FA-98671 / Model quantization arithmetic / Open access

Quantization: Clipping occurs before zero-point shift · case 01

Clipping occurs before zero-point shift.

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

ROOT CAUSE

Quantize, add zero point, then clamp the resulting integer code to inclusive [lo,hi].

THE FAILURE

Quantize, add zero point, then clamp the resulting integer code to inclusive [lo,hi].

Unsuccessful approach: The attempted code clamps the unshifted scaled value before adding zero point; a positive offset can pull a low saturated value back into range.

Case contract

Quantize, add zero point, then clamp the resulting integer code to inclusive [lo,hi]. 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(values, scale, zero, lo, hi):
    return [max(lo,min(hi,round(x/scale)))+zero for x in values]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('shift reaches top',solve([3*N],1,2,-4,5),[5])
check('shift reaches bottom',solve([-3*N],1,-2,-4,5),[-4])
check('inside',solve([1],1,1,-4,5),[2])
check('empty',solve([],1,0,-4,5),[])
check('zero point outside code range',solve([0],1,8,-4,5),[5])
check('negative scale-space',solve([-1],1,0,-4,5),[-1])
check('preclip low with positive offset',solve([-100],1,3,-4,5),[-4])
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
shift reaches top	`[5]`	`[5]`	Passed
shift reaches bottom	`[-5]`	`[-4]`	Failed
inside	`[2]`	`[2]`	Passed
empty	`[]`	`[]`	Passed
zero point outside code range	`[8]`	`[5]`	Failed
negative scale-space	`[-1]`	`[-1]`	Passed
preclip low with positive offset	`[-1]`	`[-4]`	Failed

SHA-256 / 09cea373cffa066aebf523e8b44b5974c65452c6d5cc689ad5b594c13925b30d

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(values, scale, zero, lo, hi):
    return [max(lo,min(hi,round(max(lo,min(hi,x/scale)))+zero)) for x in values]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('shift reaches top',solve([3*N],1,2,-4,5),[5])
check('shift reaches bottom',solve([-3*N],1,-2,-4,5),[-4])
check('inside',solve([1],1,1,-4,5),[2])
check('empty',solve([],1,0,-4,5),[])
check('zero point outside code range',solve([0],1,8,-4,5),[5])
check('negative scale-space',solve([-1],1,0,-4,5),[-1])
check('preclip low with positive offset',solve([-100],1,3,-4,5),[-4])
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
shift reaches top	`[5]`	`[5]`	Passed
shift reaches bottom	`[-4]`	`[-4]`	Passed
inside	`[2]`	`[2]`	Passed
empty	`[]`	`[]`	Passed
zero point outside code range	`[5]`	`[5]`	Passed
negative scale-space	`[-1]`	`[-1]`	Passed
preclip low with positive offset	`[-1]`	`[-4]`	Failed

SHA-256 / b6f3d466809bbf378741e3b2e4843530d010a692297461809b5d4c5e5257d006

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

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 2026-09-30T16:48:15.404708+00:00.

Case digest / 49a9ece43c52180b28d071526f4d651e3d294a2e488c82a2490ad26796ac6690