FA-97956 / Training label alignment / Open access

Filtering examples leaves the original label list attached · case 01

Filtering examples leaves the original label list attached.

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

ROOT CAUSE

The filter is applied to examples after positional pairs have been discarded.

THE FAILURE

The filter is applied to examples after positional pairs have been discarded.

Unsuccessful approach: Filtering labels by truthiness drops legitimate zero labels and still misaligns later rows.

Case contract

Examples and labels have equal length. Keep pairs where the example passes predicate flag; preserve order and all label values.

Why this case matters

A small deterministic training-data operation makes label ownership and ordering explicit.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(examples, labels):
    kept=[e for e in examples if e[1]]
    return [[e[0],y] for e,y in zip(kept,labels)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('zero label kept',solve([['a',True],['b',False]],[0,N]),[['a',0]])
check('middle retained',solve([['a',False],['b',True],['c',True]],[N,N+1,N+2]),[['b',N+1],['c',N+2]])
check('all filtered',solve([[0,False]],[N]),[])
check('empty input',solve([],[]),[])
check('false payload',solve([[0,True]],[False]),[[0,False]])
check('single retained',solve([['x',True]],['label']),[['x','label']])
check('order preserved',solve([['c',True],['a',True]],[N+1,N]),[['c',N+1],['a',N]])
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
zero label kept	`[['a', 0]]`	`[['a', 0]]`	Passed
middle retained	`[['b', 1], ['c', 2]]`	`[['b', 2], ['c', 3]]`	Failed
all filtered	`[]`	`[]`	Passed
empty input	`[]`	`[]`	Passed
false payload	`[[0, False]]`	`[[0, False]]`	Passed
single retained	`[['x', 'label']]`	`[['x', 'label']]`	Passed
order preserved	`[['c', 2], ['a', 1]]`	`[['c', 2], ['a', 1]]`	Passed

SHA-256 / 3b3662413262fe4721309a106212d8f0a8643e8de31dae9651c8b74293b97b88

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(examples, labels):
    return [[e,y] for e,y in zip(examples,labels) if e[1] and y]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('zero label kept',solve([['a',True],['b',False]],[0,N]),[['a',0]])
check('middle retained',solve([['a',False],['b',True],['c',True]],[N,N+1,N+2]),[['b',N+1],['c',N+2]])
check('all filtered',solve([[0,False]],[N]),[])
check('empty input',solve([],[]),[])
check('false payload',solve([[0,True]],[False]),[[0,False]])
check('single retained',solve([['x',True]],['label']),[['x','label']])
check('order preserved',solve([['c',True],['a',True]],[N+1,N]),[['c',N+1],['a',N]])
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
zero label kept	`[]`	`[['a', 0]]`	Failed
middle retained	`[[['b', True], 2], [['c', True], 3]]`	`[['b', 2], ['c', 3]]`	Failed
all filtered	`[]`	`[]`	Passed
empty input	`[]`	`[]`	Passed
false payload	`[]`	`[[0, False]]`	Failed
single retained	`[[['x', True], 'label']]`	`[['x', 'label']]`	Failed
order preserved	`[[['c', True], 2], [['a', True], 1]]`	`[['c', 2], ['a', 1]]`	Failed

SHA-256 / 0d84f0185e8fc1cdb3a0dd7c486daacb8c6ecc12d48d4627635a24da1c880950

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

Example rows are [payload,keep_flag], and lengths are equal. 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 2026-09-30T16:48:07.548646+00:00.

Case digest / d012f5ec976db8254252314ca1bb41d70fcf489b409255a3b515be49718f5756