FA-99261 / Distributed training aggregation / Open access

All-gather output follows arrival order · case 01

All-gather output follows arrival order.

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

ROOT CAUSE

Asynchronous arrival order differs from the logical rank order expected by consumers.

THE FAILURE

Asynchronous arrival order differs from the logical rank order expected by consumers.

Unsuccessful approach: Sorting by payload value is deterministic but loses rank ownership.

Case contract

Given rank-tagged payloads and world size, return payloads in rank order; missing ranks are omitted and duplicate rank uses last listed payload.

Why this case matters

A toy distributed-training contract makes replica and shard contribution rules explicit without claiming framework compatibility or performance.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(items, world):
    return [x[1] for x in items]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
it=[(2,'c'),(0,'a'),(1,'b')]
check('arrival shuffled',solve(it,3),['a','b','c'])
check('missing rank',solve([(2,'c'),(0,'a')],3),['a','c'])
check('empty gather',solve([],N),[])
check('singleton',solve([(0,'x')],1),['x'])
check('duplicate rank last wins',solve([(1,'old'),(1,'new'),(0,'a')],2),['a','new'])
check('ignore out of world',solve([(0,'a'),(4,'x')],1),['a'])
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
arrival shuffled	`['c', 'a', 'b']`	`['a', 'b', 'c']`	Failed
missing rank	`['c', 'a']`	`['a', 'c']`	Failed
empty gather	`[]`	`[]`	Passed
singleton	`['x']`	`['x']`	Passed
duplicate rank last wins	`['old', 'new', 'a']`	`['a', 'new']`	Failed
ignore out of world	`['a', 'x']`	`['a']`	Failed

SHA-256 / 19ba050de0bd5328321ede949266e0295119d64b159ab2db8769d90a808384fe

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(items, world):
    return [v for r,v in sorted(items,key=lambda x:x[1])]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
it=[(2,'c'),(0,'a'),(1,'b')]
check('arrival shuffled',solve(it,3),['a','b','c'])
check('missing rank',solve([(2,'c'),(0,'a')],3),['a','c'])
check('empty gather',solve([],N),[])
check('singleton',solve([(0,'x')],1),['x'])
check('duplicate rank last wins',solve([(1,'old'),(1,'new'),(0,'a')],2),['a','new'])
check('ignore out of world',solve([(0,'a'),(4,'x')],1),['a'])
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
arrival shuffled	`['a', 'b', 'c']`	`['a', 'b', 'c']`	Passed
missing rank	`['a', 'c']`	`['a', 'c']`	Passed
empty gather	`[]`	`[]`	Passed
singleton	`['x']`	`['x']`	Passed
duplicate rank last wins	`['a', 'new', 'old']`	`['a', 'new']`	Failed
ignore out of world	`['a', 'x']`	`['a']`	Failed

SHA-256 / 3518dedc690bcd44de1e6c0df5ce03a8726bc74fe7f5efaaae16bc1f19382854

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 materialized Python values from a fixed toy collective; no transport, failure recovery, optimizer state, or real model behavior is represented. 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 2026-09-30T16:48:22.178632+00:00.

Case digest / 7aa54db62b09850d78a60af3d527fa198c61e3dfa3bb3cbbd8583a6044be8eaa