FA-98181 / Attention mask construction / Open access

Cross-attention mask swaps source and query dimensions · case 01

Cross-attention mask swaps source and query dimensions.

Verified by executionVariant 1 · 6 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

Rows use the source count and columns use the query count, while source-key padding is ignored.

THE FAILURE

Rows use the source count and columns use the query count, while source-key padding is ignored.

Unsuccessful approach: Broadcasting a query-indexed source validity bit across the row loses the individual source-key validity pattern.

Case contract

Return QxS visibility from target queries to every valid source key; query values do not affect visibility.

Why this case matters

A small explicit visibility matrix makes sequence-attention boundary behavior reviewable.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(source_valid, query_count):
    return [[True for _ in range(query_count)] for _ in source_valid]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
s=[True]*N+[False,True]
check('rectangular source',solve(s,2),[[True]*N+[False,True] for _ in range(2)])
check('no sources',solve([],3),[[],[],[]])
check('no queries',solve([True],0),[])
check('all padded sources',solve([False,False],1),[[False,False]])
check('single valid source',solve([False,True],2),[[False,True],[False,True]])
check('queries independent',solve([True,False],3),[[True,False]]*3)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
rectangular source	`[[True, True], [True, True], [True, True]]`	`[[True, False, True], [True, False, True]]`	Failed
no sources	`[]`	`[[], [], []]`	Failed
no queries	`[[]]`	`[]`	Failed
all padded sources	`[[True], [True]]`	`[[False, False]]`	Failed
single valid source	`[[True, True], [True, True]]`	`[[False, True], [False, True]]`	Failed
queries independent	`[[True, True, True], [True, True, True]]`	`[[True, False], [True, False], [True, False]]`	Failed

SHA-256 / dca0c4ee52f87787b2a2c6a43f8fdceedfc41a8cade0bd7e5e85779fb15ee584

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(source_valid, query_count):
    return [[bool(source_valid[q%len(source_valid)]) for _ in source_valid] for q in range(query_count)] if source_valid else [[] for _ in range(query_count)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
s=[True]*N+[False,True]
check('rectangular source',solve(s,2),[[True]*N+[False,True] for _ in range(2)])
check('no sources',solve([],3),[[],[],[]])
check('no queries',solve([True],0),[])
check('all padded sources',solve([False,False],1),[[False,False]])
check('single valid source',solve([False,True],2),[[False,True],[False,True]])
check('queries independent',solve([True,False],3),[[True,False]]*3)
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
rectangular source	`[[True, True, True], [False, False, False]]`	`[[True, False, True], [True, False, True]]`	Failed
no sources	`[[], [], []]`	`[[], [], []]`	Passed
no queries	`[]`	`[]`	Passed
all padded sources	`[[False, False]]`	`[[False, False]]`	Passed
single valid source	`[[False, False], [True, True]]`	`[[False, True], [False, True]]`	Failed
queries independent	`[[True, True], [False, False], [True, True]]`	`[[True, False], [True, False], [True, False]]`	Failed

SHA-256 / 8e792556b0623826d1adfbba12f220b670f092c6a7ed69f19e2fe6ae5eb2b2a7

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 6 recorded checks per implementation. The open-access tier publishes the failure and the unsuccessful fix; the repaired source that passes every check, and the observations that prove it, are available to members.

Every case sharing this mechanism uses the same contract and the same repair, so this one record is held back for all of them.

Member access is invitation-based. Sign in with your invited account to inspect the repair.

Verification & scope

These are toy Boolean matrix contracts; they do not define a production model architecture or tensor library behavior. This reproducer isolates one failure mechanism. Results cover the supplied fixtures. Variants within a family share a test contract and should remain grouped when constructing evaluation splits. Related mechanisms with a shared evaluation_group must also remain together; these controlled models are not independent production incidents.

Observations recorded using Python 3.12.14 at 2026-09-30T16:48:10.193045+00:00.

Case digest / 8a51eb82147a1d4a0b0b3b0062c36e1c3c9bb77aaf72225cc82e8962e822817a