FA-99226 / Checkpoint tensor mapping / Open access

Checkpoint mapping: partial tensor restoration copies the wrong interval · case 01

partial tensor restoration copies the wrong interval.

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

ROOT CAUSE

Restoration copies a prefix rather than the requested overlap window.

THE FAILURE

Restoration copies a prefix rather than the requested overlap window.

Unsuccessful approach: Clipping handles the right boundary but does not reject a negative destination index.

Case contract

Overlay source values into target beginning at start; preserve the rest, clipping at target end; invalid negative start returns [].

Why this case matters

A bounded offline model of checkpoint tensor bookkeeping in a training workflow.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(target, source, start):
    out=target[:]
    out[:min(len(source),len(target))]=source[:len(target)]
    return out
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('middle', solve(*([0, 0, 0, 0], [3, 2], 1)), [0, 3, 2, 0])
check('at start', solve(*([0, 0], [3], 0)), [3, 0])
check('clipped', solve(*([0, 0, 0], [3, 2, 3, 4], 2)), [0, 0, 3])
check('empty source', solve(*([1, 2], [], 1)), [1, 2])
check('empty target', solve(*([], [3], 0)), [])
check('negative', solve(*([1], [3], -1)), [])
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
middle	`[3, 2, 0, 0]`	`[0, 3, 2, 0]`	Failed
at start	`[3, 0]`	`[3, 0]`	Passed
clipped	`[3, 2, 3]`	`[0, 0, 3]`	Failed
empty source	`[1, 2]`	`[1, 2]`	Passed
empty target	`[]`	`[]`	Passed
negative	`[3]`	`[]`	Failed

SHA-256 / 1a89bff75805f6f88868a70ee8841f92782d7db14cb68ade9b697814fc0a87c6

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(target, source, start):
    out=target[:]
    end=min(len(target),start+len(source))
    out[start:end]=source[:end-start]
    return out
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('middle', solve(*([0, 0, 0, 0], [3, 2], 1)), [0, 3, 2, 0])
check('at start', solve(*([0, 0], [3], 0)), [3, 0])
check('clipped', solve(*([0, 0, 0], [3, 2, 3, 4], 2)), [0, 0, 3])
check('empty source', solve(*([1, 2], [], 1)), [1, 2])
check('empty target', solve(*([], [3], 0)), [])
check('negative', solve(*([1], [3], -1)), [])
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
middle	`[0, 3, 2, 0]`	`[0, 3, 2, 0]`	Passed
at start	`[3, 0]`	`[3, 0]`	Passed
clipped	`[0, 0, 3]`	`[0, 0, 3]`	Passed
empty source	`[1, 2]`	`[1, 2]`	Passed
empty target	`[]`	`[]`	Passed
negative	`[3, 1]`	`[]`	Failed

SHA-256 / 9a8639d15c64540400179bcfaf4c2f38d64eac42e9fbc56e8c231fe2f83866e9

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 already decoded Python values; this toy contract makes no claim about a framework, file format, or production checkpoint. 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 2026-09-30T16:48:21.700575+00:00.

Case digest / 65a11c483c574040fff23f3ccc787e5d671298ac61337187c1e3b4fb9ef3f16c