FA-100591 / Mixture of experts dispatch / Open access

Expert load counts routes rather than distinct tokens · case 01

Expert load counts routes rather than distinct tokens.

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

ROOT CAUSE

Duplicate route records inflate per-expert load.

THE FAILURE

Duplicate route records inflate per-expert load.

Unsuccessful approach: Deduplicating expert IDs counts only one token for the whole expert.

Case contract

Rows [token,expert] may repeat; return sorted [expert,distinct-token-count] pairs.

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):
    d={}
    for t,e in routes: d[e]=d.get(e,0)+1
    return sorted([e,c] for e,c in d.items())
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N)
rows=[[1,a],[1,a],[2,a],[3,b],[3,b]]
check('duplicate ignored',solve(rows),[[a,2],[b,1]])
check('one token',solve([[N,a],[N,a]]),[[a,1]])
check('empty',solve([]),[])
check('distinct tokens',solve([[1,a],[2,a],[3,a]]),[[a,3]])
check('multiple experts',solve([[1,a],[1,b]]),[[a,1],[b,1]])
check('N token',solve([[i,a] for i in range(N)]),[[a,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
duplicate ignored	`[['a1', 3], ['b1', 2]]`	`[['a1', 2], ['b1', 1]]`	Failed
one token	`[['a1', 2]]`	`[['a1', 1]]`	Failed
empty	`[]`	`[]`	Passed
distinct tokens	`[['a1', 3]]`	`[['a1', 3]]`	Passed
multiple experts	`[['a1', 1], ['b1', 1]]`	`[['a1', 1], ['b1', 1]]`	Passed
N token	`[['a1', 1]]`	`[['a1', 1]]`	Passed

SHA-256 / 468f0eceb48e412071ca051618a01603505bb39764a490972135dd9f6aaed002

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(routes):
    return [[e,1] for e in sorted({e for t,e in routes})]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N)
rows=[[1,a],[1,a],[2,a],[3,b],[3,b]]
check('duplicate ignored',solve(rows),[[a,2],[b,1]])
check('one token',solve([[N,a],[N,a]]),[[a,1]])
check('empty',solve([]),[])
check('distinct tokens',solve([[1,a],[2,a],[3,a]]),[[a,3]])
check('multiple experts',solve([[1,a],[1,b]]),[[a,1],[b,1]])
check('N token',solve([[i,a] for i in range(N)]),[[a,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
duplicate ignored	`[['a1', 1], ['b1', 1]]`	`[['a1', 2], ['b1', 1]]`	Failed
one token	`[['a1', 1]]`	`[['a1', 1]]`	Passed
empty	`[]`	`[]`	Passed
distinct tokens	`[['a1', 1]]`	`[['a1', 3]]`	Failed
multiple experts	`[['a1', 1], ['b1', 1]]`	`[['a1', 1], ['b1', 1]]`	Passed
N token	`[['a1', 1]]`	`[['a1', 1]]`	Passed

SHA-256 / 64d8b27d43b5ab4358e8552eb71eab5ff14bd36693b4222254ded199e85e55aa

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 2026-09-30T16:48:37.194350+00:00.

Case digest / 33518889bc0530f88917ef99056dd108e571674cc4b790861fc100676466b16d