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
Weighted Misra-Gries frequent items: weighted decrement uses one unit · case 01
A heavy newcomer loses only one unit and stale counters linger, violating the weighted algorithm.
Weighted Misra-Gries frequent items: weighted decrement uses one unit · case 02
A heavy newcomer loses only one unit and stale counters linger, violating the weighted algorithm.
Weighted Misra-Gries frequent items: weighted decrement uses one unit · case 03
A heavy newcomer loses only one unit and stale counters linger, violating the weighted algorithm.
Weighted Misra-Gries frequent items: weighted decrement uses one unit · case 04
A heavy newcomer loses only one unit and stale counters linger, violating the weighted algorithm.
Weighted Misra-Gries frequent items: weighted decrement uses one unit · case 05
A heavy newcomer loses only one unit and stale counters linger, violating the weighted algorithm.
Weighted Misra-Gries frequent items: discarded mass omits the newcomer · case 01
The reported error bound is smaller than the mass actually discarded.
Weighted Misra-Gries frequent items: discarded mass omits the newcomer · case 02
The reported error bound is smaller than the mass actually discarded.
Weighted Misra-Gries frequent items: discarded mass omits the newcomer · case 03
The reported error bound is smaller than the mass actually discarded.
Weighted Misra-Gries frequent items: discarded mass omits the newcomer · case 04
The reported error bound is smaller than the mass actually discarded.
Weighted Misra-Gries frequent items: discarded mass omits the newcomer · case 05
The reported error bound is smaller than the mass actually discarded.
Weighted Misra-Gries frequent items: one extra counter admitted · case 01
The summary holds k+1 counters, exceeding its memory budget and changing every later decrement.
Weighted Misra-Gries frequent items: one extra counter admitted · case 02
The summary holds k+1 counters, exceeding its memory budget and changing every later decrement.
Weighted Misra-Gries frequent items: one extra counter admitted · case 03
The summary holds k+1 counters, exceeding its memory budget and changing every later decrement.
Weighted Misra-Gries frequent items: one extra counter admitted · case 04
The summary holds k+1 counters, exceeding its memory budget and changing every later decrement.
Weighted Misra-Gries frequent items: one extra counter admitted · case 05
The summary holds k+1 counters, exceeding its memory budget and changing every later decrement.
Space-Saving heavy hitters: newcomer restarts at one · case 01
Evicted mass disappears, so counts underestimate and the overestimate-only property is lost.
Space-Saving heavy hitters: newcomer restarts at one · case 02
Evicted mass disappears, so counts underestimate and the overestimate-only property is lost.
Space-Saving heavy hitters: newcomer restarts at one · case 03
Evicted mass disappears, so counts underestimate and the overestimate-only property is lost.
Space-Saving heavy hitters: newcomer restarts at one · case 04
Evicted mass disappears, so counts underestimate and the overestimate-only property is lost.
Space-Saving heavy hitters: newcomer restarts at one · case 05
Evicted mass disappears, so counts underestimate and the overestimate-only property is lost.
Space-Saving heavy hitters: inherited error not recorded · case 01
Newcomers look like guaranteed heavy hitters although their count is mostly inherited.
Space-Saving heavy hitters: inherited error not recorded · case 02
Newcomers look like guaranteed heavy hitters although their count is mostly inherited.
Space-Saving heavy hitters: inherited error not recorded · case 03
Newcomers look like guaranteed heavy hitters although their count is mostly inherited.
Space-Saving heavy hitters: inherited error not recorded · case 04
Newcomers look like guaranteed heavy hitters although their count is mostly inherited.
Space-Saving heavy hitters: inherited error not recorded · case 05
Newcomers look like guaranteed heavy hitters although their count is mostly inherited.
Space-Saving heavy hitters: ties evict the most recently touched entry · case 01
A burst of distinct items keeps replacing the newest entry, churning the summary.
Space-Saving heavy hitters: ties evict the most recently touched entry · case 02
A burst of distinct items keeps replacing the newest entry, churning the summary.
Space-Saving heavy hitters: ties evict the most recently touched entry · case 03
A burst of distinct items keeps replacing the newest entry, churning the summary.
Space-Saving heavy hitters: ties evict the most recently touched entry · case 04
A burst of distinct items keeps replacing the newest entry, churning the summary.
Space-Saving heavy hitters: ties evict the most recently touched entry · case 05
A burst of distinct items keeps replacing the newest entry, churning the summary.
Space-Saving heavy hitters: hits do not refresh recency · case 01
Recently active entries are evicted as if they were stale.
Space-Saving heavy hitters: hits do not refresh recency · case 02
Recently active entries are evicted as if they were stale.
Space-Saving heavy hitters: hits do not refresh recency · case 03
Recently active entries are evicted as if they were stale.
Space-Saving heavy hitters: hits do not refresh recency · case 04
Recently active entries are evicted as if they were stale.
