FA-98231 / Attention mask construction / Open access

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

Block sparse mask attends to non-neighbor blocks.

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

ROOT CAUSE

The predicate orders block IDs but never bounds their distance or checks causal token order.

THE FAILURE

The predicate orders block IDs but never bounds their distance or checks causal token order.

Unsuccessful approach: Checking only signed block difference lets a lower-numbered query block accept an earlier higher-numbered block.

Case contract

Positions have block IDs; causal pair visible iff absolute block ID difference<=block_radius.

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(blocks, block_radius):
    return [[blocks[j]<=blocks[i] for j in range(len(blocks))] for i in range(len(blocks))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
b=[i//2 for i in range(N+4)]
check('block radius',solve(b,1),[[j<=i and abs(b[i]-b[j])<=1 for j in range(len(b))] for i in range(len(b))])
check('same block',solve([0,0],0),[[True,False],[True,True]])
check('adjacent blocks',solve([0,1],1),[[True,False],[True,True]])
check('excluded block',solve([1,0,2],0),[[True,False,False],[False,True,False],[False,False,True]])
check('empty',solve([],2),[])
check('negative radius',solve([0,0],-1),[[False,False],[False,False]])
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
block radius	`[[True, True, False, False, False], [True, True, False, False, False], [True, True, True, True, False], [True, True, True, True, False], [True, True, True, True, True]]`	`[[True, False, False, False, False], [True, True, False, False, False], [True, True, True, False, False], [True, True, True, True, False], [False, False, True, True, True]]`	Failed
same block	`[[True, True], [True, True]]`	`[[True, False], [True, True]]`	Failed
adjacent blocks	`[[True, False], [True, True]]`	`[[True, False], [True, True]]`	Passed
excluded block	`[[True, True, False], [False, True, False], [True, True, True]]`	`[[True, False, False], [False, True, False], [False, False, True]]`	Failed
empty	`[]`	`[]`	Passed
negative radius	`[[True, True], [True, True]]`	`[[False, False], [False, False]]`	Failed

SHA-256 / 5b6b6556fdc4e5dad68b6751c6c17ad525208a0849016e4c1461378762eb4644

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(blocks, block_radius):
    return [[j<=i and blocks[i]-blocks[j]<=block_radius for j in range(len(blocks))] for i in range(len(blocks))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
b=[i//2 for i in range(N+4)]
check('block radius',solve(b,1),[[j<=i and abs(b[i]-b[j])<=1 for j in range(len(b))] for i in range(len(b))])
check('same block',solve([0,0],0),[[True,False],[True,True]])
check('adjacent blocks',solve([0,1],1),[[True,False],[True,True]])
check('excluded block',solve([1,0,2],0),[[True,False,False],[False,True,False],[False,False,True]])
check('empty',solve([],2),[])
check('negative radius',solve([0,0],-1),[[False,False],[False,False]])
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
block radius	`[[True, False, False, False, False], [True, True, False, False, False], [True, True, True, False, False], [True, True, True, True, False], [False, False, True, True, True]]`	`[[True, False, False, False, False], [True, True, False, False, False], [True, True, True, False, False], [True, True, True, True, False], [False, False, True, True, True]]`	Passed
same block	`[[True, False], [True, True]]`	`[[True, False], [True, True]]`	Passed
adjacent blocks	`[[True, False], [True, True]]`	`[[True, False], [True, True]]`	Passed
excluded block	`[[True, False, False], [True, True, False], [False, False, True]]`	`[[True, False, False], [False, True, False], [False, False, True]]`	Failed
empty	`[]`	`[]`	Passed
negative radius	`[[False, False], [False, False]]`	`[[False, False], [False, False]]`	Passed

SHA-256 / f982a77641fa3cca5d84ce62bfc5dc781e8b0fc2ecec01812f4787476925f27f

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 2026-09-30T16:48:10.630175+00:00.

Case digest / 797b1fdd0d74e83e089c8c9924ffc4b3416ab6c587aa9c4a6cc973fe9af46560