FA-98716 / Model quantization arithmetic / Open access

Quantization: Dot product omits input zero-point correction · case 01

Dot product omits input zero-point correction.

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

ROOT CAUSE

For equal-length code vectors compute sum((a_i-zero_a)*(b_i-zero_b)); inputs are small toy integers.

THE FAILURE

For equal-length code vectors compute sum((a_i-zero_a)*(b_i-zero_b)); inputs are small toy integers.

Unsuccessful approach: The attempted code centers the left operand only; right-side zero point still contributes a spurious term.

Case contract

For equal-length code vectors compute sum((a_i-zero_a)*(b_i-zero_b)); inputs are small toy integers. 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(a, b, zero_a, zero_b):
    return sum(x*y for x,y in zip(a,b))
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('both offsets',solve([N+2, N],[N+1,N+3],N,N),2)
check('zero offsets',solve([1,2],[3,4],0,0),11)
check('empty vectors',solve([],[],N,N),0)
check('zero centered',solve([N],[N],N,N),0)
check('negative centered',solve([N-1],[N+2],N,N),-2)
check('two terms',solve([N+1,N+2],[N+3,N+4],N,N),11)
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
both offsets	`10`	`2`	Failed
zero offsets	`11`	`11`	Passed
empty vectors	`0`	`0`	Passed
zero centered	`1`	`0`	Failed
negative centered	`0`	`-2`	Failed
two terms	`23`	`11`	Failed

SHA-256 / d8117875dab80398f7c6160fd7c9ad4c7ef920f5237175db755765ba0ddb38ab

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(a, b, zero_a, zero_b):
    return sum((x-zero_a)*y for x,y in zip(a,b))
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('both offsets',solve([N+2, N],[N+1,N+3],N,N),2)
check('zero offsets',solve([1,2],[3,4],0,0),11)
check('empty vectors',solve([],[],N,N),0)
check('zero centered',solve([N],[N],N,N),0)
check('negative centered',solve([N-1],[N+2],N,N),-2)
check('two terms',solve([N+1,N+2],[N+3,N+4],N,N),11)
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
both offsets	`4`	`2`	Failed
zero offsets	`11`	`11`	Passed
empty vectors	`0`	`0`	Passed
zero centered	`0`	`0`	Passed
negative centered	`-3`	`-2`	Failed
two terms	`14`	`11`	Failed

SHA-256 / 5a6a0cbbf45aba37888008867a8520b29be07a7ff591205379abff84a274ec07

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 2026-09-30T16:48:15.842493+00:00.

Case digest / 8ff5aa270b56d8a26fb872c4528e5e17f773b80e17fa6583029790013b2ab48f