FA-98266 / Autoregressive target shifting / Open access

The end marker receives a spurious outgoing prediction · case 01

The end marker receives a spurious outgoing prediction.

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

ROOT CAUSE

The pair builder adds an artificial EOS-to-EOS transition after the valid terminal target.

THE FAILURE

The pair builder adds an artificial EOS-to-EOS transition after the valid terminal target.

Unsuccessful approach: Filtering EOS-valued inputs also drops legitimate content when a token value equals the configured EOS sentinel.

Case contract

Return one pair per observed token, with EOS as the final target and never as an input.

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, eos):
    s=tokens+[eos]
    return [[s[i],s[i+1]] for i in range(len(s)-1)] + ([[eos,eos]] if tokens else [])
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('no eos input',solve([1,2],9),[[1,2],[2,9]])
check('single token',solve([N],8),[[N,8]])
check('empty',solve([],8),[])
check('eos repeated',solve([3,3],3),[[3,3],[3,3]])
check('several',solve([4,5,6],0),[[4,5],[5,6],[6,0]])
check('eos first token',solve([0,1],0),[[0,1],[1,0]])
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
no eos input	`[[1, 2], [2, 9], [9, 9]]`	`[[1, 2], [2, 9]]`	Failed
single token	`[[1, 8], [8, 8]]`	`[[1, 8]]`	Failed
empty	`[]`	`[]`	Passed
eos repeated	`[[3, 3], [3, 3], [3, 3]]`	`[[3, 3], [3, 3]]`	Failed
several	`[[4, 5], [5, 6], [6, 0], [0, 0]]`	`[[4, 5], [5, 6], [6, 0]]`	Failed
eos first token	`[[0, 1], [1, 0], [0, 0]]`	`[[0, 1], [1, 0]]`	Failed

SHA-256 / b6688ec079063cdbfbd05febbd349a7c1c39cc109f8994f55e8a37862b479164

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, eos):
    s=tokens+[eos]
    return [[s[i],s[i+1]] for i in range(len(s)-1) if s[i]!=eos]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('no eos input',solve([1,2],9),[[1,2],[2,9]])
check('single token',solve([N],8),[[N,8]])
check('empty',solve([],8),[])
check('eos repeated',solve([3,3],3),[[3,3],[3,3]])
check('several',solve([4,5,6],0),[[4,5],[5,6],[6,0]])
check('eos first token',solve([0,1],0),[[0,1],[1,0]])
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
no eos input	`[[1, 2], [2, 9]]`	`[[1, 2], [2, 9]]`	Passed
single token	`[[1, 8]]`	`[[1, 8]]`	Passed
empty	`[]`	`[]`	Passed
eos repeated	`[]`	`[[3, 3], [3, 3]]`	Failed
several	`[[4, 5], [5, 6], [6, 0]]`	`[[4, 5], [5, 6], [6, 0]]`	Passed
eos first token	`[[1, 0]]`	`[[0, 1], [1, 0]]`	Failed

SHA-256 / dae5781ed5a60d366a1931b5a2758a4f191b552a3131c3b8b14d71b80628ace1

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 2026-09-30T16:48:11.055048+00:00.

Case digest / 0248bc5a0a21b4d4e669ad589754812b558be619c54f807a469fea41acccf8df