{"abstract":"Selected gate weights retain probability mass from discarded experts.","category":"Mixture of experts dispatch","checks":6,"contract":"Select top k (expert,weight) pairs by descending weight then name; normalize selected weights to sum one, with zero total producing an empty list.","contract_signature":"routes,k","evaluation_group":"moe-selected-weight-normalization","failed_approach":"Selecting first but returning raw selected weights still fails the per-token unit-mass contract.","family":"w3-mixture-of-experts-dispatch-selected-normalization","id":"FA-100561","implementations":{"attempt":{"sha256":"e86e2065448291717b990a7706ce8f075420ceeafd28fdf037e926da4d11d4db","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(routes,k):\n    chosen=sorted(routes,key=lambda x:(-x[1],x[0]))[:k]\n    return chosen\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\nrows=[('a',N+1),('b',N),('c',1)]\ncheck('top two mass',solve(rows,2),[('a',(N+1)/(2*N+1)),('b',N/(2*N+1))])\ncheck('all mass',solve(rows,8),[('a',(N+1)/(2*N+2)),('b',N/(2*N+2)),('c',1/(2*N+2))])\ncheck('one route',solve(rows,1),[('a',1.0)])\ncheck('zero k',solve(rows,0),[])\ncheck('empty',solve([],N),[])\ncheck('zero total',solve([('a',0),('b',0)],2),[])\nprint(json.dumps({\"observations\": observations, \"passed\": all(x[\"passed\"] for x in observations)}, ensure_ascii=False))\nraise SystemExit(0 if all(x[\"passed\"] for x in observations) else 1)\n"},"broken":{"sha256":"1c9408032c54eff36467f3e6a8235131bcedb090d55a74f4163c2c4c61042235","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(routes,k):\n    total=sum(v for _,v in routes)\n    return [(e,w/total) for e,w in sorted(routes,key=lambda x:(-x[1],x[0]))[:k]] if total else []\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\nrows=[('a',N+1),('b',N),('c',1)]\ncheck('top two mass',solve(rows,2),[('a',(N+1)/(2*N+1)),('b',N/(2*N+1))])\ncheck('all mass',solve(rows,8),[('a',(N+1)/(2*N+2)),('b',N/(2*N+2)),('c',1/(2*N+2))])\ncheck('one route',solve(rows,1),[('a',1.0)])\ncheck('zero k',solve(rows,0),[])\ncheck('empty',solve([],N),[])\ncheck('zero total',solve([('a',0),('b',0)],2),[])\nprint(json.dumps({\"observations\": observations, \"passed\": all(x[\"passed\"] for x in observations)}, ensure_ascii=False))\nraise SystemExit(0 if all(x[\"passed\"] for x in observations) else 1)\n"}},"limitations":"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.","method":"Deterministic executable model with adversarial boundary fixtures.","provenance":{"created_by":"Failure Map","dependencies":"Python standard library","family":"w3-mixture-of-experts-dispatch-selected-normalization","generated_at":"2026-09-30T16:48:37.161019+00:00","license":"CC0-1.0","python":"3.12.14","seed":1,"split":"open-access"},"relevance":"A deterministic toy router makes this dispatch decision independently inspectable.","root_cause":"The full gate vector is normalized and then truncated, so retained weights can sum below one.","sha256":"a30265a94da688df76cce4a0dda42cb9e4da57163d1efa9f6df15bc28c76ea2d","title":"Selected gate weights retain probability mass from discarded experts · case 01","variant":1,"variant_policy":"The fixture code references N, supplied as 1 through 5. Fixed boundary checks may repeat. These records share one mechanism and must remain together in evaluation splits.","verified":true,"visibility":"public","verification":{"attempt":{"elapsed_ms":29.561,"exit_code":1,"observations":[{"actual":[["a",2],["b",1]],"check":"top two mass","expected":[["a",0.6666666666666666],["b",0.3333333333333333]],"passed":false},{"actual":[["a",2],["b",1],["c",1]],"check":"all mass","expected":[["a",0.5],["b",0.25],["c",0.25]],"passed":false},{"actual":[["a",2]],"check":"one route","expected":[["a",1.0]],"passed":false},{"actual":[],"check":"zero k","expected":[],"passed":true},{"actual":[],"check":"empty","expected":[],"passed":true},{"actual":[["a",0],["b",0]],"check":"zero total","expected":[],"passed":false}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"top two mass\", \"actual\": [[\"a\", 2], [\"b\", 1]], \"expected\": [[\"a\", 0.6666666666666666], [\"b\", 0.3333333333333333]], \"passed\": false}, {\"check\": \"all mass\", \"actual\": [[\"a\", 2], [\"b\", 1], [\"c\", 1]], \"expected\": [[\"a\", 0.5], [\"b\", 0.25], [\"c\", 0.25]], \"passed\": false}, {\"check\": \"one route\", \"actual\": [[\"a\", 2]], \"expected\": [[\"a\", 1.0]], \"passed\": false}, {\"check\": \"zero k\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"empty\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"zero total\", \"actual\": [[\"a\", 0], [\"b\", 0]], \"expected\": [], \"passed\": false}], \"passed\": false}\n"},"broken":{"elapsed_ms":28.813,"exit_code":1,"observations":[{"actual":[["a",0.5],["b",0.25]],"check":"top two mass","expected":[["a",0.6666666666666666],["b",0.3333333333333333]],"passed":false},{"actual":[["a",0.5],["b",0.25],["c",0.25]],"check":"all mass","expected":[["a",0.5],["b",0.25],["c",0.25]],"passed":true},{"actual":[["a",0.5]],"check":"one route","expected":[["a",1.0]],"passed":false},{"actual":[],"check":"zero k","expected":[],"passed":true},{"actual":[],"check":"empty","expected":[],"passed":true},{"actual":[],"check":"zero total","expected":[],"passed":true}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"top two mass\", \"actual\": [[\"a\", 0.5], [\"b\", 0.25]], \"expected\": [[\"a\", 0.6666666666666666], [\"b\", 0.3333333333333333]], \"passed\": false}, {\"check\": \"all mass\", \"actual\": [[\"a\", 0.5], [\"b\", 0.25], [\"c\", 0.25]], \"expected\": [[\"a\", 0.5], [\"b\", 0.25], [\"c\", 0.25]], \"passed\": true}, {\"check\": \"one route\", \"actual\": [[\"a\", 0.5]], \"expected\": [[\"a\", 1.0]], \"passed\": false}, {\"check\": \"zero k\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"empty\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"zero total\", \"actual\": [], \"expected\": [], \"passed\": true}], \"passed\": false}\n"}},"member_only":{"stages":["fixed"],"fields":["implementations.fixed","verification.fixed","harness","repair"],"note":"The verified repair, its recorded checks, the repair description, and the scoring harness are available to members."}}