FA-99426 / Embedding vector normalization / Open access

Max pooling selects coordinate minima · case 01

Max pooling selects coordinate minima.

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

ROOT CAUSE

The pooling reducer returns coordinate minima rather than coordinate maxima.

THE FAILURE

The pooling reducer returns coordinate minima rather than coordinate maxima.

Unsuccessful approach: Taking one maximum across the flattened batch loses the coordinate structure.

Case contract

Return coordinatewise maxima across equal-width vectors; empty input returns empty.

Why this case matters

A small embedding preparation contract can be checked without a model or retrieval system.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(vectors):
    return [min(v[j] for v in vectors) for j in range(len(vectors[0]))] if vectors else []
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('coordinate winners differ', solve([[N,5*N],[3*N,2*N]]), [3*N,5*N])
check('all negative', solve([[-N,-2*N],[-3*N,-N]]), [-N,-N])
check('singleton', solve([[N,-N]]), [N,-N])
check('empty', solve([]), [])
check('zero candidates', solve([[0,0],[0,0]]), [0,0])
check('winner in first row', solve([[4*N,N],[N,3*N]]), [4*N,3*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
coordinate winners differ	`[1, 2]`	`[3, 5]`	Failed
all negative	`[-3, -2]`	`[-1, -1]`	Failed
singleton	`[1, -1]`	`[1, -1]`	Passed
empty	`[]`	`[]`	Passed
zero candidates	`[0, 0]`	`[0, 0]`	Passed
winner in first row	`[1, 1]`	`[4, 3]`	Failed

SHA-256 / 2b7d6e4d89c0a24982503da19a247a097edca7691f68fb86405f4defb09ffe72

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(vectors):
    return [max(max(v) for v in vectors)] if vectors else []
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('coordinate winners differ', solve([[N,5*N],[3*N,2*N]]), [3*N,5*N])
check('all negative', solve([[-N,-2*N],[-3*N,-N]]), [-N,-N])
check('singleton', solve([[N,-N]]), [N,-N])
check('empty', solve([]), [])
check('zero candidates', solve([[0,0],[0,0]]), [0,0])
check('winner in first row', solve([[4*N,N],[N,3*N]]), [4*N,3*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
coordinate winners differ	`[5]`	`[3, 5]`	Failed
all negative	`[-1]`	`[-1, -1]`	Failed
singleton	`[1]`	`[1, -1]`	Failed
empty	`[]`	`[]`	Passed
zero candidates	`[0]`	`[0, 0]`	Failed
winner in first row	`[4]`	`[4, 3]`	Failed

SHA-256 / 1033bed32cd79bc4d946c910e68d787835704e41ef39f612a6a66955680930bc

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 finite toy vectors or token rows; this contract makes no claim about model quality or production embedding behavior. 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 2026-09-30T16:48:24.023606+00:00.

Case digest / 7794c52a4b76475658c00ff47e7afbb26680bf781f81be6665c57943a0061d13