FA-98241 / Autoregressive target shifting / Open access

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

The final input token has no next-token target.

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

ROOT CAUSE

The pair builder zips the stream against itself instead of offset views.

THE FAILURE

The pair builder zips the stream against itself instead of offset views.

Unsuccessful approach: Using a two-position offset fixes length but skips the immediate successor.

Case contract

For a token stream, return adjacent [input,target] pairs; streams shorter than two yield an empty list.

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(tokens):
    return [[a,b] for a,b in zip(tokens,tokens)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ordinary',solve([N,N+1,N+2]),[[N,N+1],[N+1,N+2]])
check('two tokens',solve([N,8]),[[N,8]])
check('singleton',solve([N]),[])
check('empty',solve([]),[])
check('repeated',solve([4,4,5]),[[4,4],[4,5]])
check('longer',solve([1,N,3,4]),[[1,N],[N,3],[3,4]])
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
ordinary	`[[1, 1], [2, 2], [3, 3]]`	`[[1, 2], [2, 3]]`	Failed
two tokens	`[[1, 1], [8, 8]]`	`[[1, 8]]`	Failed
singleton	`[[1, 1]]`	`[]`	Failed
empty	`[]`	`[]`	Passed
repeated	`[[4, 4], [4, 4], [5, 5]]`	`[[4, 4], [4, 5]]`	Failed
longer	`[[1, 1], [1, 1], [3, 3], [4, 4]]`	`[[1, 1], [1, 3], [3, 4]]`	Failed

SHA-256 / 8e52c278266118fb14dd8b02a6c4c19ff4d1eb91c2be7d12015357fed13518e9

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens):
    return [[tokens[i],tokens[i+2]] for i in range(max(0,len(tokens)-2))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ordinary',solve([N,N+1,N+2]),[[N,N+1],[N+1,N+2]])
check('two tokens',solve([N,8]),[[N,8]])
check('singleton',solve([N]),[])
check('empty',solve([]),[])
check('repeated',solve([4,4,5]),[[4,4],[4,5]])
check('longer',solve([1,N,3,4]),[[1,N],[N,3],[3,4]])
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
ordinary	`[[1, 3]]`	`[[1, 2], [2, 3]]`	Failed
two tokens	`[]`	`[[1, 8]]`	Failed
singleton	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
repeated	`[[4, 5]]`	`[[4, 4], [4, 5]]`	Failed
longer	`[[1, 3], [1, 4]]`	`[[1, 1], [1, 3], [3, 4]]`	Failed

SHA-256 / f3845e549ceaba498fb8d9ceda73edcef15767d3bd8c334ed99a59416aaf6194

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 2026-09-30T16:48:11.053562+00:00.

Case digest / 847ab108c550671a2a9c0d0efa66f2399c9cf09595822e565ea6c4b0572783c9