FA-100596 / Mixture of experts dispatch / Open access

Expert outputs are added without their gate weights · case 01

Expert outputs are added without their gate weights.

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

ROOT CAUSE

The combine stage sums expert values as if every route had unit weight.

THE FAILURE

The combine stage sums expert values as if every route had unit weight.

Unsuccessful approach: Using only the first route weight applies one coefficient to all experts.

Case contract

For [expert,weight,value] routes, return the scalar weighted sum.

Why this case matters

A deterministic toy router makes this dispatch decision independently inspectable.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(routes):
    return sum(v for e,w,v in routes)
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[['a',0.25,N],['b',0.75,2*N]]
check('weighted values',solve(rows),1.75*N)
check('unit weights',solve([['x',1,N]]),N)
check('zero weight',solve([['x',0,N]]),0)
check('empty',solve([]),0)
check('negative output',solve([['x',0.5,-2*N]]),-N)
check('three routes',solve([['a',0.2,N],['b',0.3,N],['c',0.5,N]]),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
weighted values	`3`	`1.75`	Failed
unit weights	`1`	`1`	Passed
zero weight	`1`	`0`	Failed
empty	`0`	`0`	Passed
negative output	`-2`	`-1`	Failed
three routes	`3`	`1`	Failed

SHA-256 / b325f9724035b9d60b904aaa349d414284bae32367969775502dc302e477880f

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(routes):
    return sum(routes[0][1]*v for e,w,v in routes) if routes else 0
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[['a',0.25,N],['b',0.75,2*N]]
check('weighted values',solve(rows),1.75*N)
check('unit weights',solve([['x',1,N]]),N)
check('zero weight',solve([['x',0,N]]),0)
check('empty',solve([]),0)
check('negative output',solve([['x',0.5,-2*N]]),-N)
check('three routes',solve([['a',0.2,N],['b',0.3,N],['c',0.5,N]]),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
weighted values	`0.75`	`1.75`	Failed
unit weights	`1`	`1`	Passed
zero weight	`0`	`0`	Passed
empty	`0`	`0`	Passed
negative output	`-1.0`	`-1`	Passed
three routes	`0.6000000000000001`	`1`	Failed

SHA-256 / b2e97ea5f5086481276f05d66a8c3d6290f2870570d1f7455e309696e9b47093

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 scores and routes; the contract does not model a production model, accelerator, or learned routing quality. 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 2026-09-30T16:48:37.194261+00:00.

Case digest / f3e45f1780cca85e74d273fa5af5d1881d3790230105d0a6862fe436c444034a