FAILURE MAP

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 ↗
100840Executable case variants
20168Distinct failure mechanisms
302520Executed implementations
20168Open-access cases

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

Python · Standard library
REFERENCEFAILURE MECHANISMDOMAINACCESS
FA-98201

Causal position does not reset at segment boundaries · case 01

Causal position does not reset at segment boundaries.

Attention mask construction● Open access↗
FA-98202

Causal position does not reset at segment boundaries · case 02

Causal position does not reset at segment boundaries.

Attention mask construction◈ Members↗
FA-98203

Causal position does not reset at segment boundaries · case 03

Causal position does not reset at segment boundaries.

Attention mask construction◈ Members↗
FA-98204

Causal position does not reset at segment boundaries · case 04

Causal position does not reset at segment boundaries.

Attention mask construction◈ Members↗
FA-98205

Causal position does not reset at segment boundaries · case 05

Causal position does not reset at segment boundaries.

Attention mask construction◈ Members↗
FA-98206

Block-causal mask lets keys cross supplied block boundaries · case 01

Block-causal mask lets keys cross supplied block boundaries.

Attention mask construction● Open access↗
FA-98207

Block-causal mask lets keys cross supplied block boundaries · case 02

Block-causal mask lets keys cross supplied block boundaries.

Attention mask construction◈ Members↗
FA-98208

Block-causal mask lets keys cross supplied block boundaries · case 03

Block-causal mask lets keys cross supplied block boundaries.

Attention mask construction◈ Members↗
FA-98209

Block-causal mask lets keys cross supplied block boundaries · case 04

Block-causal mask lets keys cross supplied block boundaries.

Attention mask construction◈ Members↗
FA-98210

Block-causal mask lets keys cross supplied block boundaries · case 05

Block-causal mask lets keys cross supplied block boundaries.

Attention mask construction◈ Members↗
FA-98211

Sliding window stops after the initial query rows · case 01

Sliding window stops after the initial query rows.

Attention mask construction● Open access↗
FA-98212

Sliding window stops after the initial query rows · case 02

Sliding window stops after the initial query rows.

Attention mask construction◈ Members↗
FA-98213

Sliding window stops after the initial query rows · case 03

Sliding window stops after the initial query rows.

Attention mask construction◈ Members↗
FA-98214

Sliding window stops after the initial query rows · case 04

Sliding window stops after the initial query rows.

Attention mask construction◈ Members↗
FA-98215

Sliding window stops after the initial query rows · case 05

Sliding window stops after the initial query rows.

Attention mask construction◈ Members↗
FA-98216

Bidirectional local attention omits future neighbors · case 01

Bidirectional local attention omits future neighbors.

Attention mask construction● Open access↗
FA-98217

Bidirectional local attention omits future neighbors · case 02

Bidirectional local attention omits future neighbors.

Attention mask construction◈ Members↗
FA-98218

Bidirectional local attention omits future neighbors · case 03

Bidirectional local attention omits future neighbors.

Attention mask construction◈ Members↗
FA-98219

Bidirectional local attention omits future neighbors · case 04

Bidirectional local attention omits future neighbors.

Attention mask construction◈ Members↗
FA-98220

Bidirectional local attention omits future neighbors · case 05

Bidirectional local attention omits future neighbors.

Attention mask construction◈ Members↗
FA-98221

Causal stripe mask includes keys from every residue · case 01

Causal stripe mask includes keys from every residue.

Attention mask construction● Open access↗
FA-98222

Causal stripe mask includes keys from every residue · case 02

Causal stripe mask includes keys from every residue.

Attention mask construction◈ Members↗
FA-98223

Causal stripe mask includes keys from every residue · case 03

Causal stripe mask includes keys from every residue.

Attention mask construction◈ Members↗
FA-98224

Causal stripe mask includes keys from every residue · case 04

Causal stripe mask includes keys from every residue.

Attention mask construction◈ Members↗
FA-98225

Causal stripe mask includes keys from every residue · case 05

Causal stripe mask includes keys from every residue.

Attention mask construction◈ Members↗
FA-98226

Global keys remain hidden from earlier queries · case 01

Global keys remain hidden from earlier queries.

Attention mask construction● Open access↗
FA-98227

Global keys remain hidden from earlier queries · case 02

Global keys remain hidden from earlier queries.

Attention mask construction◈ Members↗
FA-98228

Global keys remain hidden from earlier queries · case 03

Global keys remain hidden from earlier queries.

Attention mask construction◈ Members↗
FA-98229

Global keys remain hidden from earlier queries · case 04

Global keys remain hidden from earlier queries.

Attention mask construction◈ Members↗
FA-98230

Global keys remain hidden from earlier queries · case 05

Global keys remain hidden from earlier queries.

Attention mask construction◈ Members↗
FA-98231

Block sparse mask attends to non-neighbor blocks · case 01

Block sparse mask attends to non-neighbor blocks.

Attention mask construction● Open access↗
FA-98232

Block sparse mask attends to non-neighbor blocks · case 02

Block sparse mask attends to non-neighbor blocks.

