FA-99316 / Distributed training aggregation / Open access

Collective slot IDs are replaced by row positions · case 01

Collective slot IDs are replaced by row positions.

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

ROOT CAUSE

The reducer constructs slot IDs from the number and order of received rows instead of using declared slot IDs.

THE FAILURE

The reducer constructs slot IDs from the number and order of received rows instead of using declared slot IDs.

Unsuccessful approach: Using enumeration position as the slot key conflates shuffled payload lists.

Case contract

Rows [slot,worker,value]; return sorted [slot,sum] pairs.

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(rows):
    return [[i,sum(r[2] for r in rows if r[0]==i)] for i in range(len(rows))]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
r=[[2,'a',N],[0,'a',2*N],[2,'b',3*N]]
check('shuffled slot ids',solve(r),[[0,2*N],[2,4*N]])
check('empty',solve([]),[])
check('single slot',solve([[N,'a',N]]),[[N,N]])
check('negative sum',solve([[1,'a',-N],[1,'b',N]]),[[1,0]])
check('noncontiguous slots',solve([[5,'a',2],[2,'b',4]]),[[2,4],[5,2]])
check('three workers',solve([[N,'a',N],[N,'b',2],[N,'c',3]]),[[N,N+5]])
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
shuffled slot ids	`[[0, 2], [1, 0], [2, 4]]`	`[[0, 2], [2, 4]]`	Failed
empty	`[]`	`[]`	Passed
single slot	`[[0, 0]]`	`[[1, 1]]`	Failed
negative sum	`[[0, 0], [1, 0]]`	`[[1, 0]]`	Failed
noncontiguous slots	`[[0, 0], [1, 0]]`	`[[2, 4], [5, 2]]`	Failed
three workers	`[[0, 0], [1, 6], [2, 0]]`	`[[1, 6]]`	Failed

SHA-256 / 88e5819b1daab39575ef01e8f12060b9e8cbb358ba5eaeeadf43d3d75b684ac7

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows):
    d={}
    for i,(s,w,v) in enumerate(rows): d[i]=d.get(i,0)+v
    return [[i,d[i]] for i in sorted(d)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
r=[[2,'a',N],[0,'a',2*N],[2,'b',3*N]]
check('shuffled slot ids',solve(r),[[0,2*N],[2,4*N]])
check('empty',solve([]),[])
check('single slot',solve([[N,'a',N]]),[[N,N]])
check('negative sum',solve([[1,'a',-N],[1,'b',N]]),[[1,0]])
check('noncontiguous slots',solve([[5,'a',2],[2,'b',4]]),[[2,4],[5,2]])
check('three workers',solve([[N,'a',N],[N,'b',2],[N,'c',3]]),[[N,N+5]])
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
shuffled slot ids	`[[0, 1], [1, 2], [2, 3]]`	`[[0, 2], [2, 4]]`	Failed
empty	`[]`	`[]`	Passed
single slot	`[[0, 1]]`	`[[1, 1]]`	Failed
negative sum	`[[0, -1], [1, 1]]`	`[[1, 0]]`	Failed
noncontiguous slots	`[[0, 2], [1, 4]]`	`[[2, 4], [5, 2]]`	Failed
three workers	`[[0, 1], [1, 2], [2, 3]]`	`[[1, 6]]`	Failed

SHA-256 / 7e104a87e34fc4a6ef492c45c8b755655c9d358fdf1f00d3d70f270f35f95b3f

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 2026-09-30T16:48:22.633640+00:00.

Case digest / 96f1d31627bb0064f11579155c0bb87913cf01a15ad42e04a9706c894ec7af3e