Space-Saving heavy hitters: hits do not refresh recency · case 05
Recently active entries are evicted as if they were stale.
Space-Saving heavy hitters: guarantee ignores the error bound · case 01
Items whose count is mostly inherited error are reported as guaranteed heavy hitters.
Space-Saving heavy hitters: guarantee ignores the error bound · case 02
Items whose count is mostly inherited error are reported as guaranteed heavy hitters.
Space-Saving heavy hitters: guarantee ignores the error bound · case 03
Items whose count is mostly inherited error are reported as guaranteed heavy hitters.
Space-Saving heavy hitters: guarantee ignores the error bound · case 04
Items whose count is mostly inherited error are reported as guaranteed heavy hitters.
Space-Saving heavy hitters: guarantee ignores the error bound · case 05
Items whose count is mostly inherited error are reported as guaranteed heavy hitters.
Space-Saving heavy hitters: equal counts listed in insertion order · case 01
Top-k reports with tied counts are ordered inconsistently between runs with different arrival orders.
Space-Saving heavy hitters: equal counts listed in insertion order · case 02
Top-k reports with tied counts are ordered inconsistently between runs with different arrival orders.
Space-Saving heavy hitters: equal counts listed in insertion order · case 03
Top-k reports with tied counts are ordered inconsistently between runs with different arrival orders.
Space-Saving heavy hitters: equal counts listed in insertion order · case 04
Top-k reports with tied counts are ordered inconsistently between runs with different arrival orders.
Space-Saving heavy hitters: equal counts listed in insertion order · case 05
Top-k reports with tied counts are ordered inconsistently between runs with different arrival orders.
Reservoir sampling (Algorithm R) with supplied draws: replacement range excludes the current item · case 01
Later items are over-sampled because the acceptance probability is k/i instead of k/(i+1).
Reservoir sampling (Algorithm R) with supplied draws: replacement range excludes the current item · case 02
Later items are over-sampled because the acceptance probability is k/i instead of k/(i+1).
Reservoir sampling (Algorithm R) with supplied draws: replacement range excludes the current item · case 03
Later items are over-sampled because the acceptance probability is k/i instead of k/(i+1).
Reservoir sampling (Algorithm R) with supplied draws: replacement range excludes the current item · case 04
Later items are over-sampled because the acceptance probability is k/i instead of k/(i+1).
Reservoir sampling (Algorithm R) with supplied draws: replacement range excludes the current item · case 05
Later items are over-sampled because the acceptance probability is k/i instead of k/(i+1).
Reservoir sampling (Algorithm R) with supplied draws: draw indexed by stream position · case 01
The sampler skips the first k draws, so a replayed sample does not match the recorded random tape.
Reservoir sampling (Algorithm R) with supplied draws: draw indexed by stream position · case 02
The sampler skips the first k draws, so a replayed sample does not match the recorded random tape.
Reservoir sampling (Algorithm R) with supplied draws: draw indexed by stream position · case 03
The sampler skips the first k draws, so a replayed sample does not match the recorded random tape.
Reservoir sampling (Algorithm R) with supplied draws: draw indexed by stream position · case 04
The sampler skips the first k draws, so a replayed sample does not match the recorded random tape.
Reservoir sampling (Algorithm R) with supplied draws: draw indexed by stream position · case 05
The sampler skips the first k draws, so a replayed sample does not match the recorded random tape.
Reservoir sampling (Algorithm R) with supplied draws: accepted item evicts the oldest slot · case 01
The reservoir behaves like a FIFO window, over-representing recent items.
Reservoir sampling (Algorithm R) with supplied draws: accepted item evicts the oldest slot · case 02
The reservoir behaves like a FIFO window, over-representing recent items.
Reservoir sampling (Algorithm R) with supplied draws: accepted item evicts the oldest slot · case 03
The reservoir behaves like a FIFO window, over-representing recent items.
Reservoir sampling (Algorithm R) with supplied draws: accepted item evicts the oldest slot · case 04
The reservoir behaves like a FIFO window, over-representing recent items.
Reservoir sampling (Algorithm R) with supplied draws: accepted item evicts the oldest slot · case 05
The reservoir behaves like a FIFO window, over-representing recent items.
Reservoir sampling (Algorithm R) with supplied draws: draw tape advanced only on replacement · case 01
Rejected items do not consume randomness, so later decisions reuse the same draw.
Reservoir sampling (Algorithm R) with supplied draws: draw tape advanced only on replacement · case 02
Rejected items do not consume randomness, so later decisions reuse the same draw.
Reservoir sampling (Algorithm R) with supplied draws: draw tape advanced only on replacement · case 03
Rejected items do not consume randomness, so later decisions reuse the same draw.
Reservoir sampling (Algorithm R) with supplied draws: draw tape advanced only on replacement · case 04
Rejected items do not consume randomness, so later decisions reuse the same draw.
