FA-98021 / Training label alignment / Open access

Augmented copies receive the base weight instead of divided weight · case 01

Augmented copies receive the base weight instead of divided weight.

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

ROOT CAUSE

Each source example is duplicated into several views without conserving its total training weight.

THE FAILURE

Each source example is duplicated into several views without conserving its total training weight.

Unsuccessful approach: Applying the full weight only to the first view conserves weight but discards supervision from the other views.

Case contract

Rows are [label,weight,view_count]. Return one [label,weight/view_count] row per view; nonpositive view counts emit nothing.

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(rows):
    return [[y,w] for y,w,count in rows for _ in range(count)]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('weight conserved across views',solve([[N,6,N]]),[[N,6/N]]*N)
check('one view unchanged',solve([['x',4,1]]),[['x',4.0]])
check('zero views',solve([['x',4,0]]),[])
check('empty batch',solve([]),[])
check('multiple sources',solve([['a',8,2],['b',3,1]]),[['a',4.0],['a',4.0],['b',3.0]])
check('fractional share',solve([['c',1,4]]),[['c',0.25]]*4)
check('negative weight preserved',solve([['d',-6,3]]),[['d',-2.0]]*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
weight conserved across views	`[[1, 6]]`	`[[1, 6.0]]`	Passed
one view unchanged	`[['x', 4]]`	`[['x', 4.0]]`	Passed
zero views	`[]`	`[]`	Passed
empty batch	`[]`	`[]`	Passed
multiple sources	`[['a', 8], ['a', 8], ['b', 3]]`	`[['a', 4.0], ['a', 4.0], ['b', 3.0]]`	Failed
fractional share	`[['c', 1], ['c', 1], ['c', 1], ['c', 1]]`	`[['c', 0.25], ['c', 0.25], ['c', 0.25], ['c', 0.25]]`	Failed
negative weight preserved	`[['d', -6], ['d', -6], ['d', -6]]`	`[['d', -2.0], ['d', -2.0], ['d', -2.0]]`	Failed

SHA-256 / a2a8193648f8f5bdbcd51058bc23ef5325eeefd51bbaace00b3d491d38e982af

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows):
    return [[y,w] for y,w,count in rows for i in range(count) if i==0]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('weight conserved across views',solve([[N,6,N]]),[[N,6/N]]*N)
check('one view unchanged',solve([['x',4,1]]),[['x',4.0]])
check('zero views',solve([['x',4,0]]),[])
check('empty batch',solve([]),[])
check('multiple sources',solve([['a',8,2],['b',3,1]]),[['a',4.0],['a',4.0],['b',3.0]])
check('fractional share',solve([['c',1,4]]),[['c',0.25]]*4)
check('negative weight preserved',solve([['d',-6,3]]),[['d',-2.0]]*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
weight conserved across views	`[[1, 6]]`	`[[1, 6.0]]`	Passed
one view unchanged	`[['x', 4]]`	`[['x', 4.0]]`	Passed
zero views	`[]`	`[]`	Passed
empty batch	`[]`	`[]`	Passed
multiple sources	`[['a', 8], ['b', 3]]`	`[['a', 4.0], ['a', 4.0], ['b', 3.0]]`	Failed
fractional share	`[['c', 1]]`	`[['c', 0.25], ['c', 0.25], ['c', 0.25], ['c', 0.25]]`	Failed
negative weight preserved	`[['d', -6]]`	`[['d', -2.0], ['d', -2.0], ['d', -2.0]]`	Failed

SHA-256 / 0e7ec63a3f47cbeea94edb7738b1c5c7b9948cc9169e0de2eb46d13585258092

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

Weights are finite numbers; view_count is a nonnegative integer. 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 2026-09-30T16:48:08.406736+00:00.

Case digest / 03b765fab52f0ba804bde4a2a0a23c1bd86f05480377b475900e4bc7967448fd