FA-98246 / Autoregressive target shifting / Open access

Truncation removes a retained target at the cutoff · case 01

Truncation removes a retained target at the cutoff.

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

ROOT CAUSE

The stream is clipped to the input limit before a successor is attached.

THE FAILURE

The stream is clipped to the input limit before a successor is attached.

Unsuccessful approach: A slice-based repair treats a negative cap as a Python suffix limit and emits pairs instead of honoring the nonpositive cap.

Case contract

Given tokens and max_inputs k, return k adjacent input/next-token pairs where available; k<=0 returns empty.

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, max_inputs):
    x=tokens[:max_inputs]
    return [[x[i],x[i+1]] for i in range(len(x)-1)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('retained boundary target',solve([1,2,3,4],2),[[1,2],[2,3]])
check('full stream',solve([N,N+1,N+2],9),[[N,N+1],[N+1,N+2]])
check('zero inputs',solve([1,2],0),[])
check('negative inputs',solve([1,2],-1),[])
check('empty stream',solve([],2),[])
check('one retained input',solve([7,8],1),[[7,8]])
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
retained boundary target	`[[1, 2]]`	`[[1, 2], [2, 3]]`	Failed
full stream	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
zero inputs	`[]`	`[]`	Passed
negative inputs	`[]`	`[]`	Passed
empty stream	`[]`	`[]`	Passed
one retained input	`[]`	`[[7, 8]]`	Failed

SHA-256 / 213682be73efe68ab19e6d8d83c0e0abbc9809aef2b4c3752c884f79724da4ac

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, max_inputs):
    x=tokens[:max_inputs]
    return [[x[i],x[i+1]] for i in range(len(x)-1)] + ([[x[-1],tokens[max_inputs]] ] if x and len(tokens)>max_inputs else [])
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('retained boundary target',solve([1,2,3,4],2),[[1,2],[2,3]])
check('full stream',solve([N,N+1,N+2],9),[[N,N+1],[N+1,N+2]])
check('zero inputs',solve([1,2],0),[])
check('negative inputs',solve([1,2],-1),[])
check('empty stream',solve([],2),[])
check('one retained input',solve([7,8],1),[[7,8]])
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
retained boundary target	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
full stream	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
zero inputs	`[]`	`[]`	Passed
negative inputs	`[[1, 2]]`	`[]`	Failed
empty stream	`[]`	`[]`	Passed
one retained input	`[[7, 8]]`	`[[7, 8]]`	Passed

SHA-256 / bf6f643a3aa747c8243779b43e67d23a508ee4222f55df53884ce91e7eccdf76

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 2026-09-30T16:48:11.054415+00:00.

Case digest / 757189f1b45d88d3a842a6b331317c9442c43ed7bf8cba8b151f782452183b08