threshold.
Low-confidence routes survive the gate threshold · case 05
Low-confidence routes survive the gate threshold.
Shared expert output is lost when routed experts are empty · case 01
Shared expert output is lost when routed experts are empty.
Shared expert output is lost when routed experts are empty · case 02
Shared expert output is lost when routed experts are empty.
Shared expert output is lost when routed experts are empty · case 03
Shared expert output is lost when routed experts are empty.
Shared expert output is lost when routed experts are empty · case 04
Shared expert output is lost when routed experts are empty.
Shared expert output is lost when routed experts are empty · case 05
Shared expert output is lost when routed experts are empty.
Top-k selection crosses a group boundary · case 01
Top-k selection crosses a group boundary.
Top-k selection crosses a group boundary · case 02
Top-k selection crosses a group boundary.
Top-k selection crosses a group boundary · case 03
Top-k selection crosses a group boundary.
Top-k selection crosses a group boundary · case 04
Top-k selection crosses a group boundary.
Top-k selection crosses a group boundary · case 05
Top-k selection crosses a group boundary.
Expert load counts routes rather than distinct tokens · case 01
Expert load counts routes rather than distinct tokens.
Expert load counts routes rather than distinct tokens · case 02
Expert load counts routes rather than distinct tokens.
Expert load counts routes rather than distinct tokens · case 03
Expert load counts routes rather than distinct tokens.
Expert load counts routes rather than distinct tokens · case 04
Expert load counts routes rather than distinct tokens.
Expert load counts routes rather than distinct tokens · case 05
Expert load counts routes rather than distinct tokens.
Expert outputs are added without their gate weights · case 01
Expert outputs are added without their gate weights.
Expert outputs are added without their gate weights · case 02
Expert outputs are added without their gate weights.
Expert outputs are added without their gate weights · case 03
Expert outputs are added without their gate weights.
Expert outputs are added without their gate weights · case 04
Expert outputs are added without their gate weights.
Expert outputs are added without their gate weights · case 05
Expert outputs are added without their gate weights.
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 ↗