FA-98291 / Autoregressive target shifting / Open access

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

A supplied target prefix is aligned one token late.

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

ROOT CAUSE

Each input is paired with the next target slot, skipping the target at the same local position.

THE FAILURE

Each input is paired with the next target slot, skipping the target at the same local position.

Unsuccessful approach: Reusing the previous target index fixes the first pair only by clamping its missing predecessor to zero.

Case contract

Given equal length context and target lists, return zipped pairs in order; excess elements are ignored.

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(context, target):
    return [[context[i],target[i+1]] for i in range(min(len(context),max(0,len(target)-1)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('aligned',solve([1,2,3],[4,5,6]),[[1,4],[2,5],[3,6]])
check('unequal target',solve([1,2,3],[8,9]),[[1,8],[2,9]])
check('empty target',solve([1],[]),[])
check('empty context',solve([],[2]),[])
check('singleton',solve([N],[N+1]),[[N,N+1]])
check('distinct order',solve([7,8],[9,10]),[[7,9],[8,10]])
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
aligned	`[[1, 5], [2, 6]]`	`[[1, 4], [2, 5], [3, 6]]`	Failed
unequal target	`[[1, 9]]`	`[[1, 8], [2, 9]]`	Failed
empty target	`[]`	`[]`	Passed
empty context	`[]`	`[]`	Passed
singleton	`[]`	`[[1, 2]]`	Failed
distinct order	`[[7, 10]]`	`[[7, 9], [8, 10]]`	Failed

SHA-256 / a3285c977b84cc49c170ae619d3b9a76273312dfb7b58ec02e98c94037ad63b9

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(context, target):
    return [[context[i],target[max(0,i-1)]] for i in range(min(len(context),len(target)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('aligned',solve([1,2,3],[4,5,6]),[[1,4],[2,5],[3,6]])
check('unequal target',solve([1,2,3],[8,9]),[[1,8],[2,9]])
check('empty target',solve([1],[]),[])
check('empty context',solve([],[2]),[])
check('singleton',solve([N],[N+1]),[[N,N+1]])
check('distinct order',solve([7,8],[9,10]),[[7,9],[8,10]])
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
aligned	`[[1, 4], [2, 4], [3, 5]]`	`[[1, 4], [2, 5], [3, 6]]`	Failed
unequal target	`[[1, 8], [2, 8]]`	`[[1, 8], [2, 9]]`	Failed
empty target	`[]`	`[]`	Passed
empty context	`[]`	`[]`	Passed
singleton	`[[1, 2]]`	`[[1, 2]]`	Passed
distinct order	`[[7, 9], [8, 9]]`	`[[7, 9], [8, 10]]`	Failed

SHA-256 / 0365ef19ff7a5f06d109a64f097c83eeabdd97e415a996c72de235505ee70455

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 2026-09-30T16:48:11.500144+00:00.

Case digest / 92cc48948a9fda65a9cb3360e6e8219c01ee04297f0e69cef8f7e0e6b9eab98f