Attention mask construction◈ Members↗
FA-98233

Block sparse mask attends to non-neighbor blocks · case 03

Block sparse mask attends to non-neighbor blocks.

Attention mask construction◈ Members↗
FA-98234

Block sparse mask attends to non-neighbor blocks · case 04

Block sparse mask attends to non-neighbor blocks.

Attention mask construction◈ Members↗
FA-98235

Block sparse mask attends to non-neighbor blocks · case 05

Block sparse mask attends to non-neighbor blocks.

Attention mask construction◈ Members↗
FA-98236

Cross-attention mask omits the source interval limit · case 01

Cross-attention mask omits the source interval limit.

Attention mask construction● Open access↗
FA-98237

Cross-attention mask omits the source interval limit · case 02

Cross-attention mask omits the source interval limit.

Attention mask construction◈ Members↗
FA-98238

Cross-attention mask omits the source interval limit · case 03

Cross-attention mask omits the source interval limit.

Attention mask construction◈ Members↗
FA-98239

Cross-attention mask omits the source interval limit · case 04

Cross-attention mask omits the source interval limit.

Attention mask construction◈ Members↗
FA-98240

Cross-attention mask omits the source interval limit · case 05

Cross-attention mask omits the source interval limit.

Attention mask construction◈ Members↗
FA-98241

The final input token has no next-token target · case 01

The final input token has no next-token target.

Autoregressive target shifting● Open access↗
FA-98242

The final input token has no next-token target · case 02

The final input token has no next-token target.

Autoregressive target shifting◈ Members↗
FA-98243

The final input token has no next-token target · case 03

The final input token has no next-token target.

Autoregressive target shifting◈ Members↗
FA-98244

The final input token has no next-token target · case 04

The final input token has no next-token target.

Autoregressive target shifting◈ Members↗
FA-98245

The final input token has no next-token target · case 05

The final input token has no next-token target.

Autoregressive target shifting◈ Members↗
FA-98246

Truncation removes a retained target at the cutoff · case 01

Truncation removes a retained target at the cutoff.

Autoregressive target shifting● Open access↗
FA-98247

Truncation removes a retained target at the cutoff · case 02

Truncation removes a retained target at the cutoff.

Autoregressive target shifting◈ Members↗
FA-98248

Truncation removes a retained target at the cutoff · case 03

Truncation removes a retained target at the cutoff.

Autoregressive target shifting◈ Members↗
FA-98249

Truncation removes a retained target at the cutoff · case 04

Truncation removes a retained target at the cutoff.

Autoregressive target shifting◈ Members↗
FA-98250

Truncation removes a retained target at the cutoff · case 05

Truncation removes a retained target at the cutoff.

Autoregressive target shifting◈ Members↗
FA-98251

Start marker context is copied as its own target · case 01

Start marker context is copied as its own target.

Autoregressive target shifting● Open access↗
FA-98252

Start marker context is copied as its own target · case 02

Start marker context is copied as its own target.

Autoregressive target shifting◈ Members↗
FA-98253

Start marker context is copied as its own target · case 03

Start marker context is copied as its own target.

Autoregressive target shifting◈ Members↗
FA-98254

Start marker context is copied as its own target · case 04

Start marker context is copied as its own target.

Autoregressive target shifting◈ Members↗
FA-98255

Start marker context is copied as its own target · case 05

Start marker context is copied as its own target.

Autoregressive target shifting◈ Members↗
FA-98256

The first observed token is paired with itself after start insertion · case 01

The first observed token is paired with itself after start insertion.

Autoregressive target shifting● Open access↗
FA-98257

The first observed token is paired with itself after start insertion · case 02

The first observed token is paired with itself after start insertion.

Autoregressive target shifting◈ Members↗
FA-98258

The first observed token is paired with itself after start insertion · case 03

The first observed token is paired with itself after start insertion.

Autoregressive target shifting◈ Members↗
FA-98259

The first observed token is paired with itself after start insertion · case 04

The first observed token is paired with itself after start insertion.

Autoregressive target shifting◈ Members↗
FA-98260

The first observed token is paired with itself after start insertion · case 05

The first observed token is paired with itself after start insertion.

Autoregressive target shifting◈ Members↗
FA-98261

End marker is omitted from next-token supervision · case 01

End marker is omitted from next-token supervision.

Autoregressive target shifting● Open access↗
FA-98262

End marker is omitted from next-token supervision · case 02

End marker is omitted from next-token supervision.

Autoregressive target shifting◈ Members↗
FA-98263

End marker is omitted from next-token supervision · case 03

End marker is omitted from next-token supervision.

Autoregressive target shifting◈ Members↗
FA-98264

End marker is omitted from next-token supervision · case 04

End marker is omitted from next-token supervision.

Autoregressive target shifting◈ Members↗
FA-98265

End marker is omitted from next-token supervision · case 05

End marker is omitted from next-token supervision.

Autoregressive target shifting◈ Members↗
FA-98266

The end marker receives a spurious outgoing prediction · case 01

The end marker receives a spurious outgoing prediction.

