{"abstract":"Unequal valid-token batches receive equal token-mean weight.","category":"Gradient accumulation weighting","checks":7,"contract":"Each sum contains only valid-token contributions. Divide the combined sum by all valid tokens; no valid tokens yields zero.","contract_signature":"valid_sums, valid_counts","evaluation_group":"gradient-accumulation-sample_mean","failed_approach":"A microbatch mean still gives short and long token segments equal influence.","family":"w3-gradient-accumulation-weighting-valid-token-mean","id":"FA-98846","implementations":{"attempt":{"sha256":"8058abeb2959de7058ad9746655f6e729f762877a521a4c56bf479a8bf47af87","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(valid_sums, valid_counts):\n    return sum(valid_sums) / max(1,len(valid_sums))\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\ncheck('unequal token lengths',solve([2*N,12*N],[N,3*N]),3.5)\ncheck('equal lengths',solve([2*N,4*N],[N,N]),3)\ncheck('single segment',solve([5*N],[N]),5)\ncheck('all empty',solve([0],[0]),0)\ncheck('empty list',solve([],[]),0)\ncheck('zero segment',solve([0,9*N],[N,3*N]),2.25)\ncheck('three segments',solve([N,6*N,8*N],[1,2,4]),15*N/7)\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":"67ed16afade97665ecf88f14b5b66b1c67023daad54284f97399d7ec129b05fd","source":"\"\"\"Failure Map reference implementation. Python standard library only.\"\"\"\nimport json\n\nN = 1\nobservations = []\ndef solve(valid_sums, valid_counts):\n    return sum(s / max(1,c) for s,c in zip(valid_sums,valid_counts)) / max(1,len(valid_sums))\ndef check(label, actual, expected):\n    observations.append({\"check\": label, \"actual\": actual, \"expected\": expected, \"passed\": actual == expected})\ncheck('unequal token lengths',solve([2*N,12*N],[N,3*N]),3.5)\ncheck('equal lengths',solve([2*N,4*N],[N,N]),3)\ncheck('single segment',solve([5*N],[N]),5)\ncheck('all empty',solve([0],[0]),0)\ncheck('empty list',solve([],[]),0)\ncheck('zero segment',solve([0,9*N],[N,3*N]),2.25)\ncheck('three segments',solve([N,6*N,8*N],[1,2,4]),15*N/7)\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 scalar or short-list summaries and do not model autodiff, devices, or an optimizer. 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-gradient-accumulation-weighting-valid-token-mean","generated_at":"2026-09-30T16:48:17.606517+00:00","license":"CC0-1.0","python":"3.12.14","seed":1,"split":"open-access"},"relevance":"A deterministic toy model isolates one local accumulation contract.","root_cause":"The broken path averages segment means even when segments contain different numbers of valid tokens.","sha256":"4ff21167630a140229abc357ae31c96a91e5c5437ccf2b28f4f0b28fde241baf","title":"Unequal valid-token batches receive equal token-mean weight · 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":27.132,"exit_code":1,"observations":[{"actual":7.0,"check":"unequal token lengths","expected":3.5,"passed":false},{"actual":3.0,"check":"equal lengths","expected":3,"passed":true},{"actual":5.0,"check":"single segment","expected":5,"passed":true},{"actual":0.0,"check":"all empty","expected":0,"passed":true},{"actual":0.0,"check":"empty list","expected":0,"passed":true},{"actual":4.5,"check":"zero segment","expected":2.25,"passed":false},{"actual":5.0,"check":"three segments","expected":2.142857142857143,"passed":false}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"unequal token lengths\", \"actual\": 7.0, \"expected\": 3.5, \"passed\": false}, {\"check\": \"equal lengths\", \"actual\": 3.0, \"expected\": 3, \"passed\": true}, {\"check\": \"single segment\", \"actual\": 5.0, \"expected\": 5, \"passed\": true}, {\"check\": \"all empty\", \"actual\": 0.0, \"expected\": 0, \"passed\": true}, {\"check\": \"empty list\", \"actual\": 0.0, \"expected\": 0, \"passed\": true}, {\"check\": \"zero segment\", \"actual\": 4.5, \"expected\": 2.25, \"passed\": false}, {\"check\": \"three segments\", \"actual\": 5.0, \"expected\": 2.142857142857143, \"passed\": false}], \"passed\": false}\n"},"broken":{"elapsed_ms":29.273,"exit_code":1,"observations":[{"actual":3.0,"check":"unequal token lengths","expected":3.5,"passed":false},{"actual":3.0,"check":"equal lengths","expected":3,"passed":true},{"actual":5.0,"check":"single segment","expected":5,"passed":true},{"actual":0.0,"check":"all empty","expected":0,"passed":true},{"actual":0.0,"check":"empty list","expected":0,"passed":true},{"actual":1.5,"check":"zero segment","expected":2.25,"passed":false},{"actual":2.0,"check":"three segments","expected":2.142857142857143,"passed":false}],"passed":false,"stderr":"","stdout":"{\"observations\": [{\"check\": \"unequal token lengths\", \"actual\": 3.0, \"expected\": 3.5, \"passed\": false}, {\"check\": \"equal lengths\", \"actual\": 3.0, \"expected\": 3, \"passed\": true}, {\"check\": \"single segment\", \"actual\": 5.0, \"expected\": 5, \"passed\": true}, {\"check\": \"all empty\", \"actual\": 0.0, \"expected\": 0, \"passed\": true}, {\"check\": \"empty list\", \"actual\": 0.0, \"expected\": 0, \"passed\": true}, {\"check\": \"zero segment\", \"actual\": 1.5, \"expected\": 2.25, \"passed\": false}, {\"check\": \"three segments\", \"actual\": 2.0, \"expected\": 2.142857142857143, \"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."}}