FA-98701 / Model quantization arithmetic / Open access

Quantization: Output zero point is scaled with the accumulator · case 01

The output zero point changes with the requantization multiplier.

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

ROOT CAUSE

The broken implementation adds output zero point to the centered input before multiplying and shifting it.

THE FAILURE

The broken implementation adds output zero point to the centered input before multiplying and shifting it.

Unsuccessful approach: The attempted code moves output offset outside the multiplier but still divides the output zero point by the accumulator shift; it must be added in output-code units afterward.

Case contract

Requantize centered input codes by multiplier/2**shift, then add output zero point in output-code units. Inputs follow this explicit integer toy contract; ties use Python round-to-even where rounding is required.

Why this case matters

This deterministic model isolates one quantization arithmetic contract from tensor storage and runtime kernels.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(codes, in_zero, multiplier, shift, out_zero):
    return [round((q-in_zero+out_zero)*multiplier/(1<<shift)) for q in codes]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('offset outside scale',solve([N+2],N,2,1,3),[5])
check('input zero',solve([N],N,2,1,3),[3])
check('negative centered value',solve([N-1],N,2,1,3),[2])
check('empty',solve([],N,2,1,3),[])
check('unit shift',solve([N+1],N,3,0,-2),[1])
check('two codes',solve([N,N+2],N,2,1,0),[0,2])
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
offset outside scale	`[5]`	`[5]`	Passed
input zero	`[3]`	`[3]`	Passed
negative centered value	`[2]`	`[2]`	Passed
empty	`[]`	`[]`	Passed
unit shift	`[-3]`	`[1]`	Failed
two codes	`[0, 2]`	`[0, 2]`	Passed

SHA-256 / 10f412b03b9923a4005dcf72f490268f4d97bcd53a7de5766ea0413435e5afc8

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(codes, in_zero, multiplier, shift, out_zero):
    return [round((q-in_zero)*multiplier/(1<<shift)+out_zero/(1<<shift)) for q in codes]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('offset outside scale',solve([N+2],N,2,1,3),[5])
check('input zero',solve([N],N,2,1,3),[3])
check('negative centered value',solve([N-1],N,2,1,3),[2])
check('empty',solve([],N,2,1,3),[])
check('unit shift',solve([N+1],N,3,0,-2),[1])
check('two codes',solve([N,N+2],N,2,1,0),[0,2])
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
offset outside scale	`[4]`	`[5]`	Failed
input zero	`[2]`	`[3]`	Failed
negative centered value	`[0]`	`[2]`	Failed
empty	`[]`	`[]`	Passed
unit shift	`[1]`	`[1]`	Passed
two codes	`[0, 2]`	`[0, 2]`	Passed

SHA-256 / 3f5c83900ca634b0e88d7121a77bd06e1fe02b85ec12a840fe99d37075fbbe04

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

Toy scalar/list arithmetic only; it does not model a production quantizer, hardware instruction, or accuracy/performance gain. 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 2026-09-30T16:48:15.842390+00:00.

Case digest / ae830906da248c6d0f1417f3218e231c38f5e4bd2bf044b5e2042db64491b27a