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
Coefficient quantizer truncates toward zero · case 01
Negative coefficients in truncate mode quantize one LSB too high.
Coefficient quantizer truncates toward zero · case 02
Negative coefficients in truncate mode quantize one LSB too high.
Coefficient quantizer truncates toward zero · case 03
Negative coefficients in truncate mode quantize one LSB too high.
Coefficient quantizer truncates toward zero · case 04
Negative coefficients in truncate mode quantize one LSB too high.
Coefficient quantizer truncates toward zero · case 05
Negative coefficients in truncate mode quantize one LSB too high.
Coefficient quantizer allows the positive code 2^(B-1) · case 01
A coefficient of 1.0 in Q7 yields code 128, which does not fit in eight bits.
Coefficient quantizer allows the positive code 2^(B-1) · case 02
A coefficient of 1.0 in Q7 yields code 128, which does not fit in eight bits.
Coefficient quantizer allows the positive code 2^(B-1) · case 03
A coefficient of 1.0 in Q7 yields code 128, which does not fit in eight bits.
Coefficient quantizer allows the positive code 2^(B-1) · case 04
A coefficient of 1.0 in Q7 yields code 128, which does not fit in eight bits.
Coefficient quantizer allows the positive code 2^(B-1) · case 05
A coefficient of 1.0 in Q7 yields code 128, which does not fit in eight bits.
Coefficient quantizer counts codes that land exactly on a rail · case 01
A coefficient that quantizes exactly to the most negative code is reported as saturated.
Coefficient quantizer counts codes that land exactly on a rail · case 02
A coefficient that quantizes exactly to the most negative code is reported as saturated.
Coefficient quantizer counts codes that land exactly on a rail · case 03
A coefficient that quantizes exactly to the most negative code is reported as saturated.
Coefficient quantizer counts codes that land exactly on a rail · case 04
A coefficient that quantizes exactly to the most negative code is reported as saturated.
Coefficient quantizer counts codes that land exactly on a rail · case 05
A coefficient that quantizes exactly to the most negative code is reported as saturated.
Coefficient quantizer measures error before clipping · case 01
max_error understates the real error for saturated coefficients.
Coefficient quantizer measures error before clipping · case 02
max_error understates the real error for saturated coefficients.
Coefficient quantizer measures error before clipping · case 03
max_error understates the real error for saturated coefficients.
Coefficient quantizer measures error before clipping · case 04
max_error understates the real error for saturated coefficients.
Coefficient quantizer measures error before clipping · case 05
max_error understates the real error for saturated coefficients.
SOS cascade sections share one aliased state list · case 01
With two or more sections, each section reads the delay values written by the previous one.
SOS cascade sections share one aliased state list · case 02
With two or more sections, each section reads the delay values written by the previous one.
SOS cascade sections share one aliased state list · case 03
With two or more sections, each section reads the delay values written by the previous one.
SOS cascade sections share one aliased state list · case 04
With two or more sections, each section reads the delay values written by the previous one.
SOS cascade sections share one aliased state list · case 05
With two or more sections, each section reads the delay values written by the previous one.
SOS cascade applies the gain after the last section · case 01
With saturating stages the output differs from gain-at-input: loud inputs are no longer clipped in the first stage.
SOS cascade applies the gain after the last section · case 02
With saturating stages the output differs from gain-at-input: loud inputs are no longer clipped in the first stage.
SOS cascade applies the gain after the last section · case 03
With saturating stages the output differs from gain-at-input: loud inputs are no longer clipped in the first stage.
SOS cascade applies the gain after the last section · case 04
With saturating stages the output differs from gain-at-input: loud inputs are no longer clipped in the first stage.
SOS cascade applies the gain after the last section · case 05
With saturating stages the output differs from gain-at-input: loud inputs are no longer clipped in the first stage.
SOS cascade clamps only the final output · case 01
Intermediate stages exceed the limit, so later stages see values the hardware could never produce.
SOS cascade clamps only the final output · case 02
Intermediate stages exceed the limit, so later stages see values the hardware could never produce.
SOS cascade clamps only the final output · case 03
Intermediate stages exceed the limit, so later stages see values the hardware could never produce.
SOS cascade clamps only the final output · case 04
Intermediate stages exceed the limit, so later stages see values the hardware could never produce.
SOS cascade clamps only the final output · case 05
Intermediate stages exceed the limit, so later stages see values the hardware could never produce.
