FA-98271 / Autoregressive target shifting / Open access

Ignored target labels mask the following position · case 01

Ignored target labels mask the following position.

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

ROOT CAUSE

The ignore mask is shifted with inputs instead of being derived from target labels.

THE FAILURE

The ignore mask is shifted with inputs instead of being derived from target labels.

Unsuccessful approach: Starting target-label iteration at index one omits the first shifted target position.

Case contract

For labels, return one boolean per adjacent target position: true iff that target is not ignore; pair count is len(labels)-1.

Why this case matters

A small deterministic token-stream contract makes next-token alignment errors inspectable.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(labels, ignore):
    return [x!=ignore for x in labels[:-1]]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ignored middle target',solve([1,-1,3,4],-1),[False,True,True])
check('first label ignored',solve([-1,2,3],-1),[True,True])
check('last ignored',solve([1,2,-1],-1),[True,False])
check('singleton',solve([-1],-1),[])
check('empty',solve([],-1),[])
check('all valid',solve([N,N+1,N+2],-1),[True,True])
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
ignored middle target	`[True, False, True]`	`[False, True, True]`	Failed
first label ignored	`[False, True]`	`[True, True]`	Failed
last ignored	`[True, True]`	`[True, False]`	Failed
singleton	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
all valid	`[True, True]`	`[True, True]`	Passed

SHA-256 / 48f60bc293c85c9ef89dd0301b42065b1c2839eda2b6a545e251942d9760be91

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(labels, ignore):
    return [labels[i+1]!=ignore for i in range(1,len(labels)-1)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ignored middle target',solve([1,-1,3,4],-1),[False,True,True])
check('first label ignored',solve([-1,2,3],-1),[True,True])
check('last ignored',solve([1,2,-1],-1),[True,False])
check('singleton',solve([-1],-1),[])
check('empty',solve([],-1),[])
check('all valid',solve([N,N+1,N+2],-1),[True,True])
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
ignored middle target	`[True, True]`	`[False, True, True]`	Failed
first label ignored	`[True]`	`[True, True]`	Failed
last ignored	`[False]`	`[True, False]`	Failed
singleton	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
all valid	`[True]`	`[True, True]`	Failed

SHA-256 / 8dc4d31652ddebd62dc0a76057cc1de940343a75d548e381f4b941605523a219

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

Inputs are toy integer token sequences; this contract says nothing about model quality, batching, sampling, or production training frameworks. 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 2026-09-30T16:48:11.054394+00:00.

Case digest / d59f8fed9e490e8ab6d58af941e5a63a5c463cf2e7462e48123cbf4ff6b84046