FA-98326 / Autoregressive target shifting / Open access

A segment start loses its first valid target pair · case 01

A segment start loses its first valid target pair.

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

ROOT CAUSE

The segment start is treated as a target index instead of an input index.

THE FAILURE

The segment start is treated as a target index instead of an input index.

Unsuccessful approach: Using inclusive segment ends includes the first token owned by the following segment.

Case contract

For each [start,end) segment, include i=start..end-2 when both tokens exist.

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, segments):
    return [[tokens[i],tokens[i+1]] for a,b in segments for i in range(max(0,a+1),min(b-1,len(tokens)-1))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('first pair retained',solve([0,1,2,3],[[1,4]]),[[1,2],[2,3]])
check('adjacent segments',solve([0,1,2,3],[[0,2],[2,4]]),[[0,1],[2,3]])
check('length two',solve([5,6],[[0,2]]),[[5,6]])
check('length one',solve([5],[[0,1]]),[])
check('empty',solve([],[[0,0]]),[])
check('end clipped',solve([8,9],[[0,9]]),[[8,9]])
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
first pair retained	`[[2, 3]]`	`[[1, 2], [2, 3]]`	Failed
adjacent segments	`[]`	`[[0, 1], [2, 3]]`	Failed
length two	`[]`	`[[5, 6]]`	Failed
length one	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
end clipped	`[]`	`[[8, 9]]`	Failed

SHA-256 / 1397d913584661a2227cff638629515b3163eb957cd78d6355f0f9c88c293c86

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(tokens, segments):
    return [[tokens[i],tokens[i+1]] for a,b in segments for i in range(max(0,a),min(b,len(tokens)-1))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('first pair retained',solve([0,1,2,3],[[1,4]]),[[1,2],[2,3]])
check('adjacent segments',solve([0,1,2,3],[[0,2],[2,4]]),[[0,1],[2,3]])
check('length two',solve([5,6],[[0,2]]),[[5,6]])
check('length one',solve([5],[[0,1]]),[])
check('empty',solve([],[[0,0]]),[])
check('end clipped',solve([8,9],[[0,9]]),[[8,9]])
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
first pair retained	`[[1, 2], [2, 3]]`	`[[1, 2], [2, 3]]`	Passed
adjacent segments	`[[0, 1], [1, 2], [2, 3]]`	`[[0, 1], [2, 3]]`	Failed
length two	`[[5, 6]]`	`[[5, 6]]`	Passed
length one	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
end clipped	`[[8, 9]]`	`[[8, 9]]`	Passed

SHA-256 / d4549e4b09fb8b1510ee8ce6dca51f936962e070ab5d4cbf27ff060417cae44d

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 2026-09-30T16:48:11.933252+00:00.

Case digest / d6dd66e96df8c2e934903fdb2e9f5bf34ae4f21c1a3640580a29c5880c7ff471