FA-97986 / Training label alignment / Open access

Ignored examples remain in the supervised loss mask · case 01

Ignored examples remain in the supervised loss mask.

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

ROOT CAUSE

The mask is derived from example presence rather than the sentinel label value.

THE FAILURE

The mask is derived from example presence rather than the sentinel label value.

Unsuccessful approach: Treating every zero-valued label as ignored also suppresses a valid class zero.

Case contract

Given labels and ignore_id, return 1 for trainable positions and 0 exactly where label equals ignore_id.

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(labels, ignore_id):
    return [1]*len(labels)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('sentinel masked',solve([N,-100,0],-100),[1,0,1])
check('zero is class',solve([0,1],-1),[1,1])
check('all ignored',solve([-1,-1],-1),[0,0])
check('empty labels',solve([],-1),[])
check('single ignored',solve([N],N),[0])
check('sentinel at end',solve([0,0,N],N),[1,1,0])
check('valid negative class',solve([-2,-1],-1),[1,0])
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
sentinel masked	`[1, 1, 1]`	`[1, 0, 1]`	Failed
zero is class	`[1, 1]`	`[1, 1]`	Passed
all ignored	`[1, 1]`	`[0, 0]`	Failed
empty labels	`[]`	`[]`	Passed
single ignored	`[1]`	`[0]`	Failed
sentinel at end	`[1, 1, 1]`	`[1, 1, 0]`	Failed
valid negative class	`[1, 1]`	`[1, 0]`	Failed

SHA-256 / 48319526e55d940455d2802f4f7b63141f5086a12d58909bfe6882377685260c

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(labels, ignore_id):
    return [int(bool(y) and y!=ignore_id) for y in labels]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('sentinel masked',solve([N,-100,0],-100),[1,0,1])
check('zero is class',solve([0,1],-1),[1,1])
check('all ignored',solve([-1,-1],-1),[0,0])
check('empty labels',solve([],-1),[])
check('single ignored',solve([N],N),[0])
check('sentinel at end',solve([0,0,N],N),[1,1,0])
check('valid negative class',solve([-2,-1],-1),[1,0])
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
sentinel masked	`[1, 0, 0]`	`[1, 0, 1]`	Failed
zero is class	`[0, 1]`	`[1, 1]`	Failed
all ignored	`[0, 0]`	`[0, 0]`	Passed
empty labels	`[]`	`[]`	Passed
single ignored	`[0]`	`[0]`	Passed
sentinel at end	`[0, 0, 0]`	`[1, 1, 0]`	Failed
valid negative class	`[1, 0]`	`[1, 0]`	Passed

SHA-256 / f5dbe196f5c32c51cb57b59e824f4f1e279e5aa93f67d78e40d8c4abc5e70d68

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

Labels and ignore_id are scalar values with equality semantics. 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 2026-09-30T16:48:07.976744+00:00.

Case digest / f7f5e8a4ec1c9386611e1d3a20d86a0f539b71dddc949857191146f9e2a609e8