Reservoir sampling (Algorithm R) with supplied draws: draw tape advanced only on replacement · case 05
Rejected items do not consume randomness, so later decisions reuse the same draw.
Weighted reservoir sampling by exponential keys: key raises u to the weight · case 01
Heavier items receive smaller keys and are sampled less often than light ones.
Weighted reservoir sampling by exponential keys: key raises u to the weight · case 02
Heavier items receive smaller keys and are sampled less often than light ones.
Weighted reservoir sampling by exponential keys: key raises u to the weight · case 03
Heavier items receive smaller keys and are sampled less often than light ones.
Weighted reservoir sampling by exponential keys: key raises u to the weight · case 04
Heavier items receive smaller keys and are sampled less often than light ones.
Weighted reservoir sampling by exponential keys: key raises u to the weight · case 05
Heavier items receive smaller keys and are sampled less often than light ones.
Weighted reservoir sampling by exponential keys: keeps the smallest keys · case 01
The sample favours the lightest items.
Weighted reservoir sampling by exponential keys: keeps the smallest keys · case 02
The sample favours the lightest items.
Weighted reservoir sampling by exponential keys: keeps the smallest keys · case 03
The sample favours the lightest items.
Weighted reservoir sampling by exponential keys: keeps the smallest keys · case 04
The sample favours the lightest items.
Weighted reservoir sampling by exponential keys: keeps the smallest keys · case 05
The sample favours the lightest items.
Weighted reservoir sampling by exponential keys: negative weights enter the sample · case 01
An item with negative weight gets a key above every legitimate item and is always sampled.
Weighted reservoir sampling by exponential keys: negative weights enter the sample · case 02
An item with negative weight gets a key above every legitimate item and is always sampled.
Weighted reservoir sampling by exponential keys: negative weights enter the sample · case 03
An item with negative weight gets a key above every legitimate item and is always sampled.
Weighted reservoir sampling by exponential keys: negative weights enter the sample · case 04
An item with negative weight gets a key above every legitimate item and is always sampled.
Weighted reservoir sampling by exponential keys: negative weights enter the sample · case 05
An item with negative weight gets a key above every legitimate item and is always sampled.
Weighted reservoir sampling by exponential keys: ties favour the larger id · case 01
Equal keys resolve differently from the contract, making replays disagree.
Weighted reservoir sampling by exponential keys: ties favour the larger id · case 02
Equal keys resolve differently from the contract, making replays disagree.
Weighted reservoir sampling by exponential keys: ties favour the larger id · case 03
Equal keys resolve differently from the contract, making replays disagree.
Weighted reservoir sampling by exponential keys: ties favour the larger id · case 04
Equal keys resolve differently from the contract, making replays disagree.
Weighted reservoir sampling by exponential keys: ties favour the larger id · case 05
Equal keys resolve differently from the contract, making replays disagree.
MinHash Jaccard estimation: hashes the minimum element · case 01
Every signature component comes from the same element, so the estimate is 0 or 1.
MinHash Jaccard estimation: hashes the minimum element · case 02
Every signature component comes from the same element, so the estimate is 0 or 1.
MinHash Jaccard estimation: hashes the minimum element · case 03
Every signature component comes from the same element, so the estimate is 0 or 1.
MinHash Jaccard estimation: hashes the minimum element · case 04
Every signature component comes from the same element, so the estimate is 0 or 1.
MinHash Jaccard estimation: hashes the minimum element · case 05
Every signature component comes from the same element, so the estimate is 0 or 1.
MinHash Jaccard estimation: two empty sets report zero similarity · case 01
Empty documents are reported as maximally dissimilar instead of undefined.
MinHash Jaccard estimation: two empty sets report zero similarity · case 02
Empty documents are reported as maximally dissimilar instead of undefined.
MinHash Jaccard estimation: two empty sets report zero similarity · case 03
Empty documents are reported as maximally dissimilar instead of undefined.
MinHash Jaccard estimation: two empty sets report zero similarity · case 04
Empty documents are reported as maximally dissimilar instead of undefined.
MinHash Jaccard estimation: two empty sets report zero similarity · case 05
Empty documents are reported as maximally dissimilar instead of undefined.
MinHash Jaccard estimation: agreement normalised like a set Jaccard · case 01
Similarity estimates are systematically deflated for partially overlapping sets.
MinHash Jaccard estimation: agreement normalised like a set Jaccard · case 02
Similarity estimates are systematically deflated for partially overlapping sets.
MinHash Jaccard estimation: agreement normalised like a set Jaccard · case 03
Similarity estimates are systematically deflated for partially overlapping sets.
MinHash Jaccard estimation: agreement normalised like a set Jaccard · case 04
Similarity estimates are systematically deflated for partially overlapping sets.
MinHash Jaccard estimation: agreement normalised like a set Jaccard · case 05
Similarity estimates are systematically deflated for partially overlapping sets.
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 ↗