FA-98301 / Autoregressive target shifting / Open access

Shifted pairs carry an off-by-one context position · case 01

Shifted pairs carry an off-by-one context position.

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

ROOT CAUSE

The pair emitter labels the first input with origin plus one instead of the declared origin.

THE FAILURE

The pair emitter labels the first input with origin plus one instead of the declared origin.

Unsuccessful approach: Resetting the position counter to zero repairs only streams whose origin is zero.

Case contract

Given equally aligned input and next-target lists plus an origin, return [context_position,input,target] rows; ignore unmatched suffixes.

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(inputs, targets, origin):
    return [[origin+i+1,inputs[i],targets[i]] for i in range(min(len(inputs),len(targets)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('nonzero origin',solve([5,6],[8,9],N),[[N,5,8],[N+1,6,9]])
check('zero origin',solve([8],[3],0),[[0,8,3]])
check('negative origin',solve([1,2],[7,8],-2),[[-2,1,7],[-1,2,8]])
check('empty',solve([],[],N),[])
check('unmatched target',solve([4],[9,10],7),[[7,4,9]])
check('unmatched input',solve([4,5],[9],20),[[20,4,9]])
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
nonzero origin	`[[2, 5, 8], [3, 6, 9]]`	`[[1, 5, 8], [2, 6, 9]]`	Failed
zero origin	`[[1, 8, 3]]`	`[[0, 8, 3]]`	Failed
negative origin	`[[-1, 1, 7], [0, 2, 8]]`	`[[-2, 1, 7], [-1, 2, 8]]`	Failed
empty	`[]`	`[]`	Passed
unmatched target	`[[8, 4, 9]]`	`[[7, 4, 9]]`	Failed
unmatched input	`[[21, 4, 9]]`	`[[20, 4, 9]]`	Failed

SHA-256 / 2aef22566ff4529eb9813c763c01182835784670ce08f6ce6d69a86415d1b471

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(inputs, targets, origin):
    return [[i,inputs[i],targets[i]] for i in range(min(len(inputs),len(targets)))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('nonzero origin',solve([5,6],[8,9],N),[[N,5,8],[N+1,6,9]])
check('zero origin',solve([8],[3],0),[[0,8,3]])
check('negative origin',solve([1,2],[7,8],-2),[[-2,1,7],[-1,2,8]])
check('empty',solve([],[],N),[])
check('unmatched target',solve([4],[9,10],7),[[7,4,9]])
check('unmatched input',solve([4,5],[9],20),[[20,4,9]])
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
nonzero origin	`[[0, 5, 8], [1, 6, 9]]`	`[[1, 5, 8], [2, 6, 9]]`	Failed
zero origin	`[[0, 8, 3]]`	`[[0, 8, 3]]`	Passed
negative origin	`[[0, 1, 7], [1, 2, 8]]`	`[[-2, 1, 7], [-1, 2, 8]]`	Failed
empty	`[]`	`[]`	Passed
unmatched target	`[[0, 4, 9]]`	`[[7, 4, 9]]`	Failed
unmatched input	`[[0, 4, 9]]`	`[[20, 4, 9]]`	Failed

SHA-256 / a2696e2fc3d79191124e9a2e1ee730e1c155834a81932f2ff404a74ed9d287c9

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 2026-09-30T16:48:11.500551+00:00.

Case digest / 41755f6b0021349085b1d522f28d7b7748c85d8430c51d7e625c965e04568aba