FA-100551 / Mixture of experts dispatch / Open access

One expert filling its quota blocks unrelated experts · case 01

One expert filling its quota blocks unrelated experts.

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

ROOT CAUSE

Capacity is counted across the batch rather than separately per expert.

THE FAILURE

Capacity is counted across the batch rather than separately per expert.

Unsuccessful approach: The repair admits one extra route but still uses a global counter.

Case contract

Requests [token,expert,score] are considered in order; admit at most capacity per expert and preserve request order.

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(requests,capacity):
    return [r for i,r in enumerate(requests) if i<capacity]
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=[[0,a,3],[1,a,2],[2,b,4],[3,b,1]]
check('independent quotas',solve(rows,1),[rows[0],rows[2]])
check('capacity two',solve(rows,2),rows)
check('zero capacity',solve(rows,0),[])
check('empty requests',solve([],N),[])
check('singleton',solve([[0,a,1]],1),[[0,a,1]])
check('interleaved expert',solve([[0,a,1],[1,b,1],[2,a,2]],1),[[0,a,1],[1,b,1]])
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
independent quotas	`[[0, 'a1', 3]]`	`[[0, 'a1', 3], [2, 'b1', 4]]`	Failed
capacity two	`[[0, 'a1', 3], [1, 'a1', 2]]`	`[[0, 'a1', 3], [1, 'a1', 2], [2, 'b1', 4], [3, 'b1', 1]]`	Failed
zero capacity	`[]`	`[]`	Passed
empty requests	`[]`	`[]`	Passed
singleton	`[[0, 'a1', 1]]`	`[[0, 'a1', 1]]`	Passed
interleaved expert	`[[0, 'a1', 1]]`	`[[0, 'a1', 1], [1, 'b1', 1]]`	Failed

SHA-256 / d3a782e2a789044a97fef8be1f6a1f518afea7dda3c680dc790d5e87f77eda48

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(requests,capacity):
    return [r for i,r in enumerate(requests) if i<=capacity]
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=[[0,a,3],[1,a,2],[2,b,4],[3,b,1]]
check('independent quotas',solve(rows,1),[rows[0],rows[2]])
check('capacity two',solve(rows,2),rows)
check('zero capacity',solve(rows,0),[])
check('empty requests',solve([],N),[])
check('singleton',solve([[0,a,1]],1),[[0,a,1]])
check('interleaved expert',solve([[0,a,1],[1,b,1],[2,a,2]],1),[[0,a,1],[1,b,1]])
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
independent quotas	`[[0, 'a1', 3], [1, 'a1', 2]]`	`[[0, 'a1', 3], [2, 'b1', 4]]`	Failed
capacity two	`[[0, 'a1', 3], [1, 'a1', 2], [2, 'b1', 4]]`	`[[0, 'a1', 3], [1, 'a1', 2], [2, 'b1', 4], [3, 'b1', 1]]`	Failed
zero capacity	`[[0, 'a1', 3]]`	`[]`	Failed
empty requests	`[]`	`[]`	Passed
singleton	`[[0, 'a1', 1]]`	`[[0, 'a1', 1]]`	Passed
interleaved expert	`[[0, 'a1', 1], [1, 'b1', 1]]`	`[[0, 'a1', 1], [1, 'b1', 1]]`	Passed

SHA-256 / 6fc3ffbad7ee9cbb420d34a623bf81183869bbe2d2b64390bd302db7ee8bf188

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 2026-09-30T16:48:36.731182+00:00.

Case digest / 8da2e835a05806c4682b5642bcaf92eae61e68e7a7a9401486d03267a59dc99e