{"abstract":"Combine weights are normalized across unrelated tokens.","category":"Mixture of experts dispatch","checks":6,"contract":"Rows [token,expert,weight]; return rows in order with each weight divided by its token total.","contract_signature":"routes","evaluation_group":"moe-per-token-combine","failed_approach":"Using the first route’s token as the denominator group omits other tokens.","family":"w3-mixture-of-experts-dispatch-combine-token-group","id":"FA-100621","implementations":{"attempt":{"sha256":"98ef609ee66eb4e9fd1f4ecc0dfbb2b4c2999b4499ea45af8f81d27f3a133818","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(routes):\n    first=routes[0][0] if routes else None; total=sum(w for t,e,w in routes if t==first)\n    return [[t,e,w/total] for t,e,w in routes if t==first] if total else []\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\nrows=[[0,'a',N],[0,'b',1],[1,'c',2*N],[1,'d',N]]\ncheck('separate sums',solve(rows),[[0,'a',N/(N+1)],[0,'b',1/(N+1)],[1,'c',2/3],[1,'d',1/3]])\ncheck('one token',solve([[0,'a',2],[0,'b',2]]),[[0,'a',0.5],[0,'b',0.5]])\ncheck('singleton',solve([[N,'x',N]]),[[N,'x',1.0]])\ncheck('empty',solve([]),[])\ncheck('zero token omitted',solve([[0,'a',0],[1,'b',N]]),[[1,'b',1.0]])\ncheck('three tokens',solve([[0,'a',1],[1,'b',1],[2,'c',1]]),[[0,'a',1.0],[1,'b',1.0],[2,'c',1.0]])\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":"85b69c9b499a4fa72a847a3913240a1d9cea709b10d8b0c54c6a344f487ffd16","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(routes):\n    total=sum(r[2] for r in routes)\n    return [[t,e,w/total] for t,e,w in routes] if total else []\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\nrows=[[0,'a',N],[0,'b',1],[1,'c',2*N],[1,'d',N]]\ncheck('separate sums',solve(rows),[[0,'a',N/(N+1)],[0,'b',1/(N+1)],[1,'c',2/3],[1,'d',1/3]])\ncheck('one token',solve([[0,'a',2],[0,'b',2]]),[[0,'a',0.5],[0,'b',0.5]])\ncheck('singleton',solve([[N,'x',N]]),[[N,'x',1.0]])\ncheck('empty',solve([]),[])\ncheck('zero token omitted',solve([[0,'a',0],[1,'b',N]]),[[1,'b',1.0]])\ncheck('three tokens',solve([[0,'a',1],[1,'b',1],[2,'c',1]]),[[0,'a',1.0],[1,'b',1.0],[2,'c',1.0]])\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-combine-token-group","generated_at":"2026-09-30T16:48:37.646348+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":"All routes share one denominator.","sha256":"e3d7cc884122e94297d9a13b0a2ccb43eb1435d427edb7a869c7a816ae018f0e","title":"Combine weights are normalized across unrelated tokens · 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":29.191,"exit_code":1,"observations":[{"actual":[[0,"a",0.5],[0,"b",0.5]],"check":"separate sums","expected":[[0,"a",0.5],[0,"b",0.5],[1,"c",0.6666666666666666],[1,"d",0.3333333333333333]],"passed":false},{"actual":[[0,"a",0.5],[0,"b",0.5]],"check":"one token","expected":[[0,"a",0.5],[0,"b",0.5]],"passed":true},{"actual":[[1,"x",1.0]],"check":"singleton","expected":[[1,"x",1.0]],"passed":true},{"actual":[],"check":"empty","expected":[],"passed":true},{"actual":[],"check":"zero token omitted","expected":[[1,"b",1.0]],"passed":false},{"actual":[[0,"a",1.0]],"check":"three tokens","expected":[[0,"a",1.0],[1,"b",1.0],[2,"c",1.0]],"passed":false}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"separate sums\", \"actual\": [[0, \"a\", 0.5], [0, \"b\", 0.5]], \"expected\": [[0, \"a\", 0.5], [0, \"b\", 0.5], [1, \"c\", 0.6666666666666666], [1, \"d\", 0.3333333333333333]], \"passed\": false}, {\"check\": \"one token\", \"actual\": [[0, \"a\", 0.5], [0, \"b\", 0.5]], \"expected\": [[0, \"a\", 0.5], [0, \"b\", 0.5]], \"passed\": true}, {\"check\": \"singleton\", \"actual\": [[1, \"x\", 1.0]], \"expected\": [[1, \"x\", 1.0]], \"passed\": true}, {\"check\": \"empty\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"zero token omitted\", \"actual\": [], \"expected\": [[1, \"b\", 1.0]], \"passed\": false}, {\"check\": \"three tokens\", \"actual\": [[0, \"a\", 1.0]], \"expected\": [[0, \"a\", 1.0], [1, \"b\", 1.0], [2, \"c\", 1.0]], \"passed\": false}], \"passed\": false}\n"},"broken":{"elapsed_ms":30.139,"exit_code":1,"observations":[{"actual":[[0,"a",0.2],[0,"b",0.2],[1,"c",0.4],[1,"d",0.2]],"check":"separate sums","expected":[[0,"a",0.5],[0,"b",0.5],[1,"c",0.6666666666666666],[1,"d",0.3333333333333333]],"passed":false},{"actual":[[0,"a",0.5],[0,"b",0.5]],"check":"one token","expected":[[0,"a",0.5],[0,"b",0.5]],"passed":true},{"actual":[[1,"x",1.0]],"check":"singleton","expected":[[1,"x",1.0]],"passed":true},{"actual":[],"check":"empty","expected":[],"passed":true},{"actual":[[0,"a",0.0],[1,"b",1.0]],"check":"zero token omitted","expected":[[1,"b",1.0]],"passed":false},{"actual":[[0,"a",0.3333333333333333],[1,"b",0.3333333333333333],[2,"c",0.3333333333333333]],"check":"three tokens","expected":[[0,"a",1.0],[1,"b",1.0],[2,"c",1.0]],"passed":false}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"separate sums\", \"actual\": [[0, \"a\", 0.2], [0, \"b\", 0.2], [1, \"c\", 0.4], [1, \"d\", 0.2]], \"expected\": [[0, \"a\", 0.5], [0, \"b\", 0.5], [1, \"c\", 0.6666666666666666], [1, \"d\", 0.3333333333333333]], \"passed\": false}, {\"check\": \"one token\", \"actual\": [[0, \"a\", 0.5], [0, \"b\", 0.5]], \"expected\": [[0, \"a\", 0.5], [0, \"b\", 0.5]], \"passed\": true}, {\"check\": \"singleton\", \"actual\": [[1, \"x\", 1.0]], \"expected\": [[1, \"x\", 1.0]], \"passed\": true}, {\"check\": \"empty\", \"actual\": [], \"expected\": [], \"passed\": true}, {\"check\": \"zero token omitted\", \"actual\": [[0, \"a\", 0.0], [1, \"b\", 1.0]], \"expected\": [[1, \"b\", 1.0]], \"passed\": false}, {\"check\": \"three tokens\", \"actual\": [[0, \"a\", 0.3333333333333333], [1, \"b\", 0.3333333333333333], [2, \"c\", 0.3333333333333333]], \"expected\": [[0, \"a\", 1.0], [1, \"b\", 1.0], [2, \"c\", 1.0]], \"passed\": false}], \"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."}}