FA-100576 / Mixture of experts dispatch / Open access

Low-confidence routes survive the gate threshold · case 01

Low-confidence routes survive the gate threshold.

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

ROOT CAUSE

Threshold filtering is applied only after capacity selection.

THE FAILURE

Threshold filtering is applied only after capacity selection.

Unsuccessful approach: Using a strict threshold also discards routes exactly on the allowed boundary.

Case contract

Keep each [token,expert,score] route with score greater than or equal to threshold, preserving 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(routes,threshold):
    return routes
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[[0,'a',N],[1,'b',N-1],[2,'c',N+1]]
check('below removed',solve(rows,N),[rows[0],rows[2]])
check('exact included',solve(rows,N-1),rows)
check('above',solve(rows,N+2),[])
check('empty',solve([],N),[])
check('all pass',solve(rows,0),rows)
check('negative boundary',solve([[0,'x',-1]],-1),[[0,'x',-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
below removed	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	`[[0, 'a', 1], [2, 'c', 2]]`	Failed
exact included	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	Passed
above	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	`[]`	Failed
empty	`[]`	`[]`	Passed
all pass	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	Passed
negative boundary	`[[0, 'x', -1]]`	`[[0, 'x', -1]]`	Passed

SHA-256 / 320a5b61bedaa8aee2ce2716d808698e989653fbfa927f5ff31f1b669d2dda10

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(routes,threshold):
    return [r for r in routes if r[2]>threshold]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
rows=[[0,'a',N],[1,'b',N-1],[2,'c',N+1]]
check('below removed',solve(rows,N),[rows[0],rows[2]])
check('exact included',solve(rows,N-1),rows)
check('above',solve(rows,N+2),[])
check('empty',solve([],N),[])
check('all pass',solve(rows,0),rows)
check('negative boundary',solve([[0,'x',-1]],-1),[[0,'x',-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
below removed	`[[2, 'c', 2]]`	`[[0, 'a', 1], [2, 'c', 2]]`	Failed
exact included	`[[0, 'a', 1], [2, 'c', 2]]`	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	Failed
above	`[]`	`[]`	Passed
empty	`[]`	`[]`	Passed
all pass	`[[0, 'a', 1], [2, 'c', 2]]`	`[[0, 'a', 1], [1, 'b', 0], [2, 'c', 2]]`	Failed
negative boundary	`[]`	`[[0, 'x', -1]]`	Failed

SHA-256 / 37b386f6d44a22286fd1a7544d16b79e65ad67ec877e91fafb260743cad91cf6

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 2026-09-30T16:48:37.194144+00:00.

Case digest / 3986138db861f8ee0c912e44b4412221102d0762b3032c478d8f86f628480d6c