FA-98336 / Autoregressive target shifting / Open access

A separator context incorrectly suppresses the following target · case 01

A separator context incorrectly suppresses the following target.

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

ROOT CAUSE

The filter checks the input position, although separator tokens may provide the next segment context.

THE FAILURE

The filter checks the input position, although separator tokens may provide the next segment context.

Unsuccessful approach: Suppressing both input and target occurrences loses the first valid prediction after a separator.

Case contract

Return adjacent pairs except when the target equals sep; a sep input can predict ordinary content.

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, sep):
    return [[tokens[i],tokens[i+1]] for i in range(len(tokens)-1) if tokens[i]!=sep]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('separator context valid',solve([1,9,2],9),[[9,2]])
check('separator target invalid',solve([1,2,9],9),[[1,2]])
check('multiple ordinary',solve([9,2,3],9),[[9,2],[2,3]])
check('consecutive separators',solve([9,9,2],9),[[9,2]])
check('empty',solve([],9),[])
check('one item',solve([9],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
separator context valid	`[[1, 9]]`	`[[9, 2]]`	Failed
separator target invalid	`[[1, 2], [2, 9]]`	`[[1, 2]]`	Failed
multiple ordinary	`[[2, 3]]`	`[[9, 2], [2, 3]]`	Failed
consecutive separators	`[]`	`[[9, 2]]`	Failed
empty	`[]`	`[]`	Passed
one item	`[]`	`[]`	Passed

SHA-256 / 41ccde2aa63f373c9dedacb18b1f7517a7829dad60c2d6ed61294ac260b6b060

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, sep):
    return [[tokens[i],tokens[i+1]] for i in range(len(tokens)-1) if tokens[i]!=sep and tokens[i+1]!=sep]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('separator context valid',solve([1,9,2],9),[[9,2]])
check('separator target invalid',solve([1,2,9],9),[[1,2]])
check('multiple ordinary',solve([9,2,3],9),[[9,2],[2,3]])
check('consecutive separators',solve([9,9,2],9),[[9,2]])
check('empty',solve([],9),[])
check('one item',solve([9],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
separator context valid	`[]`	`[[9, 2]]`	Failed
separator target invalid	`[[1, 2]]`	`[[1, 2]]`	Passed
multiple ordinary	`[[2, 3]]`	`[[9, 2], [2, 3]]`	Failed
consecutive separators	`[]`	`[[9, 2]]`	Failed
empty	`[]`	`[]`	Passed
one item	`[]`	`[]`	Passed

SHA-256 / 337ac8be85fb9dcc240ed35190c3d5f8beb3f6a4ea26daf59751cd5f72e76d28

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 2026-09-30T16:48:11.932993+00:00.

Case digest / cd5361b1aaa2c19f6059801071b4f69fda56cd1bcb9a8c3ea602513c40da35c0