SOS cascade stores the unclamped output in the feedback state · case 01
After a clip the recursive part continues from a value beyond the limit.
SOS cascade stores the unclamped output in the feedback state · case 02
After a clip the recursive part continues from a value beyond the limit.
SOS cascade stores the unclamped output in the feedback state · case 03
After a clip the recursive part continues from a value beyond the limit.
SOS cascade stores the unclamped output in the feedback state · case 04
After a clip the recursive part continues from a value beyond the limit.
SOS cascade stores the unclamped output in the feedback state · case 05
After a clip the recursive part continues from a value beyond the limit.
Linear-phase check skips the centre tap for antisymmetry · case 01
[1, 2, -1] is classified as type 3 although a nonzero centre breaks antisymmetry.
Linear-phase check skips the centre tap for antisymmetry · case 02
[1, 2, -1] is classified as type 3 although a nonzero centre breaks antisymmetry.
Linear-phase check skips the centre tap for antisymmetry · case 03
[1, 2, -1] is classified as type 3 although a nonzero centre breaks antisymmetry.
Linear-phase check skips the centre tap for antisymmetry · case 04
[1, 2, -1] is classified as type 3 although a nonzero centre breaks antisymmetry.
Linear-phase check skips the centre tap for antisymmetry · case 05
[1, 2, -1] is classified as type 3 although a nonzero centre breaks antisymmetry.
Linear-phase delay ignores leading zero taps · case 01
[0, 0, 1, 2, 1] reports a delay of 1 sample instead of 3.
Linear-phase delay ignores leading zero taps · case 02
[0, 0, 1, 2, 1] reports a delay of 1 sample instead of 3.
Linear-phase delay ignores leading zero taps · case 03
[0, 0, 1, 2, 1] reports a delay of 1 sample instead of 3.
Linear-phase delay ignores leading zero taps · case 04
[0, 0, 1, 2, 1] reports a delay of 1 sample instead of 3.
Linear-phase delay ignores leading zero taps · case 05
[0, 0, 1, 2, 1] reports a delay of 1 sample instead of 3.
Linear-phase classifier swaps types 1 and 2 · case 01
An odd-length symmetric filter is labelled type 2.
Linear-phase classifier swaps types 1 and 2 · case 02
An odd-length symmetric filter is labelled type 2.
Linear-phase classifier swaps types 1 and 2 · case 03
An odd-length symmetric filter is labelled type 2.
Linear-phase classifier swaps types 1 and 2 · case 04
An odd-length symmetric filter is labelled type 2.
Linear-phase classifier swaps types 1 and 2 · case 05
An odd-length symmetric filter is labelled type 2.
Linear-phase classifier does not strip leading zeros · case 01
[0, 3, 3, 0, 0] is reported as not linear phase.
Linear-phase classifier does not strip leading zeros · case 02
[0, 3, 3, 0, 0] is reported as not linear phase.
Linear-phase classifier does not strip leading zeros · case 03
[0, 3, 3, 0, 0] is reported as not linear phase.
Linear-phase classifier does not strip leading zeros · case 04
[0, 3, 3, 0, 0] is reported as not linear phase.
Linear-phase classifier does not strip leading zeros · case 05
[0, 3, 3, 0, 0] is reported as not linear phase.
Forward-backward filter pads with an even mirror · case 01
A ramp input gets a kink at the edges and the filtered ends droop.
Forward-backward filter pads with an even mirror · case 02
A ramp input gets a kink at the edges and the filtered ends droop.
Forward-backward filter pads with an even mirror · case 03
A ramp input gets a kink at the edges and the filtered ends droop.
Forward-backward filter pads with an even mirror · case 04
A ramp input gets a kink at the edges and the filtered ends droop.
Forward-backward filter pads with an even mirror · case 05
A ramp input gets a kink at the edges and the filtered ends droop.
Forward-backward filter runs the second pass forward · case 01
The result has twice the group delay instead of zero phase.
Forward-backward filter runs the second pass forward · case 02
The result has twice the group delay instead of zero phase.
Forward-backward filter runs the second pass forward · case 03
The result has twice the group delay instead of zero phase.
Forward-backward filter runs the second pass forward · case 04
The result has twice the group delay instead of zero phase.
Forward-backward filter runs the second pass forward · case 05
The result has twice the group delay instead of zero phase.
