FA-98276 / Autoregressive target shifting / Open access

The final target mask is dropped with the input slice · case 01

The final target mask is dropped with the input slice.

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

ROOT CAUSE

Mask construction zips against the unshifted input length and loses the last target.

THE FAILURE

Mask construction zips against the unshifted input length and loses the last target.

Unsuccessful approach: Appending a valid mask unconditionally mishandles a final ignored label.

Case contract

Return booleans for labels[1:] indicating whether each target is supervised.

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(labels, ignore):
    return [labels[i]!=ignore for i in range(max(0,len(labels)-2))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('terminal ignored',solve([1,2,-1],-1),[True,False])
check('terminal valid',solve([1,2,3],-1),[True,True])
check('two labels',solve([-1,4],-1),[True])
check('singleton',solve([1],-1),[])
check('empty',solve([],-1),[])
check('internal ignore',solve([9,-1,7,8],-1),[False,True,True])
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
terminal ignored	`[True]`	`[True, False]`	Failed
terminal valid	`[True]`	`[True, True]`	Failed
two labels	`[]`	`[True]`	Failed
singleton	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
internal ignore	`[True, False]`	`[False, True, True]`	Failed

SHA-256 / 8659a78cadf3e6a833091f7b26293843ba0637178181eacee721122138ad51d5

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(labels, ignore):
    return [True for x in labels[1:]]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('terminal ignored',solve([1,2,-1],-1),[True,False])
check('terminal valid',solve([1,2,3],-1),[True,True])
check('two labels',solve([-1,4],-1),[True])
check('singleton',solve([1],-1),[])
check('empty',solve([],-1),[])
check('internal ignore',solve([9,-1,7,8],-1),[False,True,True])
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
terminal ignored	`[True, True]`	`[True, False]`	Failed
terminal valid	`[True, True]`	`[True, True]`	Passed
two labels	`[True]`	`[True]`	Passed
singleton	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
internal ignore	`[True, True, True]`	`[False, True, True]`	Failed

SHA-256 / 1b4ac31602b185fb1e6cc3461feb67ea586b3a5d9b9c7f76f5ba4b4930538c7d

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 2026-09-30T16:48:11.054684+00:00.

Case digest / cd3cc1ab53a4573899efb500a630527ea8ec01d81c703e8530a48bfa323e9f86