FA-98256 / Autoregressive target shifting / Open access

The first observed token is paired with itself after start insertion · case 01

The first observed token is paired with itself after start insertion.

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

ROOT CAUSE

Targets are read from the input position instead of the next position after start-marker insertion.

THE FAILURE

Targets are read from the input position instead of the next position after start-marker insertion.

Unsuccessful approach: Dropping the marker restores observed adjacency but loses the first prediction from context.

Case contract

Given observations and bos, return adjacent pairs in [bos]+tokens, excluding any pair beyond the final observed target.

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, bos):
    return [[t,t] for t in tokens]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ordinary',solve([N,N+2],6),[[6,N],[N,N+2]])
check('singleton',solve([4],1),[[1,4]])
check('empty',solve([],1),[])
check('bos equals token',solve([3,4],3),[[3,3],[3,4]])
check('three observations',solve([0,2,5],9),[[9,0],[0,2],[2,5]])
check('large marker',solve([8,7],N+8),[[N+8,8],[8,7]])
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], [3, 3]]`	`[[6, 1], [1, 3]]`	Failed
singleton	`[[4, 4]]`	`[[1, 4]]`	Failed
empty	`[]`	`[]`	Passed
bos equals token	`[[3, 3], [4, 4]]`	`[[3, 3], [3, 4]]`	Failed
three observations	`[[0, 0], [2, 2], [5, 5]]`	`[[9, 0], [0, 2], [2, 5]]`	Failed
large marker	`[[8, 8], [7, 7]]`	`[[9, 8], [8, 7]]`	Failed

SHA-256 / 046138a7dc8e5dd55f1cb7dabcc4a5474ec4bc520d549419514cba9f4c0a4a67

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, bos):
    return [[tokens[i],tokens[i+1]] for i in range(len(tokens)-1)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('ordinary',solve([N,N+2],6),[[6,N],[N,N+2]])
check('singleton',solve([4],1),[[1,4]])
check('empty',solve([],1),[])
check('bos equals token',solve([3,4],3),[[3,3],[3,4]])
check('three observations',solve([0,2,5],9),[[9,0],[0,2],[2,5]])
check('large marker',solve([8,7],N+8),[[N+8,8],[8,7]])
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]]`	`[[6, 1], [1, 3]]`	Failed
singleton	`[]`	`[[1, 4]]`	Failed
empty	`[]`	`[]`	Passed
bos equals token	`[[3, 4]]`	`[[3, 3], [3, 4]]`	Failed
three observations	`[[0, 2], [2, 5]]`	`[[9, 0], [0, 2], [2, 5]]`	Failed
large marker	`[[8, 7]]`	`[[9, 8], [8, 7]]`	Failed

SHA-256 / 0d15fe16c331adf070b12878f66b32d93db9fe94a8c8982ba9ab13536c0bb006

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 2026-09-30T16:48:11.053543+00:00.

Case digest / 9ccf7fcadb539100894fd36ec02fb694ef8c4cf62b4b2815c95faac3ba906e8c