FA-98286 / Autoregressive target shifting / Open access

The final in-window context token is omitted · case 01

The final in-window context token is omitted.

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

ROOT CAUSE

The upper bound is derived from the truncated input window rather than its target lookahead.

THE FAILURE

The upper bound is derived from the truncated input window rather than its target lookahead.

Unsuccessful approach: Making the upper bound inclusive also introduces a pair beyond the sequence end.

Case contract

For tokens and stop, pair indexes 0 through min(stop,len(tokens)-1)-1 with their successors.

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, stop):
    return [[tokens[i],tokens[i+1]] for i in range(max(0,min(stop-1,len(tokens)-1)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('last allowed context',solve([0,1,2,3],3),[[0,1],[1,2],[2,3]])
check('one context',solve([4,5],1),[[4,5]])
check('zero context',solve([4,5],0),[])
check('oversized stop',solve([1,2,3],20),[[1,2],[2,3]])
check('empty',solve([],3),[])
check('singleton',solve([1],1),[])
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
last allowed context	`[[0, 1], [1, 2]]`	`[[0, 1], [1, 2], [2, 3]]`	Failed
one context	`[]`	`[[4, 5]]`	Failed
zero context	`[]`	`[]`	Passed
oversized stop	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
empty	`[]`	`[]`	Passed
singleton	`[]`	`[]`	Passed

SHA-256 / f72ad2fe12c4dfde9d5f81276c8f9f1258b9fc1cbcef53a1956d5ffb7243c2c1

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, stop):
    return [[tokens[i],tokens[i+1]] for i in range(max(0,min(stop+1,len(tokens)-1)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('last allowed context',solve([0,1,2,3],3),[[0,1],[1,2],[2,3]])
check('one context',solve([4,5],1),[[4,5]])
check('zero context',solve([4,5],0),[])
check('oversized stop',solve([1,2,3],20),[[1,2],[2,3]])
check('empty',solve([],3),[])
check('singleton',solve([1],1),[])
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
last allowed context	`[[0, 1], [1, 2], [2, 3]]`	`[[0, 1], [1, 2], [2, 3]]`	Passed
one context	`[[4, 5]]`	`[[4, 5]]`	Passed
zero context	`[[4, 5]]`	`[]`	Failed
oversized stop	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
empty	`[]`	`[]`	Passed
singleton	`[]`	`[]`	Passed

SHA-256 / b6096891f2318591fde8c953171e0d3125a1c585d4879a8f5be13cd4cbc9be33

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 2026-09-30T16:48:11.500126+00:00.

Case digest / 7ed3263a4b3e95dbce3f66235427b9925fb8627a01a4b2cacc12b132d669976f