FA-100616 / Mixture of experts dispatch / Open access

Eligibility filtering runs after router top-k truncation · case 01

Eligibility filtering runs after router top-k truncation.

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

ROOT CAUSE

The candidate budget is consumed by unavailable experts before eligibility is applied.

THE FAILURE

The candidate budget is consumed by unavailable experts before eligibility is applied.

Unsuccessful approach: Filtering after top-k and refilling beyond the requested count admits one extra candidate.

Case contract

Routes [expert,score,eligible]; return up to k eligible routes by score descending then expert ID.

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,k):
    return [r for r in sorted(routes,key=lambda x:(-x[1],x[0]))[:k] if r[2]]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[('a',N+3,False),('b',N+2,True),('c',N+1,True)]
check('refill after ineligible',solve(rows,2),[('b',N+2,True),('c',N+1,True)])
check('single eligible',solve(rows,1),[('b',N+2,True)])
check('zero k',solve(rows,0),[])
check('empty',solve([],N),[])
check('none eligible',solve([('x',N,False)],2),[])
check('all eligible ranked',solve([('z',N,True),('a',N+1,True)],2),[('a',N+1,True),('z',N,True)])
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
refill after ineligible	`[['b', 3, True]]`	`[['b', 3, True], ['c', 2, True]]`	Failed
single eligible	`[]`	`[['b', 3, True]]`	Failed
zero k	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
none eligible	`[]`	`[]`	Passed
all eligible ranked	`[['a', 2, True], ['z', 1, True]]`	`[['a', 2, True], ['z', 1, True]]`	Passed

SHA-256 / 83e1154a88559dbc152d3f40a558a1b13c57864e83b6dd2eb382e3888575b8b0

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(routes,k):
    return sorted([r for r in routes if r[2]],key=lambda x:(-x[1],x[0]))[:k+1]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[('a',N+3,False),('b',N+2,True),('c',N+1,True)]
check('refill after ineligible',solve(rows,2),[('b',N+2,True),('c',N+1,True)])
check('single eligible',solve(rows,1),[('b',N+2,True)])
check('zero k',solve(rows,0),[])
check('empty',solve([],N),[])
check('none eligible',solve([('x',N,False)],2),[])
check('all eligible ranked',solve([('z',N,True),('a',N+1,True)],2),[('a',N+1,True),('z',N,True)])
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
refill after ineligible	`[['b', 3, True], ['c', 2, True]]`	`[['b', 3, True], ['c', 2, True]]`	Passed
single eligible	`[['b', 3, True], ['c', 2, True]]`	`[['b', 3, True]]`	Failed
zero k	`[['b', 3, True]]`	`[]`	Failed
empty	`[]`	`[]`	Passed
none eligible	`[]`	`[]`	Passed
all eligible ranked	`[['a', 2, True], ['z', 1, True]]`	`[['a', 2, True], ['z', 1, True]]`	Passed

SHA-256 / c245e205d6e455f8374f16bd2aad0f59cd5a199281df9e927bae90381388d05c

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 2026-09-30T16:48:37.646430+00:00.

Case digest / 56b28e60c2420f0f2f99009d35fa40956b1cd98e29cacb2fd66284c2e11b826a