FA-100541 / Mixture of experts dispatch / Open access

Top-k routing follows expert names instead of gate scores · case 01

Top-k routing follows expert names instead of gate scores.

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

ROOT CAUSE

Candidates are sorted by identifier, not gate value.

THE FAILURE

Candidates are sorted by identifier, not gate value.

Unsuccessful approach: Score-only sorting leaves ties dependent on input order.

Case contract

Select up to k unique (expert,score) pairs by descending score then ascending expert name.

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(scores,k):
    return sorted(e for e,s in scores)[:k]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[('a',N),('b',N+1),('c',N-1)]
check('score wins',solve(rows,1),['b'])
check('tie name',solve([('z',N),('a',N)],1),['a'])
check('all ranked',solve(rows,8),['b','a','c'])
check('zero k',solve(rows,0),[])
check('empty',solve([],N),[])
check('negative scores',solve([('x',-N),('y',-N-1)],1),['x'])
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
score wins	`['a']`	`['b']`	Failed
tie name	`['a']`	`['a']`	Passed
all ranked	`['a', 'b', 'c']`	`['b', 'a', 'c']`	Failed
zero k	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
negative scores	`['x']`	`['x']`	Passed

SHA-256 / dec8e5a337d21008de54954bb4cd94cf9a2ac6f8b25a7de894b53355ce845864

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(scores,k):
    return sorted(scores,key=lambda x:-x[1])[:k]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[('a',N),('b',N+1),('c',N-1)]
check('score wins',solve(rows,1),['b'])
check('tie name',solve([('z',N),('a',N)],1),['a'])
check('all ranked',solve(rows,8),['b','a','c'])
check('zero k',solve(rows,0),[])
check('empty',solve([],N),[])
check('negative scores',solve([('x',-N),('y',-N-1)],1),['x'])
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
score wins	`[['b', 2]]`	`['b']`	Failed
tie name	`[['z', 1]]`	`['a']`	Failed
all ranked	`[['b', 2], ['a', 1], ['c', 0]]`	`['b', 'a', 'c']`	Failed
zero k	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
negative scores	`[['x', -1]]`	`['x']`	Failed

SHA-256 / 4d135c74a491192707b1044bdcf4c3be60eedb32f0c319c276dcd2ae3d85aa91

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 2026-09-30T16:48:36.730442+00:00.

Case digest / bbf8b9a4ac7fe1b97fac450ff781759e7683e78aec8afc95d077dcf8620b296d