Autoregressive target shifting● Open access↗
FA-98267

The end marker receives a spurious outgoing prediction · case 02

The end marker receives a spurious outgoing prediction.

Autoregressive target shifting◈ Members↗
FA-98268

The end marker receives a spurious outgoing prediction · case 03

The end marker receives a spurious outgoing prediction.

Autoregressive target shifting◈ Members↗
FA-98269

The end marker receives a spurious outgoing prediction · case 04

The end marker receives a spurious outgoing prediction.

Autoregressive target shifting◈ Members↗
FA-98270

The end marker receives a spurious outgoing prediction · case 05

The end marker receives a spurious outgoing prediction.

Autoregressive target shifting◈ Members↗
FA-98271

Ignored target labels mask the following position · case 01

Ignored target labels mask the following position.

Autoregressive target shifting● Open access↗
FA-98272

Ignored target labels mask the following position · case 02

Ignored target labels mask the following position.

Autoregressive target shifting◈ Members↗
FA-98273

Ignored target labels mask the following position · case 03

Ignored target labels mask the following position.

Autoregressive target shifting◈ Members↗
FA-98274

Ignored target labels mask the following position · case 04

Ignored target labels mask the following position.

Autoregressive target shifting◈ Members↗
FA-98275

Ignored target labels mask the following position · case 05

Ignored target labels mask the following position.

Autoregressive target shifting◈ Members↗
FA-98276

The final target mask is dropped with the input slice · case 01

The final target mask is dropped with the input slice.

Autoregressive target shifting● Open access↗
FA-98277

The final target mask is dropped with the input slice · case 02

The final target mask is dropped with the input slice.

Autoregressive target shifting◈ Members↗
FA-98278

The final target mask is dropped with the input slice · case 03

The final target mask is dropped with the input slice.

Autoregressive target shifting◈ Members↗
FA-98279

The final target mask is dropped with the input slice · case 04

The final target mask is dropped with the input slice.

Autoregressive target shifting◈ Members↗
FA-98280

The final target mask is dropped with the input slice · case 05

The final target mask is dropped with the input slice.

Autoregressive target shifting◈ Members↗
FA-98281

A context window predicts each token from itself · case 01

A context window predicts each token from itself.

Autoregressive target shifting● Open access↗
FA-98282

A context window predicts each token from itself · case 02

A context window predicts each token from itself.

Autoregressive target shifting◈ Members↗
FA-98283

A context window predicts each token from itself · case 03

A context window predicts each token from itself.

Autoregressive target shifting◈ Members↗
FA-98284

A context window predicts each token from itself · case 04

A context window predicts each token from itself.

Autoregressive target shifting◈ Members↗
FA-98285

A context window predicts each token from itself · case 05

A context window predicts each token from itself.

Autoregressive target shifting◈ Members↗
FA-98286

The final in-window context token is omitted · case 01

The final in-window context token is omitted.

Autoregressive target shifting● Open access↗
FA-98287

The final in-window context token is omitted · case 02

The final in-window context token is omitted.

Autoregressive target shifting◈ Members↗
FA-98288

The final in-window context token is omitted · case 03

The final in-window context token is omitted.

Autoregressive target shifting◈ Members↗
FA-98289

The final in-window context token is omitted · case 04

The final in-window context token is omitted.

Autoregressive target shifting◈ Members↗
FA-98290

The final in-window context token is omitted · case 05

The final in-window context token is omitted.

Autoregressive target shifting◈ Members↗
FA-98291

A supplied target prefix is aligned one token late · case 01

A supplied target prefix is aligned one token late.

Autoregressive target shifting● Open access↗
FA-98292

A supplied target prefix is aligned one token late · case 02

A supplied target prefix is aligned one token late.

Autoregressive target shifting◈ Members↗
FA-98293

A supplied target prefix is aligned one token late · case 03

A supplied target prefix is aligned one token late.

Autoregressive target shifting◈ Members↗
FA-98294

A supplied target prefix is aligned one token late · case 04

A supplied target prefix is aligned one token late.

Autoregressive target shifting◈ Members↗
FA-98295

A supplied target prefix is aligned one token late · case 05

A supplied target prefix is aligned one token late.

Autoregressive target shifting◈ Members↗
FA-98296

A longer target prefix repeats the final context in extra pairs · case 01

A longer target prefix repeats the final context in extra pairs.

Autoregressive target shifting● Open access↗
FA-98297

A longer target prefix repeats the final context in extra pairs · case 02

A longer target prefix repeats the final context in extra pairs.

Autoregressive target shifting◈ Members↗
FA-98298

A longer target prefix repeats the final context in extra pairs · case 03

A longer target prefix repeats the final context in extra pairs.

Autoregressive target shifting◈ Members↗
FA-98299

A longer target prefix repeats the final context in extra pairs · case 04

A longer target prefix repeats the final context in extra pairs.

Autoregressive target shifting◈ Members↗
FA-98300

A longer target prefix repeats the final context in extra pairs · case 05

A longer target prefix repeats the final context in extra pairs.

Autoregressive target shifting◈ Members↗

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 ↗