FA-99181 / Checkpoint tensor mapping / Open access

Checkpoint mapping: different fixed dimensions pass shard compatibility · case 01

different fixed dimensions pass shard compatibility.

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

ROOT CAUSE

Rank equality is mistaken for shape compatibility.

THE FAILURE

Rank equality is mistaken for shape compatibility.

Unsuccessful approach: Modulo-normalizing an invalid axis turns it into a valid dimension.

Case contract

Return True when all non-scalar shapes have equal rank, nonnegative axis, and equal dimensions except along axis; empty input is False.

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(shapes, axis):
    return bool(shapes) and all(len(s)==len(shapes[0]) for s in shapes)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('valid', solve(*([[3, 2], [3, 4]], 1)), True)
check('wrong fixed', solve(*([[3, 2], [4, 4]], 1)), False)
check('negative axis', solve(*([[3], [3]], -1)), False)
check('too large', solve(*([[3], [3]], 1)), False)
check('empty', solve(*([], 0)), False)
check('scalar shapes', solve(*([[], []], 0)), False)
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
valid	`True`	`True`	Passed
wrong fixed	`True`	`False`	Failed
negative axis	`True`	`False`	Failed
too large	`True`	`False`	Failed
empty	`False`	`False`	Passed
scalar shapes	`True`	`False`	Failed

SHA-256 / cc520a14edf68bec6652813539516f620eef0df8e159115a0bed1e108e1984cd

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(shapes, axis):
    return bool(shapes) and all(len(s)>0 and len(s)==len(shapes[0]) and all(s[i]==shapes[0][i] for i in range(len(s)) if i!=axis%len(s)) for s in shapes)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('valid', solve(*([[3, 2], [3, 4]], 1)), True)
check('wrong fixed', solve(*([[3, 2], [4, 4]], 1)), False)
check('negative axis', solve(*([[3], [3]], -1)), False)
check('too large', solve(*([[3], [3]], 1)), False)
check('empty', solve(*([], 0)), False)
check('scalar shapes', solve(*([[], []], 0)), False)
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
valid	`True`	`True`	Passed
wrong fixed	`False`	`False`	Passed
negative axis	`True`	`False`	Failed
too large	`True`	`False`	Failed
empty	`False`	`False`	Passed
scalar shapes	`False`	`False`	Passed

SHA-256 / 717cabef3eef20e3ff39bea040647daebca1f3748cd21932b8b1d47ebf97c4c5

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 2026-09-30T16:48:21.243593+00:00.

Case digest / c1982d770dbf9d164fe817c7fa59e22717a835386b2fa33f522a3a9dbcd765cc