FA-99286 / Distributed training aggregation / Open access

Mean reduction trusts declared world size after dropout · case 01

Mean reduction trusts declared world size after dropout.

Verified by executionVariant 1 · 7 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

The divisor is the configured world size even when fewer worker contributions arrived.

THE FAILURE

The divisor is the configured world size even when fewer worker contributions arrived.

Unsuccessful approach: Assuming exactly one dropout and dividing by world size minus one fails when the actual number of missing workers differs.

Case contract

Given received scalar contributions and configured world size, return their mean; empty set returns None.

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(values, world):
    return sum(values)/world if world else None
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('partial round',solve([N,3*N],4),2*N)
check('exactly one missing',solve([N,3*N],3),2*N)
check('full round',solve([N,3*N],2),2*N)
check('empty round',solve([],N),None)
check('single received',solve([5*N],4),5*N)
check('configured zero with values',solve([N,3*N],0),2*N)
check('negative values',solve([-N,N,4*N],9),4*N/3)
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
partial round	`1.0`	`2`	Failed
exactly one missing	`1.3333333333333333`	`2`	Failed
full round	`2.0`	`2`	Passed
empty round	`0.0`	`None`	Failed
single received	`1.25`	`5`	Failed
configured zero with values	`None`	`2`	Failed
negative values	`0.4444444444444444`	`1.3333333333333333`	Failed

SHA-256 / f96b458a3ea03450170a3f11d86b4b5878dbf17a2984b13a3a388f02053a32c9

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(values, world):
    return sum(values)/max(1,world-1) if values else None
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('partial round',solve([N,3*N],4),2*N)
check('exactly one missing',solve([N,3*N],3),2*N)
check('full round',solve([N,3*N],2),2*N)
check('empty round',solve([],N),None)
check('single received',solve([5*N],4),5*N)
check('configured zero with values',solve([N,3*N],0),2*N)
check('negative values',solve([-N,N,4*N],9),4*N/3)
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
partial round	`1.3333333333333333`	`2`	Failed
exactly one missing	`2.0`	`2`	Passed
full round	`4.0`	`2`	Failed
empty round	`None`	`None`	Passed
single received	`1.6666666666666667`	`5`	Failed
configured zero with values	`4.0`	`2`	Failed
negative values	`0.5`	`1.3333333333333333`	Failed

SHA-256 / 215cdac921ae08b8f43c8b95323d4d9ca033c8b4e23a55aaff5f07925e5e3efb

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 7 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 2026-09-30T16:48:22.632883+00:00.

Case digest / 8c608c121a0c186dfb5fd4d02360c1b327255c46acd93e87b018ced0a745f874