Forward-backward filter trims the wrong end of the padding · case 01
The output starts with padded samples and is shifted by pad samples.
Forward-backward filter trims the wrong end of the padding · case 02
The output starts with padded samples and is shifted by pad samples.
Forward-backward filter trims the wrong end of the padding · case 03
The output starts with padded samples and is shifted by pad samples.
Forward-backward filter trims the wrong end of the padding · case 04
The output starts with padded samples and is shifted by pad samples.
Forward-backward filter trims the wrong end of the padding · case 05
The output starts with padded samples and is shifted by pad samples.
Forward-backward padding reflects the edge sample itself · case 01
The padded extension repeats the edge value, creating a flat spot that biases the ends.
Forward-backward padding reflects the edge sample itself · case 02
The padded extension repeats the edge value, creating a flat spot that biases the ends.
Forward-backward padding reflects the edge sample itself · case 03
The padded extension repeats the edge value, creating a flat spot that biases the ends.
Forward-backward padding reflects the edge sample itself · case 04
The padded extension repeats the edge value, creating a flat spot that biases the ends.
Forward-backward padding reflects the edge sample itself · case 05
The padded extension repeats the edge value, creating a flat spot that biases the ends.
Resampling plan swaps the interpolation and decimation factors · case 01
48 kHz to 44.1 kHz is planned as up 160, down 147 instead of up 147, down 160.
Resampling plan swaps the interpolation and decimation factors · case 02
48 kHz to 44.1 kHz is planned as up 160, down 147 instead of up 147, down 160.
Resampling plan swaps the interpolation and decimation factors · case 03
48 kHz to 44.1 kHz is planned as up 160, down 147 instead of up 147, down 160.
Resampling plan swaps the interpolation and decimation factors · case 04
48 kHz to 44.1 kHz is planned as up 160, down 147 instead of up 147, down 160.
Resampling plan swaps the interpolation and decimation factors · case 05
48 kHz to 44.1 kHz is planned as up 160, down 147 instead of up 147, down 160.
Resampling plan sets the cutoff from the higher rate · case 01
Downsampling 48 kHz to 16 kHz keeps content up to 24 kHz, which aliases.
Resampling plan sets the cutoff from the higher rate · case 02
Downsampling 48 kHz to 16 kHz keeps content up to 24 kHz, which aliases.
Resampling plan sets the cutoff from the higher rate · case 03
Downsampling 48 kHz to 16 kHz keeps content up to 24 kHz, which aliases.
Resampling plan sets the cutoff from the higher rate · case 04
Downsampling 48 kHz to 16 kHz keeps content up to 24 kHz, which aliases.
Resampling plan sets the cutoff from the higher rate · case 05
Downsampling 48 kHz to 16 kHz keeps content up to 24 kHz, which aliases.
Resampling plan normalizes the cutoff to the input Nyquist · case 01
The normalized cutoff handed to the filter designer is L times too large.
Resampling plan normalizes the cutoff to the input Nyquist · case 02
The normalized cutoff handed to the filter designer is L times too large.
Resampling plan normalizes the cutoff to the input Nyquist · case 03
The normalized cutoff handed to the filter designer is L times too large.
Resampling plan normalizes the cutoff to the input Nyquist · case 04
The normalized cutoff handed to the filter designer is L times too large.
Resampling plan normalizes the cutoff to the input Nyquist · case 05
The normalized cutoff handed to the filter designer is L times too large.
Resampling plan multiplies the input rate by the decimation factor · case 01
48 kHz to 44.1 kHz reports an intermediate rate of 7.68 MHz instead of 7.056 MHz.
Resampling plan multiplies the input rate by the decimation factor · case 02
48 kHz to 44.1 kHz reports an intermediate rate of 7.68 MHz instead of 7.056 MHz.
Resampling plan multiplies the input rate by the decimation factor · case 03
48 kHz to 44.1 kHz reports an intermediate rate of 7.68 MHz instead of 7.056 MHz.
Resampling plan multiplies the input rate by the decimation factor · case 04
48 kHz to 44.1 kHz reports an intermediate rate of 7.68 MHz instead of 7.056 MHz.
Resampling plan multiplies the input rate by the decimation factor · case 05
48 kHz to 44.1 kHz reports an intermediate rate of 7.68 MHz instead of 7.056 MHz.
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 ↗