FA-97851 / Dataset split isolation / Open access

Dataset split: class representation is uneven across group partitions · case 01

class representation is uneven across group partitions.

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

ROOT CAUSE

Groups are selected by key order without comparing label counts on the two sides.

THE FAILURE

Groups are selected by key order without comparing label counts on the two sides.

Unsuccessful approach: Sorting groups by only their first label helps pure groups but ignores mixed-label contribution and does not balance the resulting counts.

Case contract

Each group maps to a list of labels. Return group IDs assigned to train by greedy least-filled label bins; groups stay whole, ties by ID.

Why this case matters

This toy split allocator makes membership rules explicit for reproducible evaluation fixtures.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(groups):
    return sorted(groups)[:(len(groups)+1)//2]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('mixed group stays intact',solve({'a':['x','y'],'b':['x'],'c':['y']}),['a'])
check('empty',solve({}),[])
check('singleton',solve({'g':['z']}),['g'])
check('lexical tie',solve({'b':['q'],'a':['q']}),['a'])
check('minority label',solve({'a':['m','m','r'],'b':['m'],'c':['r'],'d':['r']}),['a','d'])
check('empty labels',solve({'a':[],'b':[]}),['a','b'])
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
mixed group stays intact	`['a', 'b']`	`['a']`	Failed
empty	`[]`	`[]`	Passed
singleton	`['g']`	`['g']`	Passed
lexical tie	`['a']`	`['a']`	Passed
minority label	`['a', 'b']`	`['a', 'd']`	Failed
empty labels	`['a']`	`['a', 'b']`	Failed

SHA-256 / 63cc4b04f4dc8d6327655024d7595420ebd1b6e66b87c5ab9ddd4fd42525d469

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(groups):
    return sorted(groups,key=lambda g:groups[g][0] if groups[g] else '')[:(len(groups)+1)//2]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('mixed group stays intact',solve({'a':['x','y'],'b':['x'],'c':['y']}),['a'])
check('empty',solve({}),[])
check('singleton',solve({'g':['z']}),['g'])
check('lexical tie',solve({'b':['q'],'a':['q']}),['a'])
check('minority label',solve({'a':['m','m','r'],'b':['m'],'c':['r'],'d':['r']}),['a','d'])
check('empty labels',solve({'a':[],'b':[]}),['a','b'])
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
mixed group stays intact	`['a', 'b']`	`['a']`	Failed
empty	`[]`	`[]`	Passed
singleton	`['g']`	`['g']`	Passed
lexical tie	`['b']`	`['a']`	Failed
minority label	`['a', 'b']`	`['a', 'd']`	Failed
empty labels	`['a']`	`['a', 'b']`	Failed

SHA-256 / 1b86b42745a32f16694343681625bc9e983008aae4c349e4adb1e0e8c6d6836f

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 small in-memory records with self-contained contracts; no claim is made about production sampling quality. 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 2026-09-30T16:48:06.688947+00:00.

Case digest / c89b8d52249a74b030fd9282f8b14eb929c74ee6059042c6d3560